Automated Customer Support

2 ways AI systems today generate smarter answers. I’ve explained both in simple steps below. 𝗥𝗔𝗚 (𝗥𝗲𝘁𝗿𝗶𝗲𝘃𝗮𝗹-𝗔𝘂𝗴𝗺𝗲𝗻𝘁𝗲𝗱 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗼𝗻) (𝘴𝘵𝘦𝘱-𝘣𝘺-𝘴𝘵𝘦𝘱) RAG lets AI fetch and use real-time external information to generate fact-based, updated answers. 1. 𝗦𝘁𝗮𝗿𝘁 𝘄𝗶𝘁𝗵 𝗾𝘂𝗲𝗿𝘆 – User asks a question or gives input. 2. 𝗘𝗻𝗰𝗼𝗱𝗲 𝗶𝗻𝗽𝘂𝘁 – Convert it into a machine-readable format. 3. 𝗧𝗼𝗸𝗲𝗻𝗶𝘇𝗲 𝘁𝗲𝘅𝘁 – Break the query into small understandable pieces. 4. 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗲 𝗲𝗺𝗯𝗲𝗱𝗱𝗶𝗻𝗴𝘀 – Turn text into numeric vectors that capture meaning. 5. 𝗥𝗲𝘁𝗿𝗶𝗲𝘃𝗲 𝗸𝗻𝗼𝘄𝗹𝗲𝗱𝗴𝗲 – Search a vector database for relevant information. 6. 𝗦𝗲𝗹𝗲𝗰𝘁 𝗰𝗼𝗻𝘁𝗲𝘅𝘁 – Pick the most useful retrieved chunks. 7. 𝗙𝗶𝗹𝘁𝗲𝗿 𝗻𝗼𝗶𝘀𝗲 – Remove irrelevant or low-quality data. 8. 𝗙𝘂𝘀𝗲 𝗸𝗻𝗼𝘄𝗹𝗲𝗱𝗴𝗲 – Combine external info with the model’s internal knowledge. 9. 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗲 𝗿𝗲𝘀𝗽𝗼𝗻𝘀𝗲 – Create an answer using both retrieved data and reasoning. 10. 𝗩𝗮𝗹𝗶𝗱𝗮𝘁𝗲 𝗼𝘂𝘁𝗽𝘂𝘁 – Check for factual accuracy and coherence. 11. 𝗥𝗲𝗺𝗼𝘃𝗲 𝗯𝗶𝗮𝘀 – Eliminate misleading or biased phrasing. 12. 𝗗𝗲𝗹𝗶𝘃𝗲𝗿 𝗳𝗶𝗻𝗮𝗹 𝗼𝘂𝘁𝗽𝘂𝘁 – Provide the user with a reliable, fact-backed response. __________________________________________________ 𝗖𝗔𝗚 (𝗖𝗼𝗻𝘁𝗲𝘅𝘁-𝗔𝘂𝗴𝗺𝗲𝗻𝘁𝗲𝗱 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗼𝗻) (𝘴𝘵𝘦𝘱-𝘣𝘺-𝘴𝘵𝘦𝘱) CAG lets AI remember past interactions to provide more relevant, personalized, and context-aware responses. 1. 𝗦𝘁𝗮𝗿𝘁 𝘄𝗶𝘁𝗵 𝗾𝘂𝗲𝗿𝘆 – User provides input or a task request. 2. 𝗣𝗿𝗼𝗰𝗲𝘀𝘀 𝗶𝗻𝗽𝘂𝘁 – Convert it into a structured format for the model. 3. 𝗜𝗻𝗷𝗲𝗰𝘁 𝗰𝗼𝗻𝘁𝗲𝘅𝘁 – Add relevant background (past chats, user data, goals). 4. 𝗥𝗲𝗰𝗮𝗹𝗹 𝗱𝗼𝗺𝗮𝗶𝗻 𝗺𝗲𝗺𝗼𝗿𝘆 – Bring in domain-specific knowledge or prior interactions. 5. 𝗔𝗰𝗰𝗲𝘀𝘀 𝗸𝗻𝗼𝘄𝗹𝗲𝗱𝗴𝗲 𝗯𝗮𝘀𝗲 – Fetch related internal or external references. 6. 𝗠𝗲𝗿𝗴𝗲 𝗱𝗮𝘁𝗮 – Combine all context and knowledge sources. 7. 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗲 𝗼𝘂𝘁𝗽𝘂𝘁 – Create a response using this rich, aligned context. 8. 𝗩𝗲𝗿𝗶𝗳𝘆 𝗿𝗲𝘀𝗽𝗼𝗻𝘀𝗲 – Check the result for logical and contextual accuracy. 9. 𝗘𝘅𝗽𝗮𝗻𝗱 𝗰𝗼𝗻𝘁𝗲𝘅𝘁 – Enrich the response with more relevant details if needed. 10. 𝗔𝗹𝗶𝗴𝗻 𝗰𝗼𝗻𝘁𝗲𝘅𝘁 – Ensure the output fits the user’s prior goals or conversation. 11. 𝗖𝗵𝗲𝗰𝗸 𝗰𝗼𝗻𝘀𝗶𝘀𝘁𝗲𝗻𝗰𝘆 – Confirm that everything stays coherent and connected. 12. 𝗗𝗲𝗹𝗶𝘃𝗲𝗿 𝗳𝗶𝗻𝗮𝗹 𝗼𝘂𝘁𝗽𝘂𝘁 – Provide a complete, context-aware, and consistent answer. In short: • 𝗥𝗔𝗚 gives models access to the 𝗿𝗶𝗴𝗵𝘁 𝗶𝗻𝗳𝗼𝗿𝗺𝗮𝘁𝗶𝗼𝗻. • 𝗖𝗔𝗚 helps them use it 𝗶𝗻 𝘁𝗵𝗲 𝗿𝗶𝗴𝗵𝘁 𝗰𝗼𝗻𝘁𝗲𝘅𝘁. Together, they make AI systems: more accurate, more reliable, more personalized and more useful in real-world workflows. ✅ Repost for others in your network who can benefit from this.

