Got BDE?⚡️

Hi data baddies 💅🏼

Which is the correct way to calculate average order value? Here’s the data:
↳ $10,000 in sales
↳ 100 orders
↳ $2,000 refunds

Is the answer $100, $80, or ✨depends on the business metric definition✨?

A human data analyst would ask that question before calculating anything. AI agents don’t. They act autonomously, which is why they need CONTEXT: what the data means for your business.

That’s the focus of CData’s context engine, which combines systems knowledge, data modeling, semantic definitions, and institutional knowledge. At Foundations 2026 on November 5, Dr. Nick Golovin walks through how it works: how each request gets paired with its intent, how definitions get captured, and how approved corrections become knowledge every agent reuses.

Register and Join me HERE!

Typing into AI tools is a bottleneck

Have you ever written an AI prompt with no information and just thought in your head, “Oh, AI will know what I mean. It’ll figure it out,” and then gotten frustrated by the results?

Can’t relate 🤪

Sometimes I can be so lazy with AI giving it the bare minimum and expecting it to produce magic, perfect results. Even if I don’t provide it with proper instructions, information, and context.

We have to remember that AI is really not magic. It only has access to the data and information you give it. It has access to the data and information in its training data and any sort of data context and tools you give it access to, and that’s it.

It can’t read your mind. It doesn’t have feelings, and it doesn’t know what you mean if you don’t tell it!

As a very impatient girlie w/ (undiagnosed) ADHD, I’ve realized that my biggest bottleneck to providing sufficient context to AI is actually typing. Not only does typing take so long to do, but it can be very frustrating when your brain is moving faster than your fingers can type. Many times, it leaves me typing less and not fully explaining my thoughts and ideas because I simply am just too lazy and impatient to type it all out.

Come on and admit it. We’ve all been there!!

Which is why this year I’ve become a complete EXPERT in using voice mode, voice-to-text transcription, and even tools like Wispr Flow. And it’s completely changed the way that I use AI.

How to use your voice, what does that mean?

There are multiple ways you can use your voice in AI tools.

1. The first one is through the tool’s native voice-to-text transcription feature. It will likely be a microphone symbol or button next to the chat. This is a way for you to fully put your voice to work. While the voice-to-text model transcribes everything you say into written words, it’s basically the same type of data as typing. The model is converting your spoken words into text for you, so you don’t have to type it all.

2. The second way you can use your voice is by using an AI tool’s official voice mode, where the AI tool speaks spoken language back at you. This feels the most conversational, like speaking to a friend. Note that very often, the AI models used in voice mode are very different than traditional ones. Voice mode might not be the most robust solution depending on your task, but it can be very good for speaking and talking through topics hands-free. I do this all the time while doing my makeup because it makes it super easy to talk through business ideas and brainstorm!

   

3. The third way you can use your voice is with an external tool like Wispr Flow, and this method is by far my favorite and most used. Not only does it work for typing into tools like ChatGPT and Claude, but I also use it to write my newsletters, create brain dumps of written context for my team to use in AI for their work, and to quickly type emails and take notes. Wispr Flow is, in my opinion, the most robust and advanced voice-to-text tool out there because it also understands context from your computer and screen. It also learns new words and context the more you use it.

   

   this picture feels a little bit like Inception, showing you the history of my Wispr Flow, where I used Wispr Flow to type the paragraph above. LOL

The science behind using your voice (instead of typing!)

When you start using your voice in AI tools one of two things happens, and most of the time, it’s actually both:

1. You work faster & more efficiently:
   Perhaps you provide the exact same amount of context, and the quality of your context and prompting does not change. By using your voice instead of typing, now you’re working a lot more efficiently.
   
   Let’s look at the math. The average person types around 40 words per minute, but when you’re talking, you can speak 120 to 150 words per minute!!!! That means that you’re able to provide context at a speed 3 to 4 times faster. Imagine how much faster you can work and how much time you can save. Remember, this is assuming you’re providing the exact same amount of context that you would have provided while typing.
   

2. You provide MORE and BETTER context, resulting in better AI outputs:
   What’s true for most people is that they end up providing MORE context by using their voice versus typing. This is where the true value lies, because now you’re not just saving time and being more efficient (although you likely are doing that as well), but now you’re actually improving your prompting, which will inevitably improve your AI outputs. Now you’re giving AI a lot more relevant information and background knowledge, so it’s going to have more useful information to work with when producing the output

Not only are you probably working more quickly and providing more and better context AI, but it’s also going to save you a lot of energy and mental load. I don’t know about you, but for me, typing is a very daunting and dreadful task. Yapping takes a lot less energy from me and makes me feel a lot less fatigued.

anything that saves me time, energy, and mental load... and also improves my quality of work is a MASSIVE win for me as a lean team doing BIG things.

How I use my voice with AI right now

• Brain Dumps of Context

  • I create a brain dump where I let my thoughts flow about what I want the topic to be, the research I’ve done, and the angle I want to tackle. I keep in my full personality and the typical verbiage I use, and I don’t stop to organize anything or fix a sentence halfway through. Then I take this massive brain dump and put it into AI tools to further streamline and organize my thoughts, or I pass it along to my team to be used in things such as my newsletter. It’s a great way for me to give them a piece of my brain to use for content.

• Content Scripting & Content Ideation

  • When I’m creating content (whether it be video, newsletter, or something else) I first start by yapping all of my thoughts into an AI tool. The beauty of this is that I don’t have to think about the exact words, the structure, the flow, the hook, etc. I first think through all the things I want to talk about in the content and whatever pops in my head. Then I can put it into AI to really streamline my ideas and tighten the messaging. It’s a great way for me to quickly spend 2 minutes brainstorming a topic and talking through it, and then I can quickly turn it into a script or outline for me or my team to use.

• Team feedback

  • When someone on my team sends me content for review (such as a newsletter or a video), instead of hand typing each thing that they should edit, I use voice-to-text to quickly provide feedback by speaking a few sentences. Sometimes I also film a Loom video where my team can grab the transcript and then use AI tools to summarize the feedback, turn it into actionable steps, and then QA the new version against the feedback at the very end. It not only helps me provide and document the feedback, but it also serves as a way for my team to QA their edits at the end.

Across all three, the pattern is the same: I give more context in less time, and the output gets better because of it.

The reminder: always QA

Since we’re giving AI more context, it’s always important to remember that more context is not always better. If you mention something completely irrelevant and say, “Oh, never mind,” AI is not going to understand that. If you tell AI to NOT look at something but it’s in the context, it’s going to still look at it because it’s there.

If you provide irrelevant context, it can actually make AI drift away from your original idea. Sometimes AI can latch on to the most specific things and run with it, so you should always QA your work and results in the end, and always have a human in the loop providing input and reading the final result.

See the example below where I describe using Wispr Flow on my computer one night while watching an episode of Desperate Housewives. It was a very loud scene in the show where Gabby Solis was yelling at Carlos (IYKYK). Wispr Flow heard all of that and included it in the transcription, mixed with my words. I was so confused when I got the output because I had no idea what it was talking about. Then, when I went and looked at my source data (aka my transcript), I realized what had happened and spent the rest of the night laughing 🤣

so it’s okay to use your voice if you’re lazy like me, but don’t be so lazy that you don’t QA 😉

Treat every AI output as a first draft. You have to check names, numbers, quotes, and anything stated as fact against a real source.

But note to self: record my brain dump AFTER the episode ends 😅

Try this today

Next time you open an AI tool, hit the mic button and talk through the task for two full minutes before typing anything.

Say who it’s for, what it’s for, what good looks like, and everything you’d normally skip when typing in a rush. Then read the output with a fine-tooth comb and verify every claim, name, and number before you use it.

And remember, always QA, and don’t use Wispr Flow while watching Desperate Housewives!

Bye BDEs 💅🏼

Jess Ramos 💕

⚡️ Social Highlights:

⚡️If you’re new here:

💁🏽‍♀️ Who Am I?

I’m Jess Ramos, the founder of Big Data Energy and the creator of the BEST SQL course and community: Big SQL Energy⚡️. Check me out on socials: 🔗YouTube, 🔗LinkedIn, 🔗Instagram, and 🔗TikTok. And of course subscribe to my 🔗newsletter here for all my upcoming lessons and updates— all for free!

Got BDE?⚡️

Hi data baddies 💅🏼

Lately I’ve been obsessed with data... and more than usual 😏 Over the past year, I’ve been really digging into how data needs to be ready for AI. I’m hearing from executives across domains and industries why it’s so important to have GOOD data, now more than ever in the age of AI. One company doing it really well is CData.

I just read their whitepaper on their new Sync performance benchmarks, and the numbers are insane: they cut replication time by 76% using the same infrastructure and workload… with zero schema changes 🤯

Sync is CData’s pipeline tool that moves your data from source systems into warehouses like Snowflake and Databricks, keeping it fresh instead of stuck in slow overnight batch jobs.

Check out the whitepaper to learn about:

↳ How parallel partitioned reads split large tables across CPU cores instead of reading serially
↳ Where write improvements shave the most time (hint: wide tables w/ hundreds of columns)
↳ The exact configuration that got Snowflake, Databricks, AND Redshift all under 3 hours

CHECK IT OUT HERE!

Tokenmaxxing is officially over.

For the last couple of years, every company with a pulse has been OBSESSED with AI adoption. Almost in an unhealthy, unnecessary way….. forcing AI into every team, project, and corner of the company without really thinking about it.

We’ve been building AI agents, throwing them everywhere, moving fast, only to figure out why we built them… later. The mindset was simple: the more AI you use, the better, and that is called ✨tokenmaxxing✨.

What’s a Token (And Why Companies Started Maxxing Them)

A token is basically a unit of text that an AI model processes. Think of it like this: one word might be one token, but sometimes a single word gets broken into multiple tokens. It’s how we measure AI usage and how much computational work is being done through an LLM. Measuring tokens is basically measuring how much AI you used.

The problem is that tokens are expensive. Like, REALLY expensive. Each API call to Claude, ChatGPT, or any other large language model costs money based on tokens consumed. The more tokens you use, the higher your bill. That’s right… you’re not paying based on the number of prompts you send or the computational time… you’re paying based on the number of tokens you use. Which can add up quickly and be very costly for large enterprises with tons of teams and people. Especially people who (no offense 😇) aren’t super AI-literate and or giving a d*mn about company costs (it’s ok we’ve all been there!).

And that brings us to….

✨Misaligned Metrics✨

But somewhere along the way, companies started equating token consumption with productivity. More tokens used = more AI being used = more value being created. It became the metric for success. You can’t really measure success and get a clear ROI sometimes, so token usage became a PROXY metric for “success and good ROI”. Engineering teams were approved based on token projections. Managers and leaders were applauded when their teams used a ton of tokens. They must be really smart and really future-thinking and adopting all the coolest tools and trends and producing a lot and being super ROI-heavy!!!! 🤡

The assumption was that heavy AI adoption would naturally lead to better outcomes, more efficiency, and bigger ROI.

It was a seductive metric, easy to measure, to brag about, and easy to track.

It was also completely wrong.

(gosh who could’ve seen this coming, maybe they should’ve had a data analyst choose their north star success metrics! Don’t worry guys, this is job security for us!!)

The Three Problems Tokenmaxxing Created

Problem #1: Nobody Knows What Agents They’re Actually Running

I sat down with the executives at Saviynt and learned that most enterprises have absolute ZERO visibility into their own AI agents. Not just the powerful ones. ALL OF THEM.

And worse, they don’t know who they belong to and what data they can access. They’re just running around rogue, collecting data they shouldn’t have access to, summarizing it, and spitting it out to real people who also probably shouldn’t have access to that data.

When you ask your average enterprise these questions, you’ll get blank stares.

• How many agents are running in your infrastructure right now?

• What permissions do they have?

• What data can they access?

  

This has opened up MASSIVE cybersecurity concerns. Companies were spinning up agents so fast, trying to tokenmaxx that they weren’t implementing basic security controls. Security has become an afterthought. No least-privilege permissions, no audit trails, and no accountability for what agents were doing.

Security teams have started freaking out, and they had every right to. Data teams should also be freaking out because our data should not be out willy nilly in the wrong hands.

Problem #2: Your Data Isn’t Actually Ready for AI

Here's something most companies don't want to admit: their data is a mess.

Please do not argue with me. Having your data in Excel files on someone’s computer and transferring from Salesforce to Snowflake with no oversight and having broken batch pipelines is NOT A FLEX PPL!!

When I talked to Shireesh Tota at Microsoft about getting data AI-ready, he broke down what he’s seeing: a lot of enterprises have data that’s incomplete, inaccurate, or missing crucial context. KNOWLEDGE has to be the word of 2026. Agents don’t have visibility into what the data means. They don’t have semantic definitions. They don’t understand your niche company metrics and proprietary systems.

They’re just pattern-matching on noisy, poorly organized information missing a bunch of details….. yikes 😬

When you give an agent access to bad data without context, you don’t get productivity. You get confident, nonsensical answers. You get decisions made on faulty premises and output that looks good but is fundamentally wrong.

The agents weren’t the problem; the data was. And if your data isn’t enterprise-ready, tokenmaxxing just means you’re doing bad analysis at scale and at speed.

Which is even worse than just doing bad analysis.

Problem #3: Not Every Problem Needs to Be an AI Problem

Here’s the uncomfortable truth for leaders: companies rushed to implement AI everywhere they possibly could without asking if AI was actually the right solution.

“Should we AI this?” became the default question instead of “Do we need AI for this?”

Some tasks don’t need autonomous agents. Some workflows are fine with the old way. Some processes SHOULD have a real human doing them to QA and handle sensitive information.

But in the tokenmaxxing era, if there was a problem, there had to be an AI solution. People decided on the solution without looking at the problem. It was the hammer, and everything started looking like a nail.

We’ve seen this even in past years with trying to apply machine learning and fancy data science to basic exploratory data analysis. Same concept just new tools.

This results in wasted tokens, wasted time, and wasted money on solutions that didn’t actually solve anything.

The Shift: From Tokenmaxxing to Valuemaxxing

I’m starting to see a huge shift in all of the executive interviews I’m doing, from data companies to cybersecurity companies to big traditional companies. Executives are starting to speak less on pushing for AI adoption and more around building data infrastructure and addressing cybersecurity concerns.

The boring non-sexy stuff no one wants to usually talk about. But it’s now become quite necessary. That’s what I learned speaking with executives at SailPoint.

Executives started asking different questions.

• What ROI are we actually getting from this?

• How much real work is being completed?

• How much developer time are we actually saving?

• What’s our total cost of AI implementation?

• Where’s the business case?

Tokenmaxxing became less about how many tokens are being consumed and more about what the tokens actually accomplished.

It’s not about using less AI but about using AI intelligently. It’s about measuring what matters:

• Efficiency gains

• Actual productivity improvements

• Business outcomes

• Risk reductions

Valuemaxxing means you’re looking at the full picture. Yes, we’re using AI, but we’re also measuring whether it’s creating value. Aside from that, we’re being intentional about where we deploy agents and how they’re implemented. We’re making sure our data is actually ready, and we're implementing security from day one, not as an afterthought.

What’s wild is that the companies that shifted to valuemaxxing aren’t moving slower. They’re moving faster because they’re not cleaning up disasters retroactively. They’re not dealing with rogue agents or data integrity issues or compliance violations. They’re building sustainable AI systems from the start.

What this means for you as a leader

(and yes everyone should be a leader in this way regardless of title)

If you’re building AI into your business, stop asking “How do we maximize tokens?” and start asking “How do we create real, measurable value with AI?”

That means knowing what agents you’re running and who’s responsible for them. It means making sure your data is clean, contextualized, and ready for AI to interpret. It means asking whether AI is actually the right solution before you implement it. And it means measuring outcomes, not just consumption.

The tokenmaxxing era proved that AI could do things. Valuemaxxing era is about proving that AI can do things reliably, sustainably, and profitably.

That’s when you know it’s actually here to stay.

Bye BDEs 💅🏼

Jess Ramos 💕

⚡️ Social Highlights:

⚡️If you’re new here:

💁🏽‍♀️ Who Am I?

I’m Jess Ramos, the founder of Big Data Energy and the creator of the BEST SQL course and community: Big SQL Energy⚡️. Check me out on socials: 🔗YouTube, 🔗LinkedIn, 🔗Instagram, and 🔗TikTok. And of course subscribe to my 🔗newsletter here for all my upcoming lessons and updates— all for free!

Got BDE?⚡️

Hi data baddies 💅🏼

Your data strategy is your AI strategy. Here’s the 3-part breakdown.

↳ Governance & security
AI can’t be trusted with real decisions unless your data layer is protecting it through classification, real-time monitoring, and immutable logging.

↳ AI built into your data layer
Building AI capabilities onto a separate vector database means stale data AND double security work.

↳ Scaling to support AI workloads
AI adds a whole new workload, so your infrastructure needs to scale live w/ real traffic.

If you wanna see all 3 of these data strategies, check out Microsoft’s SQL Foundations series in partnership w/ AMD on Microsoft’s SQL Database Hyperscale, deploying a database, migrating with zero rewrites, building AI directly into it, and scaling live from 250 to 2,000 users.

