James Deaker

AI is re-shaping marketplaces and not just adding layers, but fundamentally enabling new value. As someone working at the intersection of product-led growth, analytics, and platform thinking, this piece reinforces that marketplace success isn’t just about network effects any more it’s increasingly about how AI can amplify both sides of the marketplace flywheel: supply onboarding + demand fulfilment. This aligns with the shift from “software eats the world” → “AI eats software and re-defines platform economics”. The biggest opportunity isn’t just adding AI features. It’s rethinking the marketplace flywheel from first principles in light of what AI now makes possible. Key take-aways from this great piece from Andreessen Horowitz who are at the centre of the investment cycle into the emerging businesses in this sector: AI doesn’t just optimise existing marketplaces, it can unlock more listings, drive higher conversion rates, and create more repeat purchase behaviour. For physical, commodity-product marketplaces (think grocery delivery, ride sharing), AI helps operations, but the core model remains relatively unchanged. Where AI really shifts the game: marketplaces for personalised products or services (e.g., artisan goods, niche services). Where sellers face high friction in onboarding, listing, pricing, inventory management. AI reduces that friction, enabling more supply and better matching. The impact of this dynamic is a value multiplier: Lower cost per transaction (because AI automates tasks humans used to do) Higher throughput (more supply + demand matched faster) Greater repeat usage (AI supports lifecycle and re-engagement) “AI should drive more liquid and more stable marketplaces ... with more listings, higher conversions … and more repeat purchases than ever before.” The biggest opportunity isn’t just adding AI features: it’s rethinking the marketplace flywheel from first principles in light of what AI now makes possible. https://lnkd.in/g6EEmh3v #marketplaces #AI #platformstrategy #PLG #DigitalTransformation #Ecosystems #BusinessModel

2

Four notable pieces went out on AIVO Journal this month. Different angles, same throughline: the gap between being seen by an AI system and being chosen by one. "Measuring the Wrong Era" took apart the IAB's new AI visibility framework. It gives the market a shared vocabulary for presence, prominence, and share of voice. It has almost nothing to say about what happens after a brand is mentioned. A framework built for single-turn search doesn't see the turn where a brand actually gets dropped. "The Bubble Standard" looked at what the current AI visibility market is actually being valued on. A lot of it is presence data dressed up as insight. Presence was never the hard problem. "The Differentiation That Doesn't Travel" asked why brand equity built over decades of human perception often doesn't survive contact with an AI system's reasoning. Strong human reputation and strong AI-mediated outcome are not the same asset, and treating them as one is costing brands the moment it matters most. "Measuring the AIVO Paradox Was Never the Hard Part" closed out the four pieces. It's the operational follow-up to Agentic Brand Control: four steps for actually doing something once you know a brand is being displaced. Diagnosis, remediation, resilience testing, monitoring. Read all four on AIVO Journal. Link in comments. #AgenticBrandControl #AIVOStandard #AIVOJournal

Another sign of the power of bringing rich AI features all the way into your funnel. Customers who use Albertsons' conversational search spend about 10% more per order. Customers who use its fuller assistants, the ones that find recipes and build a list against dietary preferences, spend 26% more. Jill Pavlovich, Albertsons' senior vice president of digital shopping experiences, gave the Wall Street Journal the reason: "When they're using a more comprehensive experience, they're adding even more, because they're not forgetting items." Read those two numbers next to each other and the variable is not whether there is a conversation. It is how much of the job the conversation is allowed to finish. A chat box bolted onto a search index gets you the 10%. The version that takes a constraint, holds it, and assembles the whole order gets you the 26%. Same underlying technology, sixteen points apart, and the difference is depth of integration rather than model quality. Almost every conversational travel implementation I have looked at is the 10% version. Natural language at the entrance, then the traveler is handed back to a date picker and a destination field, and most of what the conversation collected is discarded on the way down. Groceries are not trips. A weekly shop is a repeat purchase against a list somebody already knows, and no hotel booking will ever have that shape, so I would not expect a clean 26% to show up in travel. But the shape of the finding is directionally useful and matches much of what we are seeing with our travel clients testing our Agentic Commerce Platform. Going from an AI search box to a full AI experience is powerful stuff. And I suspect it will be even more so as agents start to navigate those funnels on their own. https://lnkd.in/gznsdtEw