60% of support tickets are repetitive. And, customers expect immediate responses. That creates pressure on teams and frustration for customers. This is why support is one of the most practical and now proven places to apply AI. AI can handle common, repeat questions instantly, in your tone, using your knowledge base and CRM data. That frees up humans to focus on situations that require judgment, empathy, and creativity. One of our customers, The Knowledge Society (TKS) Society, did exactly that. Every enrollment season, they saw a surge of messages across email, Facebook Messenger, and WhatsApp. The busiest time of year was also the most overwhelming for their team. They implemented the Customer agent to answer common enrollment questions around the clock. Today, close to 80% of inquiries are handled automatically. Their team now spends more time on complex conversations and less time copying and pasting the same answers. The (ISSA) International Sports Sciences Association also scaled with Customer Agent. They were managing multiple support channels across different tools. The experience was fragmented for their team and inconsistent for customers. By introducing an AI agent to handle repetitive questions across channels, they cut response times in half and created a more consistent experience. Over 8,000 companies are already using HubSpot’s Customer Agent, with resolution rates above 67%. This is the real opportunity with AI in support.

You don't need a 2 trillion parameter model to tell you the capital of France is Paris. Be smart and route between a panel of models according to query difficulty and model specialty! New paper proposes a framework to train a router that routes queries to the appropriate LLM to optimize the trade-off b/w cost vs. performance. Overview: Model inference cost varies significantly: Per one million output tokens: Llama-3-70b ($1) vs. GPT-4-0613 ($60), Haiku ($1.25) vs. Opus ($75) The RouteLLM paper propose a router training framework based on human preference data and augmentation techniques, demonstrating over 2x cost saving on widely used benchmarks. They define the problem as having to choose between two classes of models: (1) strong models - produce high quality responses but at a high cost (GPT-4o, Claude3.5) (2) weak models - relatively lower quality and lower cost (Mixtral8x7B, Llama3-8b) A good router requires a deep understanding of the question’s complexity as well as the strengths and weaknesses of the available LLMs. Explore different routing approaches: - Similarity-weighted (SW) ranking - Matrix factorization - BERT query classifier - Causal LLM query classifier Neat Ideas to Build From: - Users can collect a small amount of in-domain data to improve performance for their specific use cases via dataset augmentation. - Can expand this problem from routing between a strong and weak LLM to a multiclass model routing approach where we have specialist models(language vision model, function calling model etc.) - Larger framework controlled by a router - imagine a system of 15-20 tuned small models and the router as the n+1'th model responsible for picking the LLM that will handle a particular query at inference time. - MoA architectures: Routing to different architectures of a Mixture of Agents would be a cool idea as well. Depending on the query you decide how many proposers there should be, how many layers in the mixture, what the aggregate models should be etc. - Route based caching: If you get redundant queries that are slightly different then route the query+previous answer to a small model to light rewriting instead of regenerating the answer