It’s FREE here on YouTube and it’s a MUST WATCH!!

Watch HERE!

What nobody tells you when you integrate AI into your existing systems

Most companies are treating AI like it's a software problem when it's actually a MODEL problem.

Until you fix your data strategy, the AI is just going to amplify whatever problems you already have.

I’ve been thinking a lot about this lately, and it keeps coming back to the same three things. The actual foundation everything else sits on. Let’s get into it!!

1. Data governance & security: You can’t secure AI if you can’t secure your data first

Did you know, people are out here tricking AI agents into leaking data they don’t have access to (😱)? It’s called prompt injection, and it’s exactly what it sounds like: someone hides an instruction inside a totally normal-looking input (a form field, a document, a support ticket, whatever), the AI reads it as a command instead of just data, and now it’s handing over information it shouldn’t.

This isn’t a hypothetical; it’s happening right now to companies who rushed an AI feature into production without thinking about what that AI actually has access to. This only one potential security issue.

The fix isn’t “be more careful” or “write a better prompt.” It has to be built into the system itself. Here are the three things that should be top of mind when it comes to securing your data or AI:

• Classification and masking so that sensitive data is labeled by sensitivity and blocked from anyone (human or AI) without the right permissions

• Real-time monitoring on every single request between the AI and your data, so anything suspicious gets shut down before it ever reaches something sensitive (like a prompt injection!!!)

• Immutable logging, meaning every action an AI takes gets recorded somewhere that literally cannot be edited or deleted after the fact. When something goes wrong (and eventually it will), there’s no chance someone, or something, went back and cleaned up the trail.

Locking things down at the base level is great, but we still have to consider who gets permission to access the data. Identity matters just as much as the data itself: centralized login systems with multi-factor authentication, and passwordless connections for your apps so credentials aren’t sitting around in some config file waiting to get leaked.

Some fields need to go even further than masking. Sensitive columns can be encrypted BEFORE they ever hit the database engine, so not even the people administering it can see the raw values.

If you’re not protecting your data first, you should not be giving AI access to it. No matter how good the use case is.

2. AI built into your data layer: Your AI is only as good as how connected your data actually is

Ever gotten a super confident answer from an AI that was just… wrong? (everyone nod yes🙂‍↕️)

Here’s what’s happening under the hood at some companies: their real data, the source of truth, lives in one system. Their AI search runs on a completely separate one. And those two systems don’t talk to each other in real time.

This creates three problems:

• Stale Data: Whenever your source data changes, there’s a lag before that change syncs to wherever your AI is actually searching. So the AI is confidently answering questions using outdated information and has no idea it’s outdated.

• Duplicated security work: Now you’re securing two systems instead of one. Whatever masking, permissions, and access rules you built for your main database, you have to rebuild all of it again for the AI system. A small mistake or oversight could be very costly (can’t be losing your job in this economy!🤢)

• Search Difficulties: You can’t combine a normal query filter in your primary relational database with a semantic AI vector search. So you run the search in one system, the filter in the other, and stitch the results together yourself. That’s slower, messier, and one more place for something to go wrong.

The fix is building your AI capabilities directly into your primary data layer instead of slapping a separate system on top of it after the fact. One source of truth and one set of permissions to maintain, not two.

This is the unsexy infrastructure decision that determines whether your AI feature is actually embedded in your company data.

3. Scaling for AI workloads: infrastructure that “worked fine” as a demo might break in production

A normal new feature doesn’t really change how hard your database is working. An AI feature is different. It adds a whole new category of load on your database and system: more queries, more read traffic, more strain, and it doesn’t show up until real users start hitting it. The demo works flawlessly with five people in a conference room. Then once it ships, it falls apart.

And it's worse with agents specifically. An agent isn't just answering one prompt and moving on. It's interpreting intent, gathering context, taking multiple steps, checking its own output, before it ever gives you a final answer. Every one of those steps hits your infrastructure. So the load isn't just higher; it's multiplied.

What actually has to be true for your infrastructure to hold up:

• Compute and storage need to be able to scale independently. Compute is your processing power; storage is how much data you can hold. If your data (or your embeddings) keep growing but your traffic hasn’t, you should only need more storage, not more compute, and vice versa. If those two are bundled together, you’re stuck paying for both every single time, even when only one is actually under strain.

• You need to be able to scale compute live, while the system is under real load, not during a scheduled maintenance window. If you can’t scale in real time, your users are the ones stuck dealing with lag, timeouts, and errors while you wait for a window to catch up.

• Your AI traffic shouldn’t be competing with your core application for resources. Searching through a lot of stored data to find the right answer is expensive, and if that runs on the same system as your core application, everything slows down. You need a way to route that traffic somewhere else so it doesn’t take the rest of your app down with it.

None of this is glamorous, but it’s VITAL to make sure it’s ready for AI.

Why this actually matters to you

You don’t have to be the one architecting any of this to care about it. If your company is racing to ship AI features (and let’s be honest, everyone is right now), these three things are what determine whether those features actually work once real users show up.

Governance and security decide whether you can trust what the AI is telling you and who else can see your data. A connected data layer decides whether the AI even has the full picture and context to provide informed decisions. Scalable infrastructure decides whether any of it holds up past the demo with real users.

Most companies are moving fast on the AI part and treating the data part as an afterthought. That’s backwards. You need to fix the foundation first, or you’re just building something impressive but SHIT 💩.

Bye BDEs 💅🏼

Jess Ramos 💕

⚡️ Social Highlights:

⚡️If you’re new here:

💁🏽‍♀️ Who Am I?

I’m Jess Ramos, the founder of Big Data Energy and the creator of the BEST SQL course and community: Big SQL Energy⚡️. Check me out on socials: 🔗YouTube, 🔗LinkedIn, 🔗Instagram, and 🔗TikTok. And of course subscribe to my 🔗newsletter here for all my upcoming lessons and updates— all for free!

Got BDE?⚡️

Hi, data baddies 💅🏼

Why is it that AI Agents sometimes give the wrong answers? Shouldn’t AI be more accurate by now instead of confidently giving us WRONG information?

Well guess what… AI is only as “right” as the data going into it. How can AI give correct answers if you can’t even give it correct data? Stale snapshots, missing context, & source codes your model can’t interpret will destroy your AI outputs and erode trust in your company.

This is what CData’s upcoming FREE 45-minute session is on: Why Your AI Agent Gives Wrong Answers (And How to Fix the Data Layer). You’ll learn:

• The 3 data layer failures that cause AI agents to hallucinate

• How to assemble cross-source context in a single pipeline

• How to enforce a data quality gate before the AI layer

• How to keep your data pipeline version-controlled and auditable

SAVE YOUR SEAT and I’ll see you there

SAVE YOUR SEAT HERE!

The setup for how I got the interviews

Everyone keeps asking me how I’m interviewing so many executives.

The truth is that I just ask! I put myself out there, and I make it happen.

To be fully transparent: around half to 60% of these interviews are tied to a paid partnership where companies want me to come interview their execs as part of official deliverables. The other half is me raising my hand and volunteering.

Plus, it makes really good content! You get to see how executives actually think and how they’re approaching AI. It also opens up SO many doors because once executives know me and know my name, it helps me build deeper relationships with their companies. It’s kind of a win-win all across the board.

If you’re wondering, yes, sometimes I am nervous to interview executives. I have a lot of imposter syndrome, and sometimes I can’t believe that I’ve built this type of career and that executives see the value in interviewing with me.

But in a weird way, it’s helped me grow my confidence a lot and realize that my voice matters, and the community I’ve built in data and AI really matters.

Today I’m breaking down all of the executives I’ve interviewed so far this year, how they happened, and what I learned from each one. Of course, I have more exec interviews on the roadmap for later this year, so I’ll have to update this at the end of the year. 🥳

1. Dave Steer, the CMO of Webflow

Dave was my first real sit-down exec interview in an official capacity, and I still can't believe it happened the way it did.

I live in Atlanta (which comes with crippling FOMO by the way because everything feels like it's happening in SF or NYC) and totally didn’t expect anything big. Then Webflow reached out and said their CMO would be in town and wanted to do content.

And because I live in Atlanta, traffic is genuinely worse than every joke you’ve heard about it. I was so paranoid about being late that I stayed in a hotel the night before. Ate Brussels sprouts and French fries in bed in my bathrobe (very glamorous, I know) because that's how I mentally prepare for things I'm nervous about.

Dave is one of those executives who just makes everyone feel like they're the only person in the room. He's kind and real and doesn't do that fake corporate energy. We talked about AEO (answer engine optimization) and how AI is changing the whole marketing and discoverability game.

He's also a Taylor Swift fan, which I learned mid-interview and just made everything feel more human.

I posted the polished interview AND a behind-the-scenes video of me in my bathrobe eating fries in bed. And Webflow loved it so much they wanted to partner again. That taught me something huge about what brands actually want: it’s not the polish. It’s the authenticity. It’s the real version of you.

That shifted my entire approach to content after that.

By the way, I interviewed him again, and here’s the full sit-down YouTube video from when we worked together on Flow TV.

2. Arthur Hu, SVP/Global CIO of Lenovo

This one was at NVIDIA GTC in San Jose. NVIDIA scheduled the interview with me at the Lenovo activation, and I was excited to talk with Art about how a company like Lenovo thinks about enterprise-scale technology.

We talked about F1, specifically Lenovo’s technology partnership with F1 and how the data is changing as the rules in the sport are changing. Since I went to my first F1 race in Las Vegas at the end of 2025, I’ve been genuinely curious about the tech side of it, not just the sport.

Art was really cool about diving into the nitty-gritty of how data moves in that sport and what changes when regulations shift. It was the kind of conversation where you realize how differently different industries have to think about the exact same problems.

The realization that F1 is basically the Wild West of data right now. Everything is changing so fast, rules are being rewritten, and the data architecture has to move just as fast. That’s actually a metaphor for AI in 2026 too. We’re all just trying to keep up.

3. Shireesh Thota, CVP, Azure Databases at Microsoft

I went to the Microsoft Fabric Conference in Atlanta (FabCon SQLCon, which is basically nerd heaven if you're obsessed with data and SQL), and I got to interview Shireesh. I was so excited for this one because this is Microsoft, and I've been a hardcore Microsoft user my entire career. This person helps build the infrastructure I use. I really wanted to impress Microsoft and do a great job on this.

I asked him about the future of SQL and specifically how AI is changing the way databases need to be built. Because here's the thing: AI agents are consumers of data now, not just humans. So the rules have to completely change. Databases can't be structured the way they used to be when only humans were reading them.

Shireesh was thoughtful and deep about it and apparently everyone at Microsoft saw the video because he told me people kept coming up to him saying, "Dude, you're famous now." Which is hilarious but also? Big credit to him for being willing to have a real conversation on camera.

The second time I interviewed him was more formal (Microsoft actually reserved a studio for us!) but also deeper because we’d actually built some rapport.

This time I asked him the hard questions I’d been wondering about for months: “What does it actually take to get your data ready for AI agents?” and “What are the red flags that your data isn’t ready?” These felt like the questions that mattered but I wasn’t sure anyone would actually answer them honestly.

Shireesh gave me some of the most realistic answers I’ve heard from any exec.

Data readiness for AI is not a technical problem. It’s an organizational problem. Because a lot of the time your data is fine, but it’s buried in someone’s head or locked in a spreadsheet or living in institutional knowledge that nobody wrote down. You can’t just throw AI at that.

4. Kim Manis, CVP of Product, Microsoft Fabric:

Same conference, FabCon SQLCon in Atlanta. I got to interview Kim, and I was genuinely excited because I don't get to interview women in leadership in data as often as I should, and Kim is like, THE person in the Fabric community.

She gave me her best tips for building data projects using Power BI and Fabric, but more importantly, she talked about being a woman in data at the executive level in a way that actually mattered. She's got this radiant personality and you'd never know she's this high up at Microsoft because she's just genuinely warm.

This interview still gets the highest engagement of any exec interview I've done all year. I think it's because people could actually feel how real the conversation was. Kim wasn't performing executive-ness. She was just being a person who happens to be an executive. Big difference.

5. Marissa Kraines, VP Brand Marketing of Webflow

Webflow brought me back for their FlowTV launch and Marissa was part of that project. I got to be on an actual TV set (surreal) and interview both Marissa and Dave as they were working on their new series.

Marissa talked about how Webflow is building their new series and the behind-the-scenes strategy. It was cool to see how a brand builds content from the inside and also just to see her leadership style in action.

What started with one Dave interview turned into a whole ongoing partnership with Webflow. And that relationship keeps opening new doors. That's what happens when that first interaction is genuine.

6. Eniko Rozsa, Global CTO of AI at IBM Business Consulting Services

Red Hat Summit was the conference where everything kind of exploded for me. I partnered with IBM and got to interview a couple of their executives there.

Eniko talked about IBM’s approach to agentic AI and more importantly, the absolute messiness of real-world data in enterprises. Like, she gave me the most honest answer I’ve ever heard from any executive about why AI projects fail… It’s never the AI. It’s the data context.

She talked about how context lives everywhere, in spreadsheets, in people’s heads, and in documentation that doesn’t exist. You can’t just hand AI a dataset and expect it to work; it needs context. And getting that context in a place where AI can actually use it is the bottleneck.

A lot of the AI problems we talk about (hallucinations, inaccuracies, weird outputs) are actually just data context problems. We blame the model when we should be blaming the lack of semantic structure.

7. Robin Hernandez, VP of IBM Consulting Offerings

During the Red Hat Summit, I interviewed Robin, and she gave me some of the best advice I've heard about being a woman in AI and data. But she also complimented my lip gloss and asked me what product I was using, and I got to recommend my favorite Rare Beauty lip oil.

That's the moment that really stuck with me. Here's this VP at IBM in these high-level strategy conversations and she still cares about good lip gloss. Which sounds small but it's actually everything. It's the reminder that executives are just people.

Permission to be whole. Executives don't have to be all business. They're allowed to care about lip gloss and strategy at the same time. That gave me permission to stop trying so hard to be the "serious data person" and just be, like, a full human who also knows her stuff.

8. Matt Hicks, CEO of Red Hat

This is the crazy one! I posted about going to Red Hat Summit and one of my engaged followers commented "Another exec interview?" with the 👀 emoji. I joked, "I don't know, maybe" and tagged my contact at Red Hat.

24 hours later, I got an email from her saying, “CEO interview, yes or no.” And of course it’s the CEO of Red Hat! I’ve never seen a social media team work so fast to set up a CEO interview; it was CRAZY.

I showed up VERY prepared, and he was one of the kindest, least arrogant executives I’ve talked to. He was genuinely excited to talk about his early coding projects and the history of open source.

I think because I was prepared and respectful, but also genuine, he met me halfway and actually gave me a real conversation instead of corporate talking points.

That interview taught me more about confidence than anything else. When you’re fully prepared but also authentically curious, people respond to that. They want to actually talk to you, not perform for you.

We talked about how AI just made your coding skills worthless and how Matt started coding at 7 years old!

9. Rekha Srivatsan, CMO of Tableau

This happened during the Tableau conference, and I was genuinely excited about this because I’ve been using Tableau forever, and obviously, as a data person, I’m a HUGE fan of data visualization and business intelligence tools. To interview someone who’s worked at Tableau for so long and knows so much about the tool, I was ecstatic.

Rekha talked about how the role of data analyst is evolving because AI is automating the coding part. But that means analysts need to become knowledge architects instead. They need to think about semantic layers and data context and how to build structures that AI can actually use.

She gave me one of the most forward-thinking takes on where data analytics is going that I’ve heard from any exec.

The definition of data literacy is changing. It’s not just about knowing SQL anymore; it’s about understanding how to structure knowledge and context in a way that’s useful to AI. That’s a total mindset shift.

10. Matt Mills, President of SailPoint

SailPoint flew me to New York City for their big agentic fabric launch at NASDAQ.

Matt talked about agentic fabric, which is basically the practice of observing AI agents and making sure they’re not doing rogue things. But the deeper conversation was about how we now need to secure AI identities just like we secure human identities.

That was a huge realization for me because I hadn't really thought about cybersecurity and AI as that tightly connected.

Data and cybersecurity are the same conversation now. You can't talk about AI adoption without talking about protecting that AI and the data it touches. They're not separate problems anymore.

11. Charles Feddersen, Director of Product Management for Postgres at Microsoft

This happened during Microsoft Build. He’s not technically an exec (he made sure to clarify that!), but he’s deeply respected in the Postgres community, and I wanted to talk to him.

This was genuinely wild. We got through like one or two questions, and I got stung by a bee! Like, a bee landed on my hand, and the stinger got stuck in me. I genuinely panicked. But I wasn't going to stop the interview because that would be ridiculous.

Someone had to pull the stinger out. I took some deep breaths, and then I just… kept going, and we finished the whole thing.

Afterwards, my contact took me to the first aid tent, and honestly, I was just relieved it was over, and I could go back inside where there are no bees.