Two weeks ago I said AI Agents are handling 95% of our sales and support and I replaced $300k of salaries with a $99/mo Delphi clone. 25+ founders DM’d me… “HOW?” Here’s the 6 things you MUST do if you want to run your entire customer-facing business with AI: 1. Create a truly excellent knowledge base. Your AI is only as good as the content you feed it. If you’re starting from zero, aim for one post per day. Answer a support question by writing a post, reply with the post. After 6mo you have 180 posts. 2. Have Robb’s CustomGPT edit the posts to be consumed by AI. Robb created a GPT (link below) that tweaks posts according to Intercom’s guidance for creating content for Fin. The content is still legible to humans, but optimized for AI. 3. Eliminate recursive loops - because pissed off customers won’t buy If your AI can’t answer a question but sends the customer to an email address which is answered by the same AI, you are in trouble. Fin’s guidance feature can set up rules to escalate appropriately, eliminate loops, and keep customers happy. 4. Look at every single question every single day (yes, EVERY DAY). Every morning Robb looks at every Fin response and I look at every Delphi response. If they aren’t as good as they could possibly be, we either revise the response, or Robb creates a support doc to properly handle the question. 5. Make sure you have FAQs, Troubleshooting, and Changelogs. FAQs are an AI’s dream. Bonus points if you create FAQ’s written exactly how your customers ask the question. We have a main FAQ, and FAQs for each sub section of our support docs. Detailed troubleshooting gives the AI the ability to handle technical questions. Fin can solve 95% of script install issues because of our Troubleshooting section. Changelogs allow the AI to stay on top of what’s changed in the app to give context to questins about features and UI as it changes. 6. Measure your AI’s performance and keep it improving. When we started using Fin over 1y ago, we were at 25% positive resolutions. Now we’re above 70%. You can actively monitor positive resolutions, sentiment, and CSAT to make sure your AI keeps improving and delivering your customers an increasingly positive experience. TAKEAWAY: Every Founder wants to replace entire teams with AI. But nobody wants to do the actual work to make it happen. Everybody expects to flip a switch and have perfect customer service. The reality? You need to treat your AI like your best employee. Train it daily. Give it the resources it needs. Hold it accountable for results. Here’s the truth that the LinkedIn clickbait won't tell you… The KEY to successfully running entire business units with AI? Your AI is only as good as the content you feed it. P.S. Want Robb's CustomGPT? We just launched 6-part video series on how RB2B trained its agents well enough to disappear for a week and let AI run the entire business. Access it + get all our AI tools: https://www.rb2b.com/ai

Power Automate Work Queues are not built for scale! That's a fact. When you think about scalability in Power Automate, one thing that will definitely come to mind at some point is queues and workload management. While you might be able to survive without them in some event-based transactional flows that only process a single item at a time, but whenever you process tasks in batches, or when RPA gets involved, you'll need queues. Power Automate comes with Work Queues out of the box. And you would think that's your go-to queueing mechanism for scaling. After all, it's at scale that you really need those queues - to de-couple your flows and make it easier to maintain, support, debug them, as well as make them more robust and efficient. Queues is a must even at medium scale. Heck, we use them even in small scale implementations. But the surprising thing about Power Automate Work Queues is that they are not fit for high scale implementations. And that is by design! The docs themselves (link in the comments) explicitly state that if have high volumes or if you dequeue (pick up work items from the queue for processing) concurrently, you should either do it within moderate levels or use something else. If you try and use Power Automate Work Queues for high scale implementations (more than 5 concurrent dequeue operations or hundreds/thousands of any type operations involving the queues), you'll get in trouble. There can be all sorts of issues that could happen - your data may get duplicated, you may accidentally deque the same work item in multiple concurrent instances, or your flows might simply get throttled or even crash. This is because of the way they're build and the way they utilize Dataverse tables for storing work items and work queue metadata. So, if you do want to scale, it's best to use an alternative. And, obviously, Microsoft wouldn't be Microsoft if they didn't have an alternative tool to do that. The docs themselves recommend Azure Service Bus Queues for high throughput queueing mechanisms. Another alternative could also be Azure Storage Queues, but that only makes sense if the individual work items in your queue can get large (lots of data or even documents) or when you expect your queue to grow beyond 80GB (which is possible in very large scale implementations). Otherwise, Azure Service Bus Queues are absolutely perfect for very large volumes of small transactions. On top of that, they have some very advanced features for managing, tracking, auditing and otherwise handling your work items. And, of course, there's a existing connector in Power Automate to use it. So, while I do love Power Automate Work Queues, I'll only use them in relatively small scale implementations. And for everything else - my queues will go to Azure. And so should yours.