Charles talked about how Postgres has evolved over time and how open source has changed the entire data landscape. The evolution from Microsoft Access to where we are now is kind of mind-blowing when you think about how much data infrastructure has changed.

12. Mansi Jain, COO of Glance

Glance was actually a major sponsor, along with Microsoft, at the Women in AI event I hosted in Cannes with four other amazing AI creators.

It was an amazing space to bring together women in AI, create community, eat good food, and cool off in the hot south of France sun.

Mansi talked about how AI isn't just about enterprises and data analysts. It's about consumers. Glance is building an app that reaches millions of people, and it's powered by AI, making recommendations based on taste and preference.

It was a good reminder that AI adoption isn't just about Big Tech and data engineers. It's about the actual human experience of using products that are smarter.

AI is becoming invisible. We don't call it AI anymore; we just call it good product design. That's actually the goal: AI so seamlessly integrated into a product that people don't think about it.

13. Saviynt Executives, Sachin Nayyar (CEO & Founder), Paul Zolfaghari (President), & Vibhuti Sinha (CPO)

These interviews were with the CEO, CPO, and president of Saviynt. They flew me to Nasdaq in New York City for the launch of their Zuma product, which is a platform for observability and cybersecurity for AI agents.

I interviewed multiple executives from Saviynt and they dove deep into the problem of permissions and access for AI. Specifically: when you move data between platforms (like from Salesforce to Snowflake), the permissions don't move with it. You have to think about security at the data point level, not just the platform level.

They also talked about tokenmaxxing, which was this wild trend in 2025 where companies measured success by how many tokens they were using with AI. The more tokens, the more AI adoption, the better, right?

Except no. It was just expensive and increased security risk, and sometimes there was literally zero business ROI.

Companies are overcorrecting right now, but the real answer is somewhere in the middle: use AI where it actually solves problems, secure it properly, and measure actual business value instead of just usage.

One Last Thing

You might be wondering what interviews I have next.

Well, I have a busy fall that’s full of conferences ahead of me. I can’t spill all of my secrets and tell you all the tea now, but I promise you I have some really big, high-profile companies coming for you this fall. 👀

I will say, this week I’m headed to Boston for the Webflow Conf, where I’m gonna be interviewing a lot of Webflow’s executives all about AI, AEO, and discoverability. I’m so excited for that along with my other fall conference travel, and I’m really proud of how much I’ve grown and learned from interviewing executives.

Bye BDEs 💅🏼

Jess Ramos 💕

⚡️ Social Highlights:

⚡️If you’re new here:

💁🏽‍♀️ Who Am I?

I’m Jess Ramos, the founder of Big Data Energy and the creator of the BEST SQL course and community: Big SQL Energy⚡️. Check me out on socials: 🔗YouTube, 🔗LinkedIn, 🔗Instagram, and 🔗TikTok. And of course subscribe to my 🔗newsletter here for all my upcoming lessons and updates— all for free!

Got BDE?⚡️

Hi data baddies 💅🏼

Most people grind for months trying to break into data and never land the role. Usually, because they are focused on the wrong things. 🙅🏻‍♀️ My FREE 5-day Data Career Kickstart course fixes that. 🔥

You can access the FREE course:

Join HERE

DATA > AI MODELS

Your AI models aren’t underperforming because of the model itself... they’re underperforming because they’re running on BAD DATA! Most likely stale data that isn’t relevant anymore. I know, not what you want to hear 😅

Your data might not be AI-ready if you have batch jobs copying full tables every few hours and legacy sources like SAP or SQL Server.

If you wanna see how modern enterprises are solving this with CDC (Change Data Capture, meaning you only move what actually changed, not the entire table every time), CData is hosting a free live demo August 6, 10am ET to show you!! You’ll:

↳ Watch live pipelines continuously replicate from SQL Server, SAP, and DB2 into Snowflake, Databricks, and ClickHouse
↳ See how to connect legacy on-prem sources to your modern stack without re-architecting your infrastructure
↳ Learn how to keep your AI models and analytics fed with fresh, governed, production-ready data

Register Here!!

Everyone keeps saying AI is killing data analyst jobs, but the reality is bad analysts are being replaced while good ones are making more money than ever!

In this video, I’m breaking down what’s actually happening in the data world, from the 3 massive industry trends to the exact skills you need to future-proof your career and land six-figure roles. Check it out here 👇🏻

8 years of Data Analytics Advice in 5 min

One thing I’ve been really good at for the past eight years is data analytics.

I have analyzed data for some of the fastest growing tech start-ups, and I have built a community of over 650,000 data professionals just like you.

People ask me every day to be their mentor, and people have offered me thousands of dollars for a single day of coaching.

These are the lessons that took me YEARS to learn. Nobody puts them in a bootcamp or a course, but they’re the actual reason I was successful in the field.

1. Technical Skills WON’T get you promoted

Let’s start spicy: technical skills won’t get you promoted 👀

Being really good at SQL, Python, or Tableau will make you a good analyst. But it’s probably not going to make you a senior analyst, and it’s most definitely not going to make you a lead or staff analyst.

Technical skills are great. It’s how you actually get your hands dirty and get the job done. But it’s not who’s the best at their job or who knows the most that gets promoted. In data analytics, promotions happen when you’re a multiplier. It’s not just about doing the work, it’s about multiplying that work into business outcomes, mentorship, and driving leadership.

I spent so long taking more courses and trying to level up my technical skills so somebody would notice me. But business stakeholders and managers don’t care about technical skills. Those are things you can always look up along the way.

What they actually care about:

• Being proactive. You have to anticipate problems before they happen and already have recommendations and solutions ready to go.

• Relationships. You have to play the politics. If nobody’s buying what you’re selling, it doesn’t matter how good what you’re selling is. It’s not just about good insights; it’s about getting buy-in from leadership and getting them to trust you enough to actually act. It’s a popularity contest, kind of like high school, and if they don’t like you, you won’t move up

• Translating data into business action. “Conversion dropped 8%” is a data point. That’s thinking like a data analyst. Instead, translate it: “Conversion dropped 8% because all the users in segment A did this. Here’s what we can do about it.”

So it’s not the technical skills that get you promoted. It’s playing politics, making your work visible, building relationships, being proactive, all those soft skills we don’t want to think about as data analysts. But they’re necessary if you want to move up. Learn how to play the game.

2. The data you need might not exist (and that’s actually your job to fix it)

When I started as an analyst, I genuinely thought I was going to get perfect, ready-to-go databases and have everything I needed to analyze data. Which is kind of true for the most part! Most companies have big databases with plenty of data for you to use. Usually way more data than you ever need and know what to do with (but whether or not it’s clean, complete, and accurate is a whole other story)

What I wasn’t prepared for is when stakeholders ask a very specific, niche question we don’t already have data set up for. Maybe it’s something no one’s asked before, or it’s something we’re just not tracking yet.

From here, you have a few options, and you have to weigh the pros and cons of each:

• Actually create the data. Design the metrics in SQL or Python, build the calculations from scratch, or combine data from a bunch of different sources to solve the specific problem. Very often, you can create the metrics and calculations you want to track from other data you already have

• Create the data through your stakeholders. Give them a survey, or have them log data into a CRM like HubSpot or Salesforce. You’re relying on their business domain expertise to create the data you need. Of course, it can be a little messy this way, and sometimes not always be consistent and complete, but they truly are the experts in their domain. They can be the best people to ask for very specific data points

• My favorite way: track events and send those into the database. You work with engineers to set up different events on the front end of your product that fire and log data. You define the schema, the properties, the event names, and the fields. Someone clicks a button, that fires an event, and BOOM, you’ve created data that didn’t exist before.

But sometimes, even after all that, it’s still just impossible. Sorry, we don’t have the data. We really wish we did, but sorry, we just can’t. And that honestly might be my least favorite answer or maybe my favorite because it’s the least amount of work for me. 👀 Saying no may sound terrible and not super helpful, but sometimes that’s the best thing you can do. It’s better to say no if you know that you can’t provide a good sound analysis than to create an analysis from shitty data and mislead stakeholders in the business.

That’s just the everyday of being a data analyst. You don’t always have the data you need. You have to get creative and either create it, find it, or say no, because stakeholders really expect us to be data magicians. ✨

3. STOP being a reporting vending machine

A lot of companies have analysts working like reporting vending machines (IYKYK).💀 A stakeholder says “oh my gosh, emergency, I need this report right now!” You stop what you’re doing, and you do the report. That’s a reactive data analytics team, just keeping up with demands and doing EXACTLY as you’re asked. That’s a more basic, elementary level of data analytics.

A more senior team is proactive instead of reactive. They ask why. Why do you need to know? Why is this important to you and your team? How is it going to impact the business? What would signal success and failure for this project? Are there any risks involved? Instead of asking short questions, they ask deeper questions like:

• What business question do you hope to answer?

• What would signal to you that this project was a success?

• What would signal that it wasn’t, and we should roll it back?

• What’s the risk involved here?

• What does leadership and your manager ask about this project?

When I first started my career, I was a people pleaser. I was a data order taker. I say yes, ask them what they want, and I’m on it. And not just on it, I’d do everything immediately because I was scared to say no or push back.

But managing stakeholders, pushing back, and asking questions professionally is what makes you a really good data analyst. So don’t be a data order taker. Be a data advisor.

4. Nobody cares how complex your code is

Business stakeholders don’t give a care about the technical details. They don’t care how complicated your analysis is. They don’t want to see the crazy calculations, the metrics, the code, all the work you did to get to the final answer. They literally only care about the final answer; sometimes that answer is as simple as yes, or do it, or no, don’t.

I made a huge mistake on this early in my career. I built this entire report with heavy-duty nested IF functions in DAX in PowerBI. When I presented it to stakeholders, I actually pulled up the function and walked them through it and exactly how it works .

They sat there with their eyes glazed over. It was… a spectacle 😅 At the end of the day, they didn’t need to see any of what I did behind the scenes. They just needed to know I built something trustworthy they could use to track the metric they wanted. I still CRINGEEE every time I think about it.

Corporate people do not want to work harder. They don’t want to use their brains more. They don’t want to be challenged. Just tell them what to do and make it easy. If you make someone else’s job easier, they’re going to like you more. So cut the jargon and shield your non-technical stakeholders from the nitty-gritty.

Nobody wants to see all of that.

5. False positives will destroy your credibility

This is kind of like the boy who cried wolf. Every day you cry “Wolf!” when nothing was there, then one day the wolf really was there, and no one is there to believe you.

If you give too many false positives to your business stakeholders, you’re basically teaching them to ignore everything you say, making yourself irrelevant and undermining your credibility. If you’re always sounding alarms and making everything feel like an emergency when it’s not, no one’s going to take you seriously when there actually is an issue. It’s especially true with metric alerts. If an alert fires every day, is there ever really an alert? Stakeholders get so desensitized to that same alert email once a day. Even once a week is a lot.

If you’re always signaling something is wrong, they’re not going to believe you when something ACTUALLY is wrong, and it’s going to be really hard to get buy-in to act.

DON’T be the data analyst who cried all hands on deck. Protect the trust and credibility with your data organization.

6. Rogue dashboards are your worst nightmare

Have you ever been in a company-wide all hands and an executive pulls up some dashboard you’ve never seen before, with no idea where it came from, how it was calculated, what data was used, or what cleaning process was done to it? That is my ACTUAL nightmare.

This is why data governance and processes matter so much. If stakeholders have access to Tableau or, worse, SQL in the database, and they can just pull their own stuff and present it like gospel, especially in the age of AI, it’s scary. Don’t get me wrong, I love that they take the initiative, but you need checks and balances. Companies hire data analysts and data scientists because we understand the implications, the filters, and the trade-offs. When non-technical people who aren’t trained pull their own data and build their own visualizations, you don’t know where they got that data or what filters were selected. Or even if the data is clean and up-to-date.

All data should not be self-served. I’m all about democratizing self-serve insights, but at the end of the day, we should be gatekeeping a lot of this. Except it’s not really gatekeeping; it’s data governance. This will be the sexiest data domain of 2027 with AI agents!

As a data analyst, it’s your job to keep educating stakeholders on quality control, make sure the wrong people don’t have access to the wrong things, and set up a process for how requests come in and get prioritized.

Sometimes you have to push back and say no. Otherwise people will take advantage.

7. Data lives in the Gray Area

Everyone wants clear, confident, objective answers from data that say yes or no and point in an exact direction. That’s not always the case. Data can be incomplete, missing, messy, unreliable. Experiments can be contaminated or not set up well. Sometimes multiple metrics tell different stories, and you have to evaluate the risk and trade off the best you can.

Say a product change goes live and conversion rates go up, but refund rates go up too. More people are converting and buying, but more people are also asking for their money back. So what’s the real goal here? Conversion? Revenue? Something else? You have to lay out the trade-offs for stakeholders: “Yay, conversion went up, but so did refund rates.” Because stakeholders get distracted by shiny objects. They see conversion up, they think great job, and they’re ready to report that success to leadership. Your job is to put that in check and explain the full picture.

Or say you run an A/B test and it shows a small lift in the metric you’re evaluating. A little lift is good; it’s not a negative thing. But what if it takes significant resources to fully roll out that change? Is it worth weeks or months of company resources for that much of a lift?

Real data analytics lives in uncertainty, trade-offs, and gray areas. It’s not always an easy decision. But the real value comes in evaluating risk and trade-offs, and actually understanding business impact. ⚡️

8. Treat data like a product, NOT a task

You’re not just delivering a quick report that no one’s going to look at. You need to treat it like an actual product, which means thinking from the user’s point of view. You evaluate whether it is easy to use. Does it have clear definitions? Are the filters clear? How do they export?

Think through the challenges users will run into before you hand it over, and keep collecting feedback after you do. Check the history of the dashboard to see who’s using it and when. That tells you whether it’s useful, and how useful it is.

Since it’s a product, you also have to market it. If a dashboard isn’t being used much, it’s either not that useful, or people don’t know about it. So do your own marketing: host a training call, teach people how to use it, add filters, export to Excel (because we all know they’re going to do that anyway). Announce changes and iterations in Slack. Treat your data like a product and act like the product manager: handle the iterations, the feedback, the marketing, everything, to keep that thing alive.

9. Your manager shapes your data culture

Your manager can have a huge impact on the data culture at your company, and it can really impact your experience there.

You can be the best data analyst in the world, with a team full of valuable insights that could make huge business impact. But if your leadership doesn’t see the value, it’s not going to go anywhere. Data culture flows from the top down. If your manager isn’t advocating for your team and talking about the importance of the data with leadership on your behalf, you might end up running in circles copying and pasting in Excel because they don’t understand the more advanced things you’re doing.

One of the best skills a data manager can have is stakeholder management. Not only should they advocate for your team’s visibility in front of leadership, but they should also manage the crazy stakeholder requests and act as a barrier or a shield for your team. Their job is to make you look good and push back on the nonsense so you can put your head down and get things done.

I’ve seen really good data culture, and I’ve seen really bad data culture 😅. You don’t fully understand what good and bad look like until you’ve experienced both. But a lot of it comes down to your manager and their relationship with leadership.

What Actually Gets You There

I think about this a lot, and these lessons took me years to learn, but they completely changed my career.

If you're serious about breaking into data analytics or advancing into a more senior role, the thing holding you back probably isn't the technical skills. It's how you apply them to real business problems, and whether you're willing to play the politics.

So yes, take the SQL course. Build the portfolio. But start practicing the stuff that actually moves you, too. Ask better questions. Learn how your company really makes decisions. And build the relationships before you need them.

That’s what changes things.

Bye BDEs 💅🏼

Jess Ramos 💕

⚡️ Social Highlights:

⚡️If you’re new here:

💁🏽‍♀️ Who Am I?

I’m Jess Ramos, the founder of Big Data Energy and the creator of the BEST SQL course and community: Big SQL Energy⚡️. Check me out on socials: 🔗YouTube, 🔗LinkedIn, 🔗Instagram, and 🔗TikTok. And of course subscribe to my 🔗newsletter here for all my upcoming lessons and updates— all for free!

Got BDE?⚡️

Hi data baddies 💅🏼

Most people grind for months trying to break into data and never land the role. Usually, because they are focused on the wrong things. 🙅🏻‍♀️ My FREE 5-day Data Career Kickstart course fixes that. 🔥

You can access the FREE course:

Join HERE

While everyone is arguing about AI models, Postgres is building the actual data layer that makes AI reliable in production, supporting vector search, agent memory, and production RAG pipelines. And it runs on the same open source database all in SQL.

I just got to be part of POSETTE: An Event for Postgres organized by the Postgres team at Microsoft, and it was incredible.

These were my 3 favorite sessions:

↳ From Queries to Agents: how Postgres is becoming the backbone for AI agent data retrieval and why it matters for production

systems

↳ Why We Built Azure HorizonDB: Microsoft’s new cloud-native Postgres service built specifically for AI workloads

↳ pg_duckdb in Action: running analytics directly inside Postgres without ever moving your data

If you missed the event, all the sessions are available on replay here:

Watch Here!

⚠️ Content warning ⚠️

This post contains references to violence, slurs, and self-harm language in the context of AI-generated content.

I've worked with more AI tools than most people have heard of

And I love talking about the efficiencies AI brings and the cool new things it can do. But I try to focus on the foundations and concepts rather than just chasing trends and hyping everything that drops.

Yes, some things are cool proof of concepts. But I wouldn’t post tutorials on tools with no guardrails just because they’re trendy, because the average person reading my content isn’t going to know the difference. And that’s exactly the problem.

What I will always do is speak out when something is alarming because people need to understand this to have a real and honest picture of AI.

That’s what I’m here to do today.

I’m not here to blast a company publicly. What I’m here to do is talk about something that shook me, because I think it says a lot about where AI actually is right now versus where people think it is.

Dangerous hallucinations are a much bigger deal than your chatbot just confidently citing a fake study. And having no guardrails is a product choice, NOT an accident, and how a company responds when something goes wrong matters just as much as the thing that went wrong in the first place (sometimes more).

Here’s what happened to me.

I was working on a brand deal with a well-known AI avatar company for a paid campaign.

Not a small one, but a well-known company. I was hired to make a video for $7,000, and I was excited about it.

I went through the full approval process: concept, script, filming, and multiple rounds of revisions. They asked for one final edit: record myself saying a single line so their tool could capture my voice, transcribe it, and produce an AI avatar of me saying it.

Simple enough.

Except my dumba** forgot to turn my microphone on (yes, really).

The file I uploaded was 17 seconds of complete silence. But there were no words, no audio, NOTHING intelligible at all. Here’s a video of the faulty audio I recorded and submitted:

What came back

Even though the audio was completely blank and null, the voice-to-text model returned a transcription anyway.

Except I hadn’t said anything. So what it produced was in absolutely no way a reflection of the audio I gave it.

It was full of cuss words, references to violence, and self-harm language. I’m not going to show you it in full, but I need you to understand what I actually saw on my screen.

There were phrases like “k** yourself.” It was calling me a “c**t” and a “b***h.” There were multiple slurs and explicit descriptions of violence directed at a person. When this appeared on my screen at first, I felt like I was being hacked or personally threatened.

I feared for my own safety. Then I had to step back and think about what was actually happening underneath the product, and I realized: this was a hallucination. Which was still terrifying but comforting knowing there wasn’t a real human threatening me on the other end of it.

Why it happens

ASR models (Automatic Speech Recognition, the tech that converts spoken audio into text) are trained on human speech. Their whole job is to find patterns in audio and produce a transcription.

When you feed them silence or a near-null file, they don’t stop and say, “I didn’t hear anything.” They do what the model is designed to do, regardless: predict what’s most likely.

They generate whatever sequence of words is statistically most likely, with nothing real to anchor them to. That’s what hallucination actually is at the model level. It’s a model throwing darts at a wall because there’s nothing to ground it. And when there’s no input at all, it can produce anythinggggggg… and it depends on the training data and guardrails. Here’s what was produced:

The output made absolutely no sense. It was WAY longer than anything that could ever fit in a 17-second voice note, and it was full of content that had no business being produced EVER especially since it’s a product anyone can use without any AI literacy.

What I did not understand was what happened next.

The response that made me walk

When I brought this concern to the company’s marketing team that I’ve been working with on the campaign, they accused me of having a TV on in the background that produced the content. I was horrified they were blaming me and thought I was watching something so vulgar on TV. Why are we blaming the victim? I sent them the full video recording of my session, screenshots, and EVERYTHING, but they still blamed and questioned me.

Then I filed a separate ticket through customer support, explained I was a creator working on a paid campaign with them, and asked to escalate to their safety team. They responded with tips on how to use the product better. When I told them I had no interest in continuing and asked if they had even looked at the images I had sent, they still didn’t show any concern.

There was no apology, no accountability, and not even a “thanks for letting us know, we’re looking into this.”

So I walked away from the partnership and money. All the work was done, but I knew I wasn't comfortable promoting this to my audience.

Honestly, it was not a hard decision. My integrity isn't worth $7,000.

The part that kept haunting me

Even after I’d already made my decision, I kept coming back to this.

Even if I HAD intentionally typed those harmful messages (which again, this was a hallucination from a user mistake), a product that turns them into a realistic avatar of a real person should NEVER allow that output.

This type of content that is threatening, dangerous, deceptive, and manipulative, rendered in a realistic human face and voice that can be used to cause real harm to real people.

What’s why content moderation and safety guardrails are so important. They sit in between what a model can generate and what a user actually sees. They catch harmful content before it’s even rendered, regardless of whether it came from a hallucinating model or someone with bad intentions. That layer is NOT optional; it’s essential in AI products.

The fact that it can happen from a USER ERROR, from an accidentally blank audio file, makes the absence of that guardrail even less excusable.

Cool demos are not the same as responsible products.

Every technology fails eventually. The question is what a company has built to catch it and how they respond when something slips through.

This one failed both tests.

Before you partner with, promote, or build on any AI product: ask what happens when it breaks. Not if, when. Does the company have a real escalation path? Do they take accountability? Or do they question the user and blame them?

I walked away from $7,000, and I’d do it again.

Bye BDEs 💅🏼

Jess Ramos 💕

⚡️ Social Highlights:

⚡️If you’re new here:

💁🏽‍♀️ Who Am I?

I’m Jess Ramos, the founder of Big Data Energy and the creator of the BEST SQL course and community: Big SQL Energy⚡️. Check me out on socials: 🔗YouTube, 🔗LinkedIn, 🔗Instagram, and 🔗TikTok. And of course subscribe to my 🔗newsletter here for all my upcoming lessons and updates— all for free!

Got BDE?⚡️

Hi data baddies 💅🏼

DATABRICKS LOVERS!!! Guess what? Your data lakehouse isn’t slow because of compute… It’s slow because your pipelines are a MESS. I know it’s tough to hear this 😅

Your data might not be AI-ready if you have batch jobs that run every few hours, missing data sources, & legacy code no one has maintained in years. Just because it’s in a lakehouse, doesn’t mean you have quality data to support AI, Analytics, and big decision making.

On June 25th at 10 am ET, CData is hosting a live webinar to share how to connect to hundreds of sources in Databricks, build CDC (Change Data Capture) pipelines that deliver near real-time incremental updates directly into Delta Lake and Iceberg tables, and orchestrate your pipelines end-to-end without needing a separate tool to manage it all.

If you’re interested in keeping Databricks fed with fresh, reliable data… this one’s for you!!!

Register Here!!

This year, I'm so excited to be going to Cannes

And honestly, there are so many reasons why.

Yes, it’s in the south of France. It’s going to be beautiful, beachy, and super bougie. And it will be full of croissants, cappuccinos, wine, and sunshine. Some of my favorite things in the world 😍. But more importantly, Cannes has become the hottest event of the year for creators like me.

The Cannes Lions International Festival of Creativity has always been the biggest stage in the world for marketing, advertising, and creative work. But over the past couple of years, it’s been growing FAST in the B2B marketing and creator space, which is genuinely exciting for creators like me who do B2B marketing and post educational content

The best way I can describe it is that this is our version of an influencer trip.

Typically, influencers get sent to Bora Bora on private jets by makeup and beauty brands. When you’re a “boring” B2B creator, that’s not really your life. You get sent to tech conferences, which, hey, I absolutely love, so I’m not complaining. But Cannes Lions is different. It’s become the one place where the ultra elite in the creator economy and B2B marketing actually end up in the same room every single year. Top-tier creators, influencer marketing managers, and C-suite executives in marketing.

Everybody who’s going is SOMEBODY.

So I knew I was going no matter what.

I booked my flight and hotel months ago with zero confirmed opportunities. It was a huge investment of time and money that I wasn’t sure I’d get back directly. But I knew the relationships I’d build, the people I’d meet, and the experiences I’d have would be absolutely priceless. Not just for the business, but for me personally, to stay motivated and inspired by being around really awesome people.

And once I committed? The opportunities started rolling in.

The three things I’m most excited about:

1. I landed a speaking gig with LinkedIn.

This one is really full circle for me. I started posting on LinkedIn a little over four years ago, before it was cool. It was genuinely harder back then. There was more friction, more challenges, and a lot less support for creators. But LinkedIn has become such a great place for creators now, and I’m really grateful they’ve seen the work I’ve been putting in over the years.

I get to speak in front of LinkedIn brands, partners, and customers about what it actually looks like to build a creator business as an educational creator. They specifically want me to talk about my authenticity and transparency. Which, if you know me, you know I don’t have a filter. I talk about things most people won’t touch, like getting fired and every salary I’ve ever made. I’m really honored, and this is going to be a huge deal for me. You can grab more details and RSVP here.

LinkedIn Creator Exchange

2. I'm co-hosting a Women in AI lunch with four incredible women in AI & partnering with Glance & Microsoft. We're bringing together AI founders, creators, and marketing executives for an intimate lunch to talk about the future of AI and actually build real connections. If you're a cool AI person who's going to be there, we'd love to have you. It’s happening June 22, 12–2 PM at Café Glance in Cannes. Seats are limited, so apply for a spot here. 👇🏻

Women in AI Lunch

3. I'm filming content again with Webflow, the hottest discoverability product out there (hello AEO!). I’ve worked with Webflow several times before, and they are genuinely some of the nicest, most creative people I’ve gotten to partner with. Their product is also just really awesome. Full transparency: I’m actually starting to build my own website in Webflow right after Cannes, once things slow down this summer, so this felt like exactly the right timing.

What I actually packed

Okay, here’s my full prep list because I know you’re going to ask.

Tech gear, obviously, I have to take you guys with me! Makeup and beauty stuff for touch-ups on the go because the content never really stops. All my cute outfits, mostly from Anthropologie, Banana Republic, and American Eagle.

For heat prep, I went all in: tons of sunscreen, a portable Dyson fan, and an umbrella specifically to protect myself from the sun because I am NOT letting Cannes give me sun damage. Liquid IV to stay hydrated. And deodorant, because nobody’s first impression of me is going to be sweaty and gross.

And my daily diet, which I’m fully expecting to be bread, cheese, sparkling water, wine, and cappuccinos. I genuinely have zero complaints about this.

Why am I telling you all of this

Cannes Lions is important to me for a lot of reasons, but the biggest one is this: I made a bet on myself before I had any proof it would pay off. I booked the trip before the speaking gig, before the brand deals, before the event was even an idea.

There’s a version of that bet every creator at every level. The room you walk into before you feel ready. The investment you make before the ROI is guaranteed. For me, this year, that’s Cannes Lions.

I CANNOT WAIT.

If you’re going to be there, please reach out. I would love to say hi in real life. And if you’re a brand or brand partner who’s going to be in Cannes or has contacts there, let me know too.

See you there. I can't wait to pop the rosé. 🥂

Bye BDEs 💅🏼

Jess Ramos 💕

⚡️ Social Highlights:

⚡️If you’re new here:

💁🏽‍♀️ Who Am I?

I’m Jess Ramos, the founder of Big Data Energy and the creator of the BEST SQL course and community: Big SQL Energy⚡️. Check me out on socials: 🔗YouTube, 🔗LinkedIn, 🔗Instagram, and 🔗TikTok. And of course subscribe to my 🔗newsletter here for all my upcoming lessons and updates— all for free!

Got BDE?⚡️

Every time you stream a video or load an app, CDNs are processing billions of requests and generating massive amounts of log data in real time. Most teams sample that data just to keep up, but sampled data creates blind spots like the exact anomaly that caused the outage or the traffic pattern signaling a problem.

This matters even more as AI agents enter the picture. Agents can detect anomalies, trace root causes, and trigger fixes automatically. But AI is only as good as the data it’s given!!

On June 25, I’m hosting a free live practitioners summit with engineers from Hydrolix, GlobalDots, and AWS on what agentic CDN operations actually looks like in production— including a live demo.

We’ll talk about:

• Why CDN teams still struggle with visibility during critical incidents

• Why full-fidelity data matters for AI-driven operations

• How AI agents can detect anomalies and trace root causes

• And we’ll show a live demo of the workflow in action

Register Here!

Future of Data: 2026 Trends you MUST know!!

The world of data analytics is moving FAST 🔗. By 2026, professionals won’t just be cleaning data and building drag & drop dashboards— they’ll be expected to leverage AI tools, engineer analytical pipelines, define business logic through code, and prove their value through real‑world projects rather than certificates. The video below breaks it down in depth ⬇️

Here are the 3 major trends I’m seeing in data analytics going into 2026:

1. AI‑Driven Analytics

AI is becoming a built‑in feature in everything from spreadsheets to visualization tools. ChatGPT, Claude, GitHub Copilot, and other AI assistants are already helping analysts write queries, interpret results, and generate insights more quickly and accurately than ever. In just a couple of years, companies will expect their data teams to fully embrace and implement these tools into workflows rather than just see them as novelties (or competition😂). Analysts who learn to drive results, automate tedious tasks, and add insights to their own judgement with AI will be the most valuable on the job market.

Those who don’t adopt AI will be left behind and become irrelevant very quickly on the job market!

Leveraging AI video!

2. Analytics Engineering

The days when data analysts were praised for their ability to drag and drop charts into a dashboard are OVER. Nowadays companies want analytics engineers who can manage an analytical pipeline and define business logic for reporting. They’re looking for people who can build and maintain data models, implement business logic into code and data, and QA the results on a deeper level. This more technical role sits between data engineering and analytics and requires fluency in SQL 🔗. Many AE roles also require skills in Python and dbt— which I believe will explode even more next year.

The most sought‑after analysts will be those who can design data transformations and automate reporting rather than just visualize outputs. TLDR: companies will want all analysts to be able to code and do technical work.

Learn SQL video!

3. Certificates are losing their weight

Basic certificates used to be a golden ticket into 6-figure analytics roles because no one had them, and they were enough to impress hiring managers. But many of these programs teach outdated tools like R or proprietary software, and these don’t match the technologies companies actually use in industry. Employers care about proven skills and hands‑on experience way more than a shiny certificate with no backbone. Data Analytics portfolios, personal projects, and work experience speak louder than certificates, and this will only increase next year. Hiring managers are going to value projects and practical knowledge over the number of courses and certifications you’ve taken.

So stop taking 100 different certificates and start practicing!

Portfolio video!

What to do about it?

In 2026, data analytics will be more intertwined with AI, engineering, and real‑world problem solving than it ever has before. To stay relevant, start experimenting with AI tools today, invest in your technical skills, and build projects that showcase what you can actually do. The future is bright— but only for those who level up and adapt with the times.

Keep that BDE rolling this year 💅🏼

Jess Ramos 💕

⚡️ Social Highlights:

⚡️If you’re new here:

💁🏽‍♀️ Who Am I?

I’m Jess Ramos, the founder of Big Data Energy and the creator of the BEST SQL course and community: Big SQL Energy⚡️. Check me out on socials: 🔗YouTube, 🔗LinkedIn, 🔗Instagram, and 🔗TikTok. And of course subscribe to my 🔗newsletter here for all my upcoming lessons and updates— all for free!

Got BDE?⚡️

Hi data baddies 💅🏼

Before we get into the tea… if you're trying to break into data (or level up and finally get PAID), my FREE 5-Day Data Career Kickstart is the exact thing I wish someone had handed me when I was starting.

Join Here!

POSETTE: An Event for Postgres2026

Most people still think of Postgres 🔗 as a relational database. But in 2026, it’s doing A LOT more than rows and columns… but it’s all still the same SQL you already know and love. Modern Postgres handles:

↳ vector embeddings and similarity search for AI apps via pgvector
↳ semi-structured JSON data that changes shape frequently via JSONB
↳ full-text search with relevance ranking built right in

No separate databases or new query languages— just Postgres doing things most people don’t know it can do. If you wanna learn more, you have to come to POSETTE: a FREE virtual event for Postgres (June 16-18) with 44 talks from the engineers and open source contributors who actually build Postgres.

Sessions cover pgvector, RAG pipelines, Postgres 19, performance, extensions, and the newly announced HorizonDB: Microsoft’s new cloud-native Postgres service built specifically for AI apps.

Register here for access to live sessions and recordings:

Register Here!

I grew my data analyst salary from $72K to $153K

I’ve had many data analyst job titles in my career. I’ve job hopped, advocated internally for promotions, and I’ve even been FIRED. I’ve been there, done that, and seen it all.

LinkedIn shows you the highlights— the job titles and companies, but it doesn’t show how people got there. So how did I negotiate all my salaries and move on up? We’re unpacking it all today.

All of these jobs and companies prepared me for who I am today: A data & AI expert/influencer/entrepreneur. And I’m so proud I am where I am and happier than ever.

But first, receipts 🧾

Ya’ll know I LOVE transparency, so here’s every single corporate data job and salary I’ve ever had. I’m also telling you which ones I negotiated and when I changed companies):

• Predictive Analytics Intern - $7/hour

• Server - $5/hour

• Graduate Assistant - $1400/month + tuition waived

• Data Analyst Intern - $17/hr (negotiated)

• Data Analyst - $71,500 + $20K bonuses

• Senior Data Analyst - $73,000

• Data Analytics Manager - $95,000 (negotiated internally)

• Senior Risk Analyst - $130,000 + $9,000 bonus (Negotiated & changed companies)

• Senior Data Analyst - $153,000 (Negotiated & changed companies)

• Data/AI Entrepreneur - MORE 😉 (fired and started over)

Now let me tell you what those numbers DON’T show you. Let’s start with company #1 after I graduated with my masters in Business Analytics.