This is why AI agents are exploding in adoption—they deliver real business value by turning LLM intelligence into automated action. They are becoming the backbone of automation in customer support, operations, sales, and internal workflows, replacing repetitive tasks that humans perform by clicking buttons and following rules. Instead of just generating text, AI agents orchestrate actions, making them far more valuable in real business environments. A perfect example is customer-support order-tracking. Every day, support teams receive hundreds of emails asking, “Where is my order?” A human agent reads the message, extracts the order number, searches in the backend system, checks the shipment status in the carrier portal, decides what’s wrong, and finally replies or creates a follow-up ticket. This manual process takes 2–3 minutes per email—highly repetitive and expensive at scale. An AI agent can now automate this entire workflow end-to-end. It first extracts the order ID from the customer’s message, then calls the lookup_order tool to fetch order details, and the check_tracking_status tool to get carrier updates. Next, it analyzes the status and determines whether delivery is delayed, lost, or on track. Based on the result, it triggers the right action, such as create_internal_ticket, initiate_carrier_trace, or reschedule_delivery. Finally, the agent generates a personalized reply to the customer with the latest status—without any human involvement. With memory, it can even handle future follow-ups intelligently. Read more on the internal architecture of an AI Agent in detail: https://lnkd.in/gEhVX5cY Build Your First AI Agent in 10 Minutes! (No Code Needed): https://lnkd.in/gjNf5yyr

I just built an FAQ using the new Copilot powered FAQ Webpart, And a Copilot agent from the same 140-page compliance manual. Here's when to use each one because you're probably wondering which tool to pick for your next project. When you see shiny new Copilot integrations in SharePoint you tend to think you need to choose one. Nope. In this case, they solve different problems for your users. The FAQ web part is for your quick-answer people. We all know the ones, they scan, find their question, get the answer, and move on. When I tested "How is staff measured on compliance performance?" the FAQ gave me a clean, condensed response. Perfect for someone who just needs the policy details without the conversation. The Copilot agent is for your detail-seekers, your conversationalist. Same question, but the agent gave me way more context and background. It's conversational. Your users can ask follow-ups, dig deeper, get explanations. Better when someone's trying to understand how policies actually work in their day-to-day. Here's what I learned building both, so you don't have to. The FAQ took one document and created clean categories with collapsible questions. Great for your policies, procedures, anything where people need quick reference. Think employee handbook, IT support, compliance guidelines. The agent lets your people have actual conversations about that same content. Someone can ask "What happens if we miss a compliance deadline?" and get a detailed response they can build on. You might want both. People work differently. Some scan FAQs, others prefer to ask questions and get explanations. Don't make this an either-or decision for your organization. Build what matches how your users actually work.

Big sale seasons do not test your discounts. They test your ability to handle conversations at scale. When Being Human Clothing prepared for one of their largest sale windows of the year, the expectation was clear: massive traffic spikes, a surge of first-time shoppers, mounting pressure on support teams, and the risk of long wait times Most brands focus on demand generation during peak moments. Very few invest equally in demand management. Instead of waiting for support queues to overflow, Being Human chose to scale intelligence. They deployed a custom AI agent, BH Buddy, across their channels in just a few days. The assistant handled product discovery, order queries, billing, refunds, returns, and general support, all managed through a unified dashboard. During a crucial sales window, the system handled real shopper conversations in real time The outcome was not just operational stability, it was measurable impact: • 115,000 messages handled in 20 days • 2.5x surge in queries on the peak sale day • 88.6% positive customer sentiment What stands out here is not just volume. It is resilience. AI did not replace the human team. It reduced pressure, captured context, resolved routine queries, and escalated complex conversations intelligently. Peak performance today requires more than marketing firepower. It requires conversational infrastructure that can scale as confidently as your traffic does. That is the real differentiator. Farooq | Sreeraman | Ragini | Ronak | Salman | Kushan | Jigar | Sumit | Abhay| Abhishek | Faizan | Jimesh

A junior reached out to me last week. One of our APIs was collapsing under 150 requests per second. Yes — only 150. He had tried everything: * Added an in-memory cache * Scaled the K8s pods * Increased CPU and memory Nothing worked. The API still couldn’t scale beyond 150 RPS. Latency? Upwards of 1 minute. 🤯 Brain = Blown. So I rolled up my sleeves and started digging; studied the code, the query patterns, and the call graphs. Turns out, the problem wasn’t hardware. It was design. It was a bulk API processing 70 requests per call. For every request: 1. Making multiple synchronous downstream calls 2. Hitting the DB repeatedly for the same data for every request 3. Using local caches (different for each of 15 pods!) So instead of adding more pods, we redesigned the flow: 1. Reduced 350 DB calls → 5 DB calls 2. Built a common context object shared across all requests 3. Shifted reads to dedicated read replicas 4. Moved from in-memory to Redis cache (shared across pods) Results: 1. 20× higher throughput — 3K QPS 2. 60× lower latency (~60s → 0.8s) 3. 50% lower infra cost (fewer pods, better design) The insight? 1. Most scalability issues aren’t infrastructure limits; they’re architectural inefficiencies disguised as capacity problems. 2. Scaling isn’t about throwing hardware at the problem. It’s about tightening data paths, minimizing redundancy, and respecting latency budgets. Before you spin up the next node, ask yourself: Is my architecture optimized enough to earn that node?