Company 1: The startup that felt like family (until it didn’t)

My first real corporate job was as a data analyst at a small fintech startup. It was small enough for everyone to know everyone. And I loved it at first. I was a little baby intern with zero experience, and I latched onto every relationship in that building.

But the thing about “family” companies… I was getting phone calls on a SUNDAY about an urgent Power BI report. And I didn’t know how to set a single boundary, because I was young and I genuinely thought that’s just what you did. And tbh there’s never anything THAT urgent about a Power BI report 😅

I was also doing WAY more than my title. I was managing someone, leading sales calls, and building the data our execs used to take to investors for a capital raise, all while making $72K. So I asked for a raise to six figures. And they said no. At least I tried to shoot my shot haha!

I was crushed. But honestly, I also didn’t come prepared, and a tiny part of me didn’t think I was worth it yet. But they did change my title to Senior Data Analyst which was a decent win.

So the next year, I came back with receipts. I wrote a literal ONE-PAGER with every single one of my accomplishments in the past year and the market rate pulled from multiple sources, and I made it impossible to say no. I made the perfect business case for my raise and had it clearly documented.

They gave me the Data Analytics Manager title and a 29% raise to $95K 🙌 It definitely made me feel valued especially since I was managing people and so many projects.

And after three years of hard work, I got bored because I wasn’t learning anymore, and I quit to go to a new job.

Company 2: The “prestigious” job I knew was wrong on day one

Next, I jumped to a big government-sponsored enterprise as a Senior Risk Analyst. $130K plus a $9K bonus. On paper, an absolute dream. I was really excited to try a big corporation out and have a big name on my resume. I used competing offers to negotiate up 37% from my previous salary, because letting the market tell you your worth hits SO different than guessing a number in an interview and praying.

I only lasted four months.

And I'm not exaggerating, I knew on my first day. It was a giant company with endless red tape. Slow, traditional mortgage industry. The complete opposite of startup life. I hated every second of it.

Nobody was actually working, everyone was just in meetings pretending to work.

By this point, I was building my brand on LinkedIn on the side, and my manager noticed. He started questioning how I was getting all these opportunities, and then he straight-up started mocking my LinkedIn posts. And let me tell you, having your older male boss mock your social media does NOT feel good as a young woman.

He HATED having me there, and I hated being there. So in just four months, I quit for my next job.

And here’s the thing everyone is terrified of: quitting a job in under a year did NOT ruin my career. Everyone says you have to stay at a job for a year, but sometimes you have to leave sooner to be happy. It is not ideal to leave a job in 4 months, but it needed to be done. And I’m happy that I tried something new even if it wasn’t the best long term. That terrible fit taught me something fast, and it led me straight to the best-paying job I’d ever had.

Company 3: the one I loved, the one that fired me

Then I landed the $153K job. It was a Senior Data Analyst position at a more established tech startup, small enough to still move fast but big enough to feel safe. I got to lead so many projects, and for a while, it was genuinely the best job I’d had.

I was learning a ton about project management, AB testing, creating metrics, and more.

It was really good until it wasn’t.

I didn’t get a SINGLE raise or promotion the entire two and a half years I was there. And slowly, the vibe shifted. Around 1.5 years in, I started receiving long passive-aggressive Slack messages about tiny mistakes here and there. I wasn’t allowed to take PTO anymore. And I was set up to fail constantly… and then documented and scolded.

I can’t say it was because of my social media, but the timing is suspicious. Once I started partnering with huge companies like AWS, Intel, and Atlassian, and speaking and traveling and doing cool things… I think it put a target on my back.

One week, my one-on-one with my manager got “weirdly” moved to a random day. HR was on the call, and in a seven-minute Zoom, I was fired for “performance.”

I cried, I was in shock. For the six months leading up to that day, I’d been in therapy every single week, just learning how to disassociate enough to get through the workday. I was the unhealthiest I had ever been.

The next morning after being fired I woke up, and it felt like a BREATH of fresh air. I felt like I was free. I’d spent those six awful months quietly protecting myself: legally, financially, emotionally, so when the day finally came, I was prepared.

In fact, I flew to Portugal 44 hours later for a data creator meetup and then hopped on a 2 week cruise to plan what I was going to do next.

What nobody tells you about job hopping

The part that the salary screenshots leave out is that every jump up that ladder was really just me learning the same lesson over and over: nobody was going to value me until I valued myself first.

I undersold myself at my first job because I didn’t think I was worth it. I let a family company blur every boundary I had because I didn’t know how to say no. I was chasing the shiny title & company that I ended up hating. And I stayed loyal for two and a half years to a place that was quietly building a case to push me out.

Staying loyal and quiet didn’t pay me. Negotiating & job hopping did. And making myself VISIBLE did. Weirdly enough, getting fired did too. Now I work for myself, I’ve replaced that corporate salary many times over, and I haven’t applied to a single job in YEARS.

And I hope I never have to again because I’m genuinely so happy with what I’m doing now.

So if you’re sitting in a job right now feeling underpaid, overlooked, or like the walls are slowly closing in… I’m not telling you to rage-quit tomorrow.

I’m telling you to protect yourself, get clear on your worth, and start making yourself the obvious choice, so the next move happens on YOUR terms instead of theirs.

Negotiate. Get your bag, queen. And be loyal to yourself.

Bye BDEs 💅🏼

Jess Ramos 💕

⚡️ Social Highlights:

⚡️If you’re new here:

💁🏽‍♀️ Who Am I?

I’m Jess Ramos, the founder of Big Data Energy and the creator of the BEST SQL course and community: Big SQL Energy⚡️. Check me out on socials: 🔗YouTube, 🔗LinkedIn, 🔗Instagram, and 🔗TikTok. And of course subscribe to my 🔗newsletter here for all my upcoming lessons and updates— all for free!

Got BDE?⚡️

Hi data baddies 💅🏼

The data analyst role is fundamentally changing, and I got to hear all about it at Tableau Conference (which was so cool since I started my career in Tableau!!)

Data analysts aren’t just building dashboards anymore. They’re becoming agentic architects, and here’s what you need to know:

1. Knowledge is the new data because AI agents can’t act on data alone. Every metric and business definition needs to be clearly defined. Tableau’s Knowledge Engine is built on 33 million semantic models that make AI trustworthy at scale.

2. Dashboards aren’t the final destination anymore. Through headless analytics and secure MCP connections, insights now show up directly in your current workflows like Slack, Teams, Claude, & ChatGPT.

3. Anyone can now talk to their data. Conversational analytics brings natural language into Tableau across Cloud, Server, and Desktop. No SQL required.

Tableau Conference was EPIC, you have to come next year!! Catch up on all the amazing sessions here:

HERE!

I sat down with the CMO of Tableau, and she said something that stuck with me: decision-making is actually HARDER in the AI era, not easier.

Anyone can ask a question and get an answer. The problem is knowing if that answer is actually right for YOUR business. Watch the full interview to see why that gap is where everything lives right now 👇🏻

I finally made it to Tableau Conference

And it completely LIVED up to the hype. I've been wanting to go to Tableau Conference for YEARS.

Like, since I was a brand new analyst with no idea what I was doing. Because Tableau is where I started. It's the tool I was fumbling through when I was just trying to build my first dashboard and not embarrass myself.

So when I got the invitation this year, I said yes IMMEDIATELY. Duh!

It really exceeded every single expectation I had.

I knew within about five minutes of walking in that this was going to be one of my favorite conferences I've ever attended. The Tableau community is something else. Everyone was genuinely excited to be there, so warm and friendly, willing to talk to total strangers. Everybody there was a true power user who genuinely loves the product, and you could feel that in every conversation, every session, every event.

I got VIP access to product demos and exclusive tours, and I even got to meet the DJ who performed at Data Night Out (which was such a fun night). But beyond the fun stuff, getting to actually see all the new AI features and capabilities in action (not on a slide deck, but working in real time in front of me) made everything land so differently.

What struck me most was how much Tableau has evolved since I was a junior analyst using it to build reports. Back then, the job was making charts and dashboards. Now I’m watching it become the infrastructure layer for AI agents. That’s a genuinely wild thing to witness if you’ve been around long enough to remember when “working in Tableau” just meant dragging dimensions to the rows shelf.

My interview with the CMO, Rekha Srivatsan

This was one of the highlights of the entire trip. She’s incredibly sharp and refreshingly direct, and she gave me one of the most honest executive answers I’ve ever heard in a while. No fluff.

Is AI actually going to replace data analysts?

Short answer: no.

Rekha’s argument is that AI has given everyone access to answers, but access to answers is not the same as access to the right answers. Someone still needs to understand the business context, define the metrics correctly, and make sure what the AI is producing is actually accurate and grounded in reality. That person is the data analyst.

Her take: the role isn’t shrinking. It’s becoming more significant than it’s ever been.

“Your role is now way more significant in organizations than it ever will be.”

The rise of the "agentic architect"

This was the part of the conversation I found most interesting because it gets at where the analyst role is actually heading, not just what it is today.

The idea is that data alone isn’t enough for AI to make real business decisions. AI needs knowledge. The context and definitions are specific to how a particular business operates. Not generic industry terms; your company’s definitions.

She gave a concrete example from Salesforce: ACV (Average Customer Value) has a specific meaning at Salesforce that differs from how other companies define it. Their fiscal year starts in February instead of January. These things seem small, but without that context encoded somewhere, an AI agent is essentially guessing about your business.

The data analyst who understands those definitions and can layer that knowledge into the system? That’s what she’s calling the agentic architect or knowledge architect. The person who bridges raw data and trusted AI output.

“Every business, every organization has their own definition around how they talk about their business... that’s where knowledge comes in.”

Is decision making easier or harder in the AI era?

I love this question because I expected the safe corporate answer, and I did not get it.

She said decision making is harder. 1000% harder. And she explained why in a way I thought was genuinely honest.

Everyone has access to AI tools now. You can ask the same question to ChatGPT, Gemini, and Claude and get three different answers. So instead of spending time finding information, you’re spending time figuring out which answer is actually correct. That’s a new cognitive burden that didn’t exist before.

She shared a story about a senior leader who got frustrated waiting on data from his team and just built his own dashboard. It looked great, beautiful visualization, impressive design. It was also inaccurate. And he was using it to make business decisions before anyone caught it.

“It’s not the data, it’s not the access to answers — it’s the right answer at the right time.”

That line stuck with me.

What happens when insights stop living in dashboards?

We’re at an inflection point, and this part of the conversation made it real.

The future isn’t going to be a separate tool to look at data, it's headless analytics!. It’s data showing up in the tools where you already work (Slack, Teams, Claude, ChatGPT) through secure MCP connections. Rekha described this as being “embedded in the flow of work,” and it’s exactly what headless analytics is making possible.

She was also refreshingly honest about the limits: she said you can probably do about 40% of your analytical work comfortably in other tools, but if you need to go one layer deeper or make a high-stakes business decision, you still need a governed, trusted platform underneath it all. The insight can live in Slack. The decision-making foundation still needs to be solid.

The coolest way she’s ever used AI

This was my favorite answer of the whole interview, and I wasn’t expecting it.

Her 14-year-old daughter, a self-described non-planner, used ChatGPT to plan the family’s entire trip to Africa this June. Every day is mapped out. Every lunch and dinner. Reservations included. A full itinerary built by a teenager who doesn’t typically plan anything.

“She went to ChatGPT and built out this whole itinerary for us... I thought that was remarkable.”

For someone who leads AI strategy at one of the biggest companies in tech, her most mind-blowing AI moment was watching her kid use it to do something she genuinely wouldn’t have done on her own. I thought that was such a human and grounded answer. It reminded me that the most powerful thing about AI isn’t the enterprise use cases, it’s what it unlocks for regular people doing regular things.

Tableau Conference reminded me of why I got into data in the first place. The community, the curiosity, the people who care deeply about making sense of information and actually using it to do something, it’s all still there. It’s just evolved into something much bigger.

If you’ve been thinking about attending, come next year. Seriously. You don’t want to miss it.

Bye BDEs 💅🏼

Jess Ramos 💕

⚡️ Social Highlights:

⚡️If you’re new here:

💁🏽‍♀️ Who Am I?

I’m Jess Ramos, the founder of Big Data Energy and the creator of the BEST SQL course and community: Big SQL Energy⚡️. Check me out on socials: 🔗YouTube, 🔗LinkedIn, 🔗Instagram, and 🔗TikTok. And of course subscribe to my 🔗newsletter here for all my upcoming lessons and updates— all for free!

Got BDE?⚡️

Hi data baddies 💅🏼

Most people grind for months trying to break into data and never land the role. Usually, because they are focused on the wrong things. 🙅🏻‍♀️ My FREE 5-day Data Career Kickstart course fixes that. 🔥

You can access the FREE course:

Join HERE

If your data lives in SQL Server, there’s a good chance it’s on-prem (meaning it’s sitting on physical servers inside your company, not in the cloud!!). And that’s fine, a lot of enterprise data still lives there.

But if your analytics stack is in Snowflake, you’re missing a lot of critical data, and it can be hard to connect those two worlds. Most teams end up with fragile pipelines, manual workarounds, and custom code that someone has to maintain forever.

CData Sync closes that gap without starting a HUGE engineering project. It replicates SQL Server data to Snowflake using log-based CDC (Change Data Capture), so you’re only moving what changed every time. So no unnecessary load on production and more importantly… no stale data!! Game changer for analytics.

Their team is doing a free live demo on May 20th at 10 am ET, and you’ll learn how to build no-code ingestion pipelines from 270+ sources (databases, SaaS apps, APIs) in minutes and how to replicate SQL Server data to Snowflake with log-based CDC.

Register Here!

I'm a technical person who has coded for years...

And I was TERRIFIED of the terminal. 💀

Which sounds ridiculous, but hear me out. I like coding. I CAN code! But the terminal always felt like a different LEVEL of coding. Heavy-duty, no guardrails, developer-tier stuff I had no business touching.

I am not a developer, and I want that to be very clear.

So when Claude Code first launched, and I had to use the terminal to set it up, I almost didn’t do it. But I did. And I genuinely could not believe what I had been sleeping on!

I’ve since built entire data cleaning projects from scratch, complete with a messy dataset, answer key, and a full readme. I’ve made videos about it. I’m OBSESSED. And here’s what most people don’t realize: Claude Code isn’t just for writing and debugging code. It does SO much more than that.

Today I’m breaking it all down AND sharing my top 5 tips to actually become a Claude Code power user.

If the terminal scares you, I promise you can do it. And if you get stuck, just ask Claude for help. DUH. 💁🏻‍♀️

The single biggest mistake people make with Claude Code is jumping straight to execution.

You give it a vague goal, it starts doing things, and somewhere along the way, it goes in a direction you did not intend. Now you are spending more time fixing the output than you would have if you had done it yourself.

Plan Mode fixes this ENTIRELY. Before Claude touches a single file, you map out the full task together. You review the steps, approve the approach, and then Claude executes. ✨

To activate it: Hit Shift + Tab until you see 'plan mode on', enter your goal, and Claude will walk you through a structured plan before doing anything.

This matters ESPECIALLY for multi-step tasks. One wrong assumption early on, and suddenly everything downstream is a mess. We've all been there.

Here’s the difference: a vague prompt means Claude is guessing, filling in gaps, and making decisions you didn’t sign off on. A clear, approved plan means it executes end-to-end exactly the way you actually wanted.

Use Plan Mode every time you start a multi-step or ambiguous task, and I promise you will notice the difference immediately.

Set Up Your CLAUDE.md File

If you are re-explaining your project to Claude every single session, you are wasting time that compounds. Every time you start fresh, you are losing context that Claude needs to do good work, like your stack, your conventions, your rules, and your preferences. 😅

The fix is a CLAUDE.md file, which gives Claude permanent memory about your project.

There are two places you can put it:

~/.claude/CLAUDE.md ← global rules that apply everywhere
/your-project/CLAUDE.md ← project-specific context

The global file is where you put things that are always true: how you like code formatted, tools you always use, things you never want Claude to do.

The project-specific file is where you document your

• Stack

• Folder Structure

• Any conventions that are specific to that codebase

• Anything Claude needs to know to work in context

Once this is set up, Claude already knows your project before you type a single prompt. You stop repeating yourself. This is one of those things that feels like a small setup investment and pays off immediately and continuously. I HIGHLY recommend doing this for every project you plan to return to more than once.

Connect Your Tools with MCP

Most workflows do not live in one place. You are pulling context from GitHub, referencing Notion docs, checking Slack threads, managing Jira tickets, and right now, you are probably doing all of that manually. 🫣

MCP (Model Context Protocol) lets Claude connect directly to those tools so it can pull context from all of them at once, without you having to act as the middleman.

To set it up:

claude mcp add <name> <command>  
example: claude mcp add notion npx -y @notionhq/notion-mcp-server

Think about what that actually means.

No more pasting in your Notion doc or copying over context you definitely forgot half of. Claude already has it!

Once set up, Claude pulls context directly from your tools so no more copy-pasting between apps. And the output gets BETTER because Claude is finally working with the full picture instead of whatever you remembered to throw into the prompt at the last second.

Build Skills for Recurring Tasks

Here is a pattern that happens constantly: you explain how to do a specific task to Claude, it does it well, and then in the next session, you explain it all over again and again and again. 😬

Skills solve this. A skill is a set of instructions packaged as a slash command. Save it once at ~/.claude/commands/your-task.md, then invoke it any time by typing /your-task, no re-explaining required.

The file structure looks like this:

~/.claude/commands/your-task.md

Think about the tasks you do on repeat. Generating reports in a specific format, running the same type of analysis, and following the same review process for PRs every single time.

If you have explained the same task to Claude more than twice, that is your sign. 👀

Package it ONCE and stop wasting your time re-explaining yourself like Claude has amnesia. You already did that work, so you should never have to do it again.

Run Parallel Sessions

This one feels almost too simple, but it is genuinely one of the most powerful changes you can make to your workflow.

Claude Code can run in multiple terminal windows simultaneously. That means while Claude is working on one task, you can have a second session working on a completely separate task at the same time. You are not waiting, and you’re not idle.

Two things are happening in parallel.

Terminal window 1 → Task A
Terminal window 2 → Task B

Before: one task at a time, Claude sitting there idle while you review output or context-switch like it’s 2019.

After: two windows running, output doubling, and you’re spending your time on ACTUAL decisions instead of just waiting around.

Note: parallel sessions consume tokens from both windows simultaneously, so this works best on Pro or higher plans.

The only rule is that your tasks need to be genuinely independent. Different files, different features, different analyses. If task B needs task A to finish first, keep them sequential. That’s just logic. 😉

But the second you have parallel workstreams, run parallel sessions. It is one of the fastest ways to multiply what you get done with Claude Code, full stop.

Here’s the thing nobody tells you about Claude Code.

Most people use it like a fancy autocomplete. Type a thing, get a thing, move on. And yeah, it works. But you are leaving SO much on the table.

The five tips I just walked you through are not productivity hacks. They are the difference between a tool that saves you ten minutes and a tool that changes what you can actually get done in a day. 💪🏻

Plan Mode means you think before you break things. CLAUDE.md means it never forgets what you told it. MCP means you stop switching between tools like a maniac. Skills mean you stop repeating yourself. And Parallel sessions mean you stop waiting.

Together, it’s not just faster. It’s a completely different way of working.

Claude Code is only as powerful as how you use it. And now you know how to actually use it.

We are in the middle of a real shift in how technical work gets done,

and Claude Code is one of the clearest examples of what that shift actually looks like in practice. The people who learn to delegate work to agents are the ones who are going to move the fastest.

The gap between people who use AI well and people who use it casually is about to become very obvious. ⚡️

Bye BDEs 💅🏼

Jess Ramos 💕

⚡️ Social Highlights:

⚡️If you’re new here:

💁🏽‍♀️ Who Am I?

I’m Jess Ramos, the founder of Big Data Energy and the creator of the BEST SQL course and community: Big SQL Energy⚡️. Check me out on socials: 🔗YouTube, 🔗LinkedIn, 🔗Instagram, and 🔗TikTok. And of course subscribe to my 🔗newsletter here for all my upcoming lessons and updates— all for free!

Got BDE?⚡️

Hi data baddies 💅🏼

Most people grind for months trying to break into data and never land the role. Usually, because they are focused on the wrong things. 🙅🏻‍♀️ My FREE 5-day Data Career Kickstart course fixes that. 🔥

You can access the FREE course here:

Join here!

Did you miss Microsoft’s FREE event, Azure Cosmo DB Conf? If so, don’t worry, you can access all the recordings here. This event was made for devs by devs with REAL demos of how to build production applications on NoSQL to support AI.

If you’re not familiar with NoSQL, it’s a flexible database model built for data that doesn’t fit neatly into rows and columns like a regular SQL database— think JSON documents, key-value pairs, and graphs. NoSQL was designed to scale horizontally across distributed systems, which is exactly what modern AI applications need.

OpenAI’s John Lee shared how ChatGPT executes over 1.4 trillion transactions DAILY on Cosmos DB 🤯 and the architecture patterns that make it possible at that scale!

All sessions are on demand NOW, watch here!

And you can get a 100% FREE voucher for the DP‑420 Azure Cosmos DB Developer Specialty if you complete the new Cosmos Conf Skills Challenge by May 8th.

PS. Peep my cameo appearance 👀 (don’t worry I played it super cool seeing myself pop up live on a MICROSOFT conference 😮‍💨)

Your AI strategy is only as good as your data

And most companies’ data is a mess. Everyone wants to talk about AI, but nobody wants to talk about what’s underneath it (data). 😅

I get it, AI has incredible marketing and data quality work does not. It’s unglamorous work, and nobody claps for it.

But here’s what is actually happening at companies that are frustrated with their AI results: they built on top of a broken foundation, and they’re surprised that the house is shaky.

What AI actually does with your data

Before we talk about what can go wrong, it helps to understand what is actually happening under the hood (without getting too technical about it). 😉

There are two main ways AI systems use your data:

Training and Fine-Tuning

The model learns patterns from your historical data. If that data is biased, incomplete, or inconsistent, the model inherits THAT too. It does not know the data is bad. It just learns from whatever you feed it.

RAG (Retrieval-Augmented Generation)

Instead of learning from data upfront, the AI pulls from your documents and databases in real time and uses them to generate a response. If those sources are outdated or disorganized, the AI will confidently give wrong answers in polished, authoritative language that sounds completely correct.

In both cases, the model treats your data as ground truth. It does not second-guess the inputs. That is a feature when your data is good. It is serious liability when it’s not.

The 5 ways bad data quietly wrecks your AI

You might have to send this part to your manager. 😅 Bad data is not always obvious, and it shows up in AI systems in very specific, very costly ways.

1. Incomplete Data: Your churn prediction model is missing engagement history for 40% of customers. The outputs look confident, but they’re not accurate.

2. Inconsistent Data: Three teams have three different definitions of “active users” living in the three different systems. The AI picks one and runs with it, and now two-thirds of your org is arguing about numbers that are all technically wrong.

3. Stale Data: Your internal AI assistant is pulling from a product knowledge base that hasn’t been updated in 18 months. Every answer is a snapshot of a product that no longer exists.

4. Biased Data: A hiring AI trained on historical hiring decisions from a team that skewed toward a specific candidate profile will replicate that bias at scale. Thousands of times and faster than any human ever could.

5. Siloed Data: The AI only sees what one team fed it. It has no idea what’s living in another department’s system. Its understanding of your business is partial at best.

This is how trust in AI breaks down

Bad data in a spreadsheet is annoying. Bad data in an AI system is a completely different level of problem.

When AI scales the output, it scales every flaw in your data with it. You end up producing thousands of wrong answers, automatically, wrapped in language that sounds authoritative and polished.

Your users will trust it at first. Then they’ll catch a few wrong answers. And once trust is gone, it is incredibly hard to earn back. That’s the REAL cost of skipping data quality work. Not the cleanup; it’s the erosion of trust in every AI-powered thing you build on top of it.

What Good Data Foundations Look Like

Let’s not be confused, good data foundations do not mean perfect data. It means trustworthy data. There’s a real difference, and chasing perfection is honestly one of the reasons this work never gets done.

Here is what actually matters:

• You know what data you have and what it means. This sounds basic, but it’s shockingly rare! Documented definitions and someone who actually owns and maintains them are the foundation that makes everything downstream possible.

• Your data is trustworthy enough to make decisions from. It doesn’t need to be perfect; it should be trustworthy. Are the gaps documented? Are the known issues flagged? Is there a process for catching problems before they compound? That level of ownership is what separates companies that get real value from AI from the ones getting confident-sounding hallucinations.

• The AI has visibility into the data it needs. Accessibility matters just as much as data quality, and context is everything! If the right data exists somewhere but the AI cannot get to it because it lives in a siloed system or a folder nobody has touched in three years, it might as well not exist.

The real talk

The companies that will win with AI long-term are not the ones that adopted it fastest. They are the ones who treated their data infrastructure as a strategic asset before AI made it impossible to ignore.

Your AI strategy is downstream of your data strategy. It always has been. The difference now is that the consequences of getting it wrong are a lot more visible and a lot more expensive.

So before you green-light the next AI initiative, ask what the data underneath it actually looks like. That question will save you a lot of pain. 😅

Bye BDEs 💅🏼

Jess Ramos 💕

⚡️ Social Highlights:

⚡️If you’re new here:

💁🏽‍♀️ Who Am I?

I’m Jess Ramos, the founder of Big Data Energy and the creator of the BEST SQL course and community: Big SQL Energy⚡️. Check me out on socials: 🔗YouTube, 🔗LinkedIn, 🔗Instagram, and 🔗TikTok. And of course subscribe to my 🔗newsletter here for all my upcoming lessons and updates— all for free!

Got BDE?⚡️

Hi data baddies 💅🏼

Stop wasting time learning the wrong things! 🙅🏻‍♀️ I just launched my FREE 5-day Data Career Kickstart course to teach you the #1 skill you actually need to land a high-paying data role right now. 🔥

You can access the FREE course here:

Join HERE

SQL isn’t always the solution.

SQL is built for structured data with rows, columns, and a defined schema. It’s great for transactional data, reporting, and consistency.

NoSQL is built for semi or unstructured data like JSONs, documents, key-value pairs, graphs. It’s great for flexible schemas, distributed systems, real-time speed, and vector search for AI applications.

Want to see NoSQL in production? Azure Cosmos DB Conf is April 28th— FREE and virtual (recordings available after). There will be speakers from OpenAI, Vercel, and Walmart. Plus there will be real code & real demos— built for developers by developers!! I’ll be there speaking too.

Save your seat!

Data isn’t just numbers in a spreadsheet!

And honestly, for a long time, that was basically true. But the data landscape has completely shifted.

I’ve talked to so many data professionals who completely freeze when a stakeholder puts them on the spot to “pull some insights” from customer feedback. You open your SQL editor, stare at a massive blob of text in a single column, and panic because you have NO idea what to do next. 💁🏻‍♀️

But here is the reality: DATA IS EVERYWHERE. It isn’t just numbers in a spreadsheet anymore. It’s words, images, audio, and emojis. Organizations have to quickly process all of this to pull insights and power AI.

To make sense of it, you need to understand the two main categories data falls into:

Structured Data

This is the traditional data you already know. Think Excel files, Google Sheets, and SQL databases. It is highly organized information that adheres to a predefined schema, and it lives in relational databases (RDBMS), making it incredibly easy to search, pull, sort, and analyze using SQL.

What it looks like:

• Spreadsheets with organized rows and columns.

• Customer databases containing names, addresses, and contact information.

• Financial transactions such as bank deposits or credit card swipes.

How it’s used:

Structured data powers traditional business intelligence and analytics. This is what you use to analyze sales figures, track inventory levels, and monitor financial performance; this is where SQL shines.

It is absolutely critical, but here is the catch: structured data only captures what you have already decided to track. You define the schema up front, so anything that doesn’t fit neatly into a column just isn’t captured. It tells you WHAT happened, but almost never WHY.

It also doesn’t scale well for capturing the full, messy reality of human behavior. And as organizations lean harder into AI, structured data alone is nowhere near enough to build anything meaningful.

Unstructured Data

This is the messy, human side of data. It doesn’t follow a set layout, which makes it incredibly difficult to organize and analyze using your usual SQL database methods. But while it lacks formal structure, it is packed with context that structured data simply cannot capture.

What it looks like:

• Social media posts (Tweets, IG captions, LinkedIn shares, emojis)

• Email and chat logs

• Multimedia content (YouTube videos, podcasts, scanned documents)

• Server logs, website clickstreams, or IoT sensor data (these can technically be structured, but the actual content inside them acts unstructured)

How is Unstructured Data Actually Used?

Unstructured data is no longer just a messy byproduct of doing business. It is the core fuel for modern analytics and AI. According to multiple reports, 80% to 90% of all enterprise data generated today is unstructured.

If you only know how to query tables, you are missing the entire picture. Here is how companies are actually using it right now:

• Powering GenAI & RAG: When a company builds an AI chatbot that actually knows their specific business, they’re feeding it unstructured data. LLMs are trained on massive text outputs, but companies use RAG to connect AI directly to their internal PDFs, CRM logs, and internal knowledge bases (think shared wikis or documentation portals where teams store internal guides and policies).

• Deep Sentiment & Customer Intent: Structured data tells you a customer gave you 2 stars. Unstructured data tells you WHY. By processing call center transcripts and support tickets with NLP, businesses can extract their customers' actual emotions at scale.

• Predictive Maintenance: It isn’t just text; it’s machine logs. Industrial equipment and airplanes generate massive logs of unstructured data. Machine learning models analyze this data alongside unstructured technician notes to spot failure patterns before something actually breaks.

• Advanced Clinical Insights: In healthcare, the most critical context is rarely found in a structured dropdown menu. It lives in unstructured medical images (X-rays, MRIs), physician notes, and patient discharge summaries.

• Automated Legal & Financial Parsing: AI models can now ingest these unstructured documents to pull out specific clauses, numbers, and legal precedents in SECONDS!

The Real Reason You Need to Understand Both

Here’s the thing nobody is saying out loud: the entire industry is moving towards building AI-powered products and applications. Companies aren’t just analyzing data anymore; they’re using it to build things. Chatbots, recommendation engines, internal tools, predictive systems. All of it runs on data.

And the data professionals who are going to be MOST valuable aren’t necessarily the ones who can build the pipelines themselves. They’re the ones who understand the full landscape well enough to know what kind of data a problem calls for, what tools and infrastructure are needed, and how to speak intelligently with the engineers and ML teams doing the building.

If you only understand structured data, you can only contribute to part of the picture. The closer you can get to understanding how unstructured data powers AI products, the more irreplaceable you become to the teams shipping them.

Unstructured data is where AI lives. Understanding it is how you stay relevant to the work being done around you.

Learn to tame the unstructured mess, and everything else follows. ✨

Bye BDEs 💅🏼

Jess Ramos 💕

⚡️ Social Highlights:

⚡️If you’re new here:

💁🏽‍♀️ Who Am I?

I’m Jess Ramos, the founder of Big Data Energy and the creator of the BEST SQL course and community: Big SQL Energy⚡️. Check me out on socials: 🔗YouTube, 🔗LinkedIn, 🔗Instagram, and 🔗TikTok. And of course, subscribe to my 🔗newsletter here for all my upcoming lessons and updates— all for free!

Got BDE?⚡️

Hi data baddies 💅🏼

As you guys know, I have launched my new FREE 5-day Data Career Kickstart course, and this course helps you start mastering the #1 skill you need to land high-paying data roles, so you stop wasting time learning the wrong things! 🚀

You can access the FREE course here:

Join HERE

The model wasn’t wrong— the data feeding it was 3 days old, and it was confidently working with stale data.

This is the part nobody talks about when they’re hyping up RAG and AI agents— the pipeline underneath still matters:

↳ Overnight batch loads feeding a system that needs real-time accuracy
↳ Manual refresh workflows that break and nobody notices
↳ Legacy tooling that was built before AI exploded

You need to know which pipelines to modernize first and how to move from batch to incremental replication without blowing up what’s already working.

Join me live w/ Kim Fessel and Manish Patel from CData to get into exactly this: how to move from batch to incremental replication, when CDC is the right call, and what separates teams that modernize incrementally from teams stuck in planning mode.

Save your seat here!

CTEs in SQL

CTEs are one of the most important skills you need to go from beginner to intermediate, and they’re one of the most commonly misunderstood. I used them for years at my first startup job without knowing what they were called. Then I walked into interviews and got asked about CTEs like it was nothing, and I had NO IDEA what the interviewer was talking about. I was literally using CTEs every day at work and did not know the name of it. That knowledge gap probably cost me real opportunities, so today I am making sure that does not happen to you. 😉

What is a CTE?

CTEs are a common table expression, and in simple terms, a CTE is a temporary result set defined within the scope of the query. Think of it as a table you build with a query, give a name, and tell SQL to hold onto for later. You can then reference it just like any regular table, as many times as you need, but only within that same query.

Here is the basic syntax:

WITH cte_name AS (
  SELECT *
  FROM table1
)

SELECT *
FROM cte_name;

Why do you need CTEs?

There are two situations where CTEs come up: when you WANT to use them, and when you absolutely HAVE to.

1. When you WANT to use them:
   Sometimes you opt into a CTE purely for readability, even if a subquery or regular query would technically work. I do this all the time when writing queries that will end up in a dashboard or get handed off to a teammate. CTEs let you define your logic one step at a time, which makes the code significantly easier to follow. CTEs are great for breaking up complex logic into different steps and building the logic one step at a time (as opposed to putting everything into one massive query!). It is just a nice thing to do for the people who have to read your work later, including the future you.

2. When you HAVE to use them:
   The most common example is nested aggregate functions. SQL does not allow you to nest aggregate functions directly, so writing AVG(SUM(column)) will throw an error. A CTE is the fix; you calculate the sum first, then take the average in the outer query:

   WITH order_counts AS (
     SELECT
       user_id,
       SUM(ordered) AS num_orders
     FROM table1
     GROUP BY 1
   )
   
   SELECT AVG(num_orders) AS avg_per_user
   FROM order_counts;

Another really common situation is working with data at different levels of granularity in the same query. Say you have orders at the day level, but you also want to include the company’s monthly total in the same row. You build a CTE that aggregates by month, then join it back to the daily dataset:

WITH monthly AS (
  SELECT
    DATE_TRUNC('month', order_date) AS month,
    COUNT(*) AS monthly_orders
  FROM orders
  GROUP BY 1
)

SELECT
  o.order_id,
  o.order_date,
  m.monthly_orders
FROM orders o
LEFT JOIN monthly m
  ON DATE_TRUNC('month', o.order_date) = m.month;

Try doing that without a CTE or a subquery, and I guarantee you will have a hard time. 😅

Benefits of CTEs:

1. They are reusable.
   And it’s one of their biggest benefit. They’re really useful if you have to refer to an output multiple times in the same query. They’re only usable within that query, though. So if you don’t rerun the CTE, you won’t be able to use it.

2. They are readable.
   A really important part of coding is making it useful long-term, especially if someone else takes over or needs to edit your work. CTEs make it much easier to understand code because you define output and sections one part at a time. This gives the reader the advantage of slowly digesting and interpreting your code one step at a time.

3. They can be recursive.

   Although I haven’t had many use cases for this in the real world, it is a more advanced programming concept and is worth knowing and learning. This is where CTEs can stand out over subqueries since subqueries don’t have this capability.

CTEs vs. Subqueries

A CTE is essentially the inside-out version of a subquery, and almost every CTE can be rewritten as a subquery and vice versa, with recursive CTEs being the exception.

So when do you pick one over the other? There is not always a clear answer, but here is my general rule: if the logic is complex, reused multiple times, or being handed off to someone else, go with CTE. If it is a simple one-off lookup embedded in a larger query, a subquery works fine.

In my day-to-day work, I regularly write queries with three to five CTEs and sometimes more. Some are mandatory because the logic requires it, and others I add simply to keep things clean and readable for my team.

CTEs are one of the most powerful tools in your SQL toolkit, and the analysts who truly understand them, not just how to write them, but when AND why to use them, are the ones who stand out in interviews and on the job.

Master this, and you will feel the difference immediately. And if you're ready to fast-track that mastery, don't forget to grab your spot in my FREE 5-day Data Career Kickstart course! 😉

Bye BDEs 💅🏼

Jess Ramos 💕

⚡️ Social Highlights:

⚡️If you’re new here:

💁🏽‍♀️ Who Am I?

I’m Jess Ramos, the founder of Big Data Energy and the creator of the BEST SQL course and community: Big SQL Energy⚡️. Check me out on socials: 🔗YouTube, 🔗LinkedIn, 🔗Instagram, and 🔗TikTok. And of course, subscribe to my 🔗newsletter here for all my upcoming lessons and updates— all for free!

Got BDE?⚡️

Hi data baddies 💅🏼

These are my LEAST favorite questions to answer as a data analyst:

1. Can you pull this data real quick?

2. Why don’t these numbers match up?

3. Why is X down this month?

Most companies are drowning in data, but can still barely answer simple questions in a reasonable amount of time. I love Databox’s AI Analyst (Genie!) because it automates the analysis process by seamlessly combining multiple data sources and helping you build anything (metrics, reports, dashboards) with a single prompt. It pulls from your actual metrics and historical trends and gives you a clear, contextual answer in seconds— all grounded in your real data.

They also just launched a Databox MCP, which connects your live business data directly into your AI workflows. AI-driven analytics is the future, and leveraging tools like Databox is a great way to get accurate insights straight from the source of truth quickly.

Try out Genie the AI Analyst

You NEED to know which Data Role is right for you.

I have helped thousands break into tech and become data professionals. And sometimes someone comes to me describing exactly what they want from a career, and I IMMEDIATELY know that it is not the role they think it is. 💁🏻‍♀️

Picking the wrong path costs you. No, literally. One of my students spent $10,000 on a bootcamp and came out the other side with no clear direction on which role actually fit him.

That’s why I am breaking down five roles in data so that you would know what you actually do every day, what skills you need, what the salary looks like, which tools you will use, and how AI is changing each role. By the end, you will know exactly which path is right for you.

The 5 roles at a glance

Before we get into the details, here’s a quick way to think about each role:

• Data Analyst: translates business questions into insights using SQL and BI tools

• Data Scientist: builds statistical models and does predictive work using Python and math

• Data Engineer: builds the infrastructure and pipelines that everyone else depends on

• Analytics Engineer: sits between data engineering and data analytics, transforming raw data into clean, reliable datasets

• Business Analyst: bridges the gap between business teams and data teams without deep technical work

Data Analyst

• Tools: SQL, Tableau, Power BI, Excel, Google Sheets, Python (occasionally)

• Salary: Entry-level is at $65K-$85K | Senior-level is at $120K-$164K+

• AI Impact: Analysts who leverage AI to work more efficiently and who have strong soft skills are THRIVING right now. AI is not replacing data analytics; it is changing how the work gets done.

They are the people answering the business’s most important questions, such as: Why has churn increased over the past month? and Why are sales down for this customer this year? You are translating business questions into SQL queries, pulling insights from the data, and communicating those findings back to stakeholders.

SQL IS THE CORE SKILL; it’s roughly 80% of the job. You also need a BI tool like Tableau or Power BI for visualization, Excel or Google Sheets for reporting, and strong communication and business acumen to make your analysis actually useful to the people receiving it. This role is the PERFECT blend of technical and business skills, which is honestly what makes it so fun! ⚡️

Data Scientist

• Tools: Python (Pandas, Scikit-Learn, PyTorch, TensorFlow), SQL, Jupyter Notebooks, and Spark

• Salary: Entry-level is at $95K-$120K | Senior-level is at $150K-$250K+

• AI Impact: AutoML platforms are making model-building more accessible, which is actually increasing the need for data scientists who can evaluate those outputs, fine-tune them, and understand what is happening under the hood.

Data science is the role where expectations and reality diverge the most, and I say this with love! Many people think data scientists spend their days building cutting-edge machine learning models. In reality, a significant portion of people with data science titles are doing work that looks a lot more like data analysis with occasional predictive modeling.

That said, true data science work involves building statistical models, doing hypothesis testing, running regressions, and translating probabilistic outputs to non-technical stakeholders. The real differentiator is the math; if statistics does not excite you, this is not the right path. 👀

Data Engineer

• Tools: Python, SQL, dbt, Spark, Airflow, AWS/GCP/Azure, Snowflake, Databricks, BigQuery, Redshift

• Entry-level Salary: $90K-$110K | Senior-level Salary: $160K-$200K+

• AI Impact: AI is accelerating the volume of data being created, which means demand for data engineers is going up, not down. With the rise of agentic AI, data engineers also have to figure out how to give AI agents secure and reliable access to data infrastructure.

Data engineers build the infrastructure that makes everyone else’s job possible. They connect data sources, build pipelines, and make sure data flows correctly from every system into one place so that analysts, scientists, and business analysts can actually access it. Without data engineers, you have messy data or no data at all.

This is the most technical role on this list by a significant margin, and that is reflected in the salary. It is also the role most likely to require on-call rotations, and if a vital pipeline breaks at 2 am, you are the one who has to fix it. Yes, at 2 am.😭

Analytics Engineer

• Tools: dbt, SQL, Snowflake/BigQuery/Databricks, Python (occasionally), Looker, and Git

• Entry-level Salary: $85K-$110K | Senior-level Salary: $140K-$200K

• AI Impact: As AI tools make it easier to generate SQL and data transformations, analytics engineers who deeply understand data modeling principles and can QA AI-generated outputs are becoming more valuable, not less.

This is the newest of the five roles and also one of the most misunderstood. Analytics engineers sit between data engineering and data analytics, and they take raw data from pipelines and transform it into clean, reliable, well-documented data sets that analysts can actually use.

If a data engineer builds the highway, an analytics engineer builds the on-ramp. They are the ones maintaining the data models, writing transformation logic in tools like dbt, and making sure the business logic is encoded correctly so that every analyst pulling from those tables is working with trustworthy data. If this sounds like your sweet spot, it’s an incredible role to be in right now!

Business Analyst

• Tools: Excel, Google Sheets, SQL (especially at mid to senior level), Tableau, PowerPoint

• Entry-level Salary: $55K-$75K | Senior-level Salary: $110K-$140K

• AI Impact: AI is actually making business analysts more capable. They can now do technical work more quickly with AI assistance. But the human judgment layer is what cannot be replaced: navigating company politics, translating ambiguous requirements, and managing difficult stakeholders.

Business analysts are the most misunderstood role on this list. They are not deep in the code; they are the translators between business teams and data teams. They gather requirements from stakeholders, document processes, and define what success looks like, and help determine what the technical work should actually accomplish.

The barrier to entry is lower than that of any of the other four roles, which makes it a great starting point for people who want to work in data without heavy coding. The trade-off is a lower salary ceiling, BUT it is one of the most accessible paths into the data world, and the soft skills you build here are valuable EVERYWHERE. 🙌

How to Decide Which One is Right for You

You need to be asking yourself these questions:

• Do you want to build things from scratch or solve existing problems?

• How much do math and statistics excite you?

• Do you want to code all day, or use code as one of several tools?

• Do you care more about the salary ceiling or the day-to-day enjoyment of the work?

• Are you okay being on call, or do you need clear boundaries between work and personal time?

I have watched people spend YEARS in the wrong role because it looked good on a resume or paid a little more at entry level. The ones who build the best careers are the ones who were honest with themselves early. About what they like, their skills, and what kind of work actually energizes them.

Get that part right first, and everything else follows. ✨

Bye BDEs 💅🏼

Jess Ramos 💕

⚡️ Social Highlights:

⚡️If you’re new here:

💁🏽‍♀️ Who Am I?

I’m Jess Ramos, the founder of Big Data Energy and the creator of the BEST SQL course and community: Big SQL Energy⚡️. Check me out on socials: 🔗YouTube, 🔗LinkedIn, 🔗Instagram, and 🔗TikTok. And of course, subscribe to my 🔗newsletter here for all my upcoming lessons and updates— all for free!

Got BDE?⚡️

Hi data baddies 💅🏼

If you want to get your hands on this FREE mini-course that helps you start mastering the #1 skill you need to land high-paying data roles, you’ve got to move FAST and stop wasting time learning the wrong things!

You can access the FREE 5-day Data Career Kickstart course HERE!!

7 Essential SQL Window Functions You Need to Master

EVERY data analyst needs to master them. Not just know how to write, but actually understand what they are doing, why they work, and how to explain them clearly under pressure. Today, I am walking through my top 7, with real examples and the exact logic behind each one. 😉

What is a Window Function?

It lets you perform a calculation across a set of rows related to the current row without collapsing all of those rows into a single result. This is what makes them fundamentally different from aggregate functions. When you use SUM or AVG with GROUP BY, you lose the detail of every individual row because everything gets collapsed into one aggregated output. Window functions solve for that; keep the row-level detail while still performing calculations across groups. That context is what makes them so powerful for real-world analysis.

Every window function follows the same basic structure: the calculation, followed by OVER, and then two optional components inside the parentheses. PARTITION BY splits your data into groups, and ORDER BY defines the order of rows within each group. Whether to use one, both, or neither depends entirely on the business problem you are trying to solve.

The 3 Ranking Functions

Let’s walk through each one using a customer orders dataset as our example. Notice that we are ordering our data by the number of orders in descending order (highest to lowest).

ROW_NUMBER() assigns a unique sequential number to every row regardless of ties. It does not care if two rows have the same value; it just keeps counting. In our table, Customers B, C, and D all have 450 orders, but they get row numbers 2, 3, and 4 with no consideration for the fact that they are tied.

SELECT
  Customer,
  Orders,
  ROW_NUMBER() OVER (ORDER BY Orders DESC) AS Row_Number
FROM customer_orders;

RANK() does account for ties and assigns the same ranking value to tied rows. Notice how Customers B, C, and D all tied with 450 orders, so they all receive a rank of 2. But after the tie, RANK() skips values. Because three rows occupied positions 2, 3, and 4, the next customer in line (Customer E with 400 orders) is assigned a rank of 5.

SELECT
  Customer,
  Orders,
  RANK() OVER (ORDER BY Orders DESC) AS Rank
FROM customer_orders;

DENSE_RANK() also accounts for ties and assigns the same value to tied rows, but it never skips numbers or leaves gaps in the ranking values. After Customers B, C, and D tie for 2nd place, the next rank in the sequence is simply 3. So, Customer E (400 orders) gets a dense rank of 3.

SELECT
  Customer,
  Orders,
  DENSE_RANK() OVER (ORDER BY Orders DESC) AS Dense_Rank
FROM customer_orders;

You can also add PARTITION BY to any of these ranking functions to restart the ranking within each group. For example, if you added a Store_Location column to this dataset, partitioning by location would mean each store gets its own independent 1, 2, 3 ranking rather than ranking customers across the entire company.

SELECT
  Customer,
  Store_Location,
  Orders,
  RANK() OVER (PARTITION BY Store_Location ORDER BY Orders DESC) AS Location_Rank
FROM customer_orders;

Comparing all 3 side-by-side

When deciding which ranking functions to use, look at how they handle ties. In the graphic below, Customers B, C, and D all have 450 orders. Notice how each function treats that tie group differently:

• ROW_NUMBER() treats every row as unique, assigning sequential numbers (2, 3, 4) even though the order value is identical. It’s perfect for when you just need a unique ID or want to grab the first record in a set.

• RANK() recognizes the tie, so B, C, and D all get a rank of 2. However, it leaves a gap afterwards. Because three rows occupied the 2nd, 3rd, and 4th positions, the next customer (E) jumps straight to 5th place.

• DENSE_RANK() gives the tie group the same rank (2), but it doesn’t skip any numbers. The next customer in line (E) is simply assigned 3rd place.

SUM and AVG as Window Functions

You are probably already familiar with SUM and AVG as aggregate functions, but when you pair them with OVER and ORDER BY, they become something much more useful: a cumulative sum and a cumulative average.

Cumulative Sum with SUM()

Adding OVER and ORDER BY to SUM() creates a cumulative total that updates with every new row. This is incredibly useful for tracking a customer’s total orders over time.

SELECT
  Customer,
  Order_Date,
  Orders,
  SUM(Orders) OVER (ORDER BY Order_Date) AS Cumulative_Sum
FROM customer_orders
ORDER BY Order_Date;

Note: When multiple rows share the same date, they all receive the same running total: the end-of-day cumulative sum for that date. Then the next date picks up where the previous one left off.

Cumulative Average with AVG()

The same logic applies to AVG(). Instead of a cumulative sum, you now have a cumulative average that updates as each new date is added. By the time you reach the final row, the cumulative average equals the overall average of the entire dataset.

SELECT
  Customer,
  Order_Date,
  Orders,
  AVG(Orders) OVER (ORDER BY Order_Date) AS Cumulative_Avg
FROM customer_orders
ORDER BY Order_Date;

If we add PARTITION BY Customer, both calculations will run independently within each customer’s data. This means you can track a running total or cumulative average per customer rather than across the entire dataset:

SELECT
  Customer,
  Order_Date,
  Orders,
  SUM(Orders) OVER (PARTITION BY Customer ORDER BY Order_Date) AS Customer_Cumulative_Sum,
  AVG(Orders) OVER (PARTITION BY Customer ORDER BY Order_Date) AS Customer_Cumulative_Avg
FROM customer_orders;

LEAD and LAG

These are two of my personal favorites, and they are incredibly useful once you start working with time-series data. LEAD() grabs the value from the next row, and LAG() grabs the value from the previous row.

Together, they make it very easy to compare the current row to what came before or after it, which is exactly what you need for tracking order history, calculating time between events, or spotting trends per customer.

In this example, we are using PARTITION BY customer so that LEAD and LAG operate independently within each customer’s data. This means that the first and last rows reset per customer. Customer A’s earliest order will always have a null LAG value, and their most recent order will always have a null LEAD value.

SELECT
  Customer,
  Order_Date,
  LEAD(Order_Date, 1) OVER (PARTITION BY Customer ORDER BY Order_Date) AS Next_Order_Date,
  LAG(Order_Date, 1) OVER (PARTITION BY Customer ORDER BY Order_Date) AS Previous_Order_Date
FROM customer_orders;

The number after the column name, which in this case is 1, specifies how many rows ahead or behind you want to look. For daily data, use 7 to compare the same day from the prior week, or adjust the offset to match whatever interval your analysis requires.

Why this matters beyond interviews

Window functions are not just an interview topic. They are one of the most practical tools in a data analyst’s toolkit. Once you understand them, you will find yourself reaching for them constantly: ranking customers by spend, identifying the most recent record per user, calculating month-over-month growth, tracking cumulative metrics over time.

The analysts who can write these fluently and explain the reasoning behind them are genuinely more valuable to their teams, and hiring managers know that.

Bye BDEs 💅🏼

Jess Ramos 💕

⚡️ Social Highlights:

⚡️If you’re new here:

💁🏽‍♀️ Who Am I?

I’m Jess Ramos, the founder of Big Data Energy and the creator of the BEST SQL course and community: Big SQL Energy⚡️. Check me out on socials: 🔗YouTube, 🔗LinkedIn, 🔗Instagram, and 🔗TikTok. And of course subscribe to my 🔗newsletter here for all my upcoming lessons and updates— all for free!

Got BDE?⚡️

Hi BDEs 💅🏼

IT’S FINALLY HERE.

My new mini-course, The Data Career Kickstart, is officially LIVE ⚡️— and enrollment is 100% free.

Here’s a sneak peek of what you’ll learn inside:

• Why SQL is the #1 skill to break into data

• How to analyze data without overcomplicating it

• The 3 SQL commands that appear in 80% of queries

• What portfolios hiring managers actually look for

• The exact steps to land your first 6-figure data role (even if you’re a complete beginner)

My team and I spent 50+ hours putting this together over the past month 😅 so I’m so excited to finally share it with you.

All you have to do to get free access is click on the link below and sign up:

Click here to get free access to the mini-course — and join 2,101+ data baddies who already got access to it!

I built this because I know what it feels like to want a data career but have absolutely NO IDEA where to begin, and I wanted to make sure that’s never the reason you don’t go after it.

See you on the inside 🥳

Jess Ramos 💕

P.S. This course is completely free, but don’t sleep on it. If you’ve been wanting to break into data but keep getting stuck, Lesson #1 alone is going to change how you think about the whole thing. Get your free spot here.

Bye BDEs 💅🏼

Jess Ramos 💕

⚡️ Social Highlights:

⚡️If you’re new here:

💁🏽‍♀️ Who Am I?

I’m Jess Ramos, the founder of Big Data Energy and the creator of the BEST SQL course and community: Big SQL Energy⚡️. Check me out on socials: 🔗YouTube, 🔗LinkedIn, 🔗Instagram, and 🔗TikTok. And of course subscribe to my 🔗newsletter here for all my upcoming lessons and updates— all for free!

Got BDE?⚡️

Hi data baddies 💅🏼

Exciting update! This month, I’m launching a brand new, FREE mini-course to help you start mastering the #1 skill you need to land high-paying data roles.

Want early access? You can join 1248+ others on the waitlist HERE!

Your portfolio might be the reason you’re not getting callbacks, and most people have no idea! In this video, I’m breaking down exactly how to build the portfolio that got me THREE six-figure data analyst roles, from picking the right projects to the read me format that hiring managers actually want to see. All in here 👇🏻

5 Data Cleaning Functions in SQL

If there is one thing that surprises people most about working in data, it’s how much time goes into cleaning it. We are talking about 80% of your time on the job. It’s also one of those skills that everyone acknowledges is important, but very few people actually teach well.

That’s exactly what we are covering today. I am walking you through a complete SQL data cleaning workflow, step-by-step, starting with a real messy dataset and ending with a clean one. You are welcome in advance! 😉

First, let's talk about the dataset

I generated this dataset in ChatGPT, and if you’re not already using AI to create synthetic datasets for practice, I highly recommend it. This dataset contains 1,000 rows of customer data and has many issues common in real-world data.

In practice, data ends up messy for several reasons:

• It is entered manually by humans across different teams.

• It is pulled from multiple source systems that each have their own formatting standards.

• It has been collected over a long period of time as those standards evolved and changed.

The results usually look something like this 👇🏻

• FIRST_NAME and LAST_NAME are full of null values, and random single/double quotes around name values

• EMAIL column has a significant number of null values

• STATE column with inconsistent values of full state names, abbreviations, uppercase, lowercase, and combinations of all of them

• SIGNUP_DATE column with a mix of regular dates, datetimes, extra quotes, and trailing spaces

Let’s walk through how to clean each of these.

Function 1: REPLACE- Remove Special Characters from the Name Columns

The first step is to remove all of the single and double quotes from the name columns. To do this, we use the REPLACE function, which takes three arguments:

• The column you want to modify

• The character you want to remove

• What you want to replace it with

Since we need to remove both single and double quotes, we nest two REPLACE functions together. The inner one removes single quotes first, and the outer one removes double quotes from the result. Nesting functions can look intimidating at first, but they become second nature very quickly, I promise!

Once the names are clean, we use CONCAT to combine the first name and last name into a full name column. In some SQL dialects, if either name is null, CONCAT will return null for the entire full name. To prevent that, we wrap each name in COALESCE, which always returns whatever information is available.

SELECT
  first_name,
  last_name,
  REPLACE(REPLACE(first_name, '''', ''), '"', '') AS first_name_clean,
  REPLACE(REPLACE(last_name, '''', ''), '"', '') AS last_name_clean,
  CONCAT(COALESCE(first_name_clean, ''), ' ', COALESCE(last_name_clean, '')) AS full_name_clean
FROM messy_customers;

Step 2: CAST- Standardize the Date Column

Dates are consistently one of the most frustrating things to clean in SQL. I say that from personal experience! 😅 This dataset has regular dates, datetimes, dates wrapped in double quotes, and dates with trailing spaces, all stored in the same column.

We apply the same nested REPLACE approach to strip the extra characters first, then cast the result as a DATE value to remove all timestamps and standardize the format. You can do this using the traditional CAST function, which is CAST(value AS DATE), or using the ::date shorthand notation, which is only available in some SQL dialects.

SELECT
  signup_date,
  REPLACE(REPLACE(signup_date, '''', ''), '"', '')::date AS date_clean
FROM messy_customers;

Here’s how it looked before: “2023-01-08 07:32:38” with a quote and trailing space, and here it is after: 2023-01-08. Every value is now a clean, consistent date.

Function 3: COALESCE- Replace Null Emails with a Default Value

Null values in an email column can break aggregations, cause problems in downstream reporting, and create confusion for stakeholders. Nobody wants to explain to their manager why the numbers are off because of a null value that slipped through!

COALESCE returns the first non-null value from a list of inputs, which makes it ideal for replacing nulls with a meaningful default. If the email field is null, we replace it with the string ‘unknown’ so that anyone working with the data immediately understands that the email address was not captured.

SELECT
  email,
  COALESCE(email, 'unknown') AS email_clean
FROM messy_customers;

Function 4: CASE WHEN- Recode the State Column into Consistent Values

This one is tedious, but ABSOLUTELY necessary. If your dataset has “California,” “CA,” “ca,” and “california” all in the same column, your database treats them as four separate categories. This produces incorrect calculations and groupings anywhere that field is used.

We use a CASE statement to recode all variations into one consistent uppercase abbreviation. To reduce the number of conditions we need to write, we wrap the state column in LOWER first, which means we only need to account for spelling variations rather than every possible capitalization combination.

There is still some manual work involved, but LOWER makes the process significantly more manageable.

SELECT
  state,
  CASE
    WHEN LOWER(state) IN ('ca', 'california') THEN 'CA'
    WHEN LOWER(state) IN ('texas', 'tx') THEN 'TX'
    WHEN LOWER(state) IN ('fl', 'florida') THEN 'FL'
    WHEN LOWER(state) IN ('ny', 'new york') THEN 'NY'
  END AS state_code_clean
FROM messy_customers;

Function 5: ROW_NUMBER()- Identify and Remove Duplicate Records

We deduplicate on the email field since email functions as a unique identifier. Only one person can own any given email address. First, we identify the duplicates using GROUP BY with a HAVING clause, which filters after aggregation and allows us to find all emails that appear more than once.

SELECT
  email,
  COUNT(*) AS duplicate_count
FROM messy_customers
GROUP BY 1
HAVING COUNT(*) > 1;

There are 11 distinct emails, each appearing dozens of times, and one email appears 105 times. Note that null emails are grouped together here, but that does not mean they belong to the same person.

To remove the duplicates, we use ROW_NUMBER(), which assigns a sequential number to each row within a partition. We partition by email and order by customer_id, then use QUALIFY to keep only the row numbered 1 per email group. This retains exactly one record per unique email.

QUALIFY filters directly on window function results without requiring a subquery or CTE. It is only available in certain SQL dialects, so check whether your environment supports it before using it. If you are not

SELECT
  customer_id,
  email,
  ROW_NUMBER() OVER (PARTITION BY email ORDER BY customer_id) AS row_num
FROM messy_customers
QUALIFY row_num = 1;

Putting It All Together

Here is everything combined into a single query:

SELECT
  REPLACE(REPLACE(first_name, '''', ''), '"', '') AS first_name_clean,
  REPLACE(REPLACE(last_name, '''', ''), '"', '') AS last_name_clean,
  CONCAT(COALESCE(first_name_clean, ''), ' ', COALESCE(last_name_clean, '')) AS full_name_clean,
  REPLACE(REPLACE(signup_date, '''', ''), '"', '')::date AS date_clean,
  COALESCE(email, 'unknown') AS email_clean,
  CASE
    WHEN LOWER(state) IN ('ca', 'california') THEN 'CA'
    WHEN LOWER(state) IN ('texas', 'tx') THEN 'TX'
    WHEN LOWER(state) IN ('fl', 'florida') THEN 'FL'
    WHEN LOWER(state) IN ('ny', 'new york') THEN 'NY'
  END AS state_code_clean
FROM messy_customers
QUALIFY ROW_NUMBER() OVER (PARTITION BY email ORDER BY customer_id) = 1;

1,000 rows of messy data, cleaned and deduplicated down to 11 records in a single query.

Why This Skill Actually Matters

Bad data produces bad analysis. It doesn’t matter how sophisticated your model is or how polished your dashboard looks; if the underlying data is not clean, the output cannot be trusted. And presenting inaccurate analysis with confidence is significantly worse than presenting nothing at all.

Data cleaning is not the most exciting part of the job, but it is what determines whether your work is actually reliable. The analysts who take this seriously are the ones who build a reputation for producing work that people can actually depend on. 😉

Bye BDEs 💅🏼

Jess Ramos 💕

⚡️ Social Highlights:

⚡️If you’re new here:

💁🏽‍♀️ Who Am I?

I’m Jess Ramos, the founder of Big Data Energy and the creator of the BEST SQL course and community: Big SQL Energy⚡️. Check me out on socials: 🔗YouTube, 🔗LinkedIn, 🔗Instagram, and 🔗TikTok. And of course, subscribe to my 🔗newsletter here for all my upcoming lessons and updates— all for free!

Got BDE?⚡️

Hi data baddies 💅🏼

If you want to get your hands on this FREE mini-course that helps you start mastering the #1 skill you need to land high-paying data roles, you’ve got to move FAST and stop wasting time learning the wrong things!

You can access the FREE 5-day Data Career Kickstart course HERE!!

Please show my amazing sponsor some love by checking out their event 🔗:

Modern analytics doesn’t fall from bad tools; it fails from bad data. Every data team I talk to is stuck in the same spot: modern analytics layer on top, legacy systems underneath, and a pipeline holding it all together with hope, dreams, and duct tape. CData Software builds the infrastructure that actually fixes this.

Helping teams move data across SQL Server, Microsoft Fabric, Azure Data Lake Storage, and 300+ other sources in a predictable and governed way. They’re running a live demo on March 24th, 10:00 AM EST, that shows what production-grade pipelines actually look like in the real world.

save ur seat here!

Being a dashboard analyst is no longer enough.

A few years ago, you could get hired as a data analyst by dragging a few charts into Tableau, throwing in some vague KPIs, and calling it a day. The pipelines could be broken. The filters didn’t always work, and the data had errors. None of it mattered because companies were so desperate to feel data-driven that they’d take anything that looked like insight.

Data was new and sexy, and companies wanted to say they were data-driven way more than they actually wanted to be data-driven. But those days are OVER.

Here’s the reality check nobody wants to say out loud:

Literally anyone can spin up a decent-looking dashboard with AI now. Stakeholders, interns, product managers, and executives can now make it in a few minutes.

So if building dashboards is your whole thing, you’re already replaceable.

I’m not saying that to be mean, I’m saying it because it’s true and you deserve to know it. The analysts who are actually staying relevant, getting the best jobs, and pulling the biggest salaries right now are not the ones who can make the prettiest visuals. They’re the ones who can do everything else:

⭐️ Translate a business question into an actual analysis
⭐️ Write real SQL, not just SELECT *
⭐️ Understand the data model, not just the dashboard on top of it
⭐️ Find the insight nobody asked for but everybody needed
⭐️ Communicate what the data actually means, not just what it shows

Companies aren’t hiring dashboard monkeys anymore. They’re hiring business translators, strategic thinkers, and full-stack data professionals. The bar is higher now, and that’s honestly a good thing. It means the people who put in the work are going to get very well rewarded for it.

What do you ACTUALLY do with this information?

You stop treating SQL like a nice-to-have and start treating it like the foundation of your entire career. You get comfortable with messy data, hard questions, and stakeholders who don’t know what they want. You build the skills that AI can assist with but can’t replace, such as critical thinking, business context, and knowing when the numbers are lying to you.

The days of drag-and-drop dashboards being enough are OVER. But the opportunity for people who go deeper has never been bigger. 🤷🏻‍♀️

If you’re ready to build those skills, my free mini-course is EXACTLY where to start. It’s built around the #1 thing I see separating analysts who get promoted from the ones who get automated. Want early access? You can join 1134+ others on the waitlist HERE!

Bye BDEs 💅🏼

Jess Ramos 💕

⚡️ Social Highlights:

⚡️If you’re new here:

💁🏽‍♀️ Who Am I?

I’m Jess Ramos, the founder of Big Data Energy and the creator of the BEST SQL course and community: Big SQL Energy⚡️. Check me out on socials: 🔗YouTube, 🔗LinkedIn, 🔗Instagram, and 🔗TikTok. And of course, subscribe to my 🔗newsletter here for all my upcoming lessons and updates— all for free!

Got BDE?⚡️

Hi data baddies 💅🏼

If you want to get your hands on this FREE mini-course that helps you start mastering the #1 skill you need to land high-paying data roles, you’ve got to move FAST and stop wasting time learning the wrong things!

You can access the FREE 5-day Data Career Kickstart course HERE!!

The US government blacklisted an AI company for the first time EVER, Claude said no to the Pentagon, Trump called them left-wing nut jobs, and somehow Katy Perry got involved? 💀

I broke down the ENTIRE story, plus an honest Claude vs ChatGPT breakdown. You can watch it here 👇🏻

How to connect to FastF1 data & build projects

I went to my first F1 race last year in Las Vegas, and left completely hooked!

Seeing the speed and the atmosphere in person is a different experience from watching on TV. So naturally… I had to find a way to combine my two interests. Data and F1, and that’s when I found FastF1.

It pulls real F1 data straight from the official F1 timing API, so you get actual race data. Lap times, sector splits, tyre strategy, pit stops, full car telemetry, weather, race results, and all of it come back as Pandas DataFrames, so it fits right into any Python workflow you’re already running. 🏁

Here’s how I used it in 3, 2, 1, GO!! 🏎️

Step 1: Install + Set Up Caching

Pip install FastF1, then import it alongside Pandas. The first thing you wanna do is enable caching and point it to a local folder. That way, it saves your data there, so you’re not re-downloading everything every single run.

Step 2: Pull Your Session

Pass in three things: year, circuit name, and session type. Here I pulled the 2025 Las Vegas Race: fastf1.get_session(2025, ‘Las Vegas’, ‘Race’) then call session.load () and you pull in all that session’s data.

Step 3: Here’s What You Can Get for Each Session’s Data

• session.results: Is a full DataFrame with every driver, their team, finishing position, and status. If you wanna see who DNF’d or which constructor had a bad day, it’s all right here. (Verstappen won btw 👀)

  

• session.laps: Every lap for every driver. Lap times, sector times, tyre compound, tyre life, pit-in and out times, and stint number.

  

• session.weather_data: Air temp, track temp, humidity, pressure, wind speed, wind direction, and rainfall are all mapped across the session.

  

• And then telemetry: Loop through every driver and every lap, call lap.get_telemetry(), and you get speed, RPM, throttle, brake, gear, and exact X/Y coordinates of the car on track.

  

  

This is real data that’s not only free but actually INTERESTING, and that combo is so rare I had to share it immediately.

Time to try it out with your fave race!

Bye BDEs 💅🏼

Jess Ramos 💕

⚡️ Social Highlights:

⚡️If you’re new here:

💁🏽‍♀️ Who Am I?

I’m Jess Ramos, the founder of Big Data Energy and the creator of the BEST SQL course and community: Big SQL Energy⚡️. Check me out on socials: 🔗YouTube, 🔗LinkedIn, 🔗Instagram, and 🔗TikTok. And of course, subscribe to my 🔗newsletter here for all my upcoming lessons and updates— all for free!