14

3 Comments

Over the past several months, our team at Lyft has been deeply focused on a question advertisers ask all the time: 𝐖𝐡𝐢𝐜𝐡 𝐬𝐩𝐞𝐜𝐢𝐟𝐢𝐜 𝐯𝐢𝐬𝐮𝐚𝐥, 𝐭𝐞𝐱𝐭𝐮𝐚𝐥, 𝐚𝐧𝐝 𝐜𝐨𝐧𝐭𝐞𝐱𝐭𝐮𝐚𝐥 𝐞𝐥𝐞𝐦𝐞𝐧𝐭𝐬 𝐢𝐧𝐬𝐢𝐝𝐞 𝐚𝐧 𝐚𝐝 𝐚𝐜𝐭𝐮𝐚𝐥𝐥𝐲 𝐝𝐫𝐢𝐯𝐞 𝐞𝐧𝐠𝐚𝐠𝐞𝐦𝐞𝐧𝐭 𝐢𝐧 𝐚 𝐫𝐢𝐝𝐞𝐬𝐡𝐚𝐫𝐞 𝐞𝐧𝐯𝐢𝐫𝐨𝐧𝐦𝐞𝐧𝐭? Unlike crowded social feeds, riders spend ~24 minutes in focused attention during a Lyft trip — creating one of the most unique media surfaces in advertising. But this environment also demands creatives that are built for the moment, not borrowed from other platforms. To answer this, our cross-functional team analyzed nearly 1,000 display creatives using advanced statistical modeling, multimodal feature extraction (computer vision + OCR + semantic tagging), and context-aware controls. 𝐓𝐡𝐞 𝐫𝐞𝐬𝐮𝐥𝐭: 𝐚 𝐬𝐜𝐢𝐞𝐧𝐜𝐞-𝐛𝐚𝐜𝐤𝐞𝐝 𝐬𝐞𝐭 𝐨𝐟 𝐜𝐫𝐞𝐚𝐭𝐢𝐯𝐞 𝐛𝐞𝐬𝐭 𝐩𝐫𝐚𝐜𝐭𝐢𝐜𝐞𝐬 𝐭𝐡𝐚𝐭 𝐦𝐚𝐭𝐞𝐫𝐢𝐚𝐥𝐥𝐲 𝐢𝐦𝐩𝐫𝐨𝐯𝐞 𝐞𝐧𝐠𝐚𝐠𝐞𝐦𝐞𝐧𝐭 𝐚𝐧𝐝 𝐚𝐧𝐝 𝐞𝐥𝐢𝐦𝐢𝐧𝐚𝐭𝐞 𝐠𝐮𝐞𝐬𝐬𝐰𝐨𝐫𝐤. https://lnkd.in/gfjJZ6ZV A few highlights: ✅ Clarity drives action — explicit offers, clear CTAs, and prominent buttons remove decision friction. ✅ Visual hierarchy matters — logos and products should support, not compete with the core message. ✅ Short, high-impact, or benefit-rich copy performs best — concise text wins in rideshare environments. ✅ Timing is critical — content aligned to rider mindset (e.g., practical mornings vs. aspirational evenings) performs significantly better. ✅ Strategic face placement builds trust and boosts action when paired near the CTA. ✅ Cool color palettes & high contrast outperform in variable in-car lighting conditions. 📈 These insights aren’t heuristics — they’re grounded in statistical modeling that isolates the impact of each creative element while controlling for other variables, enabling clear attribution of what truly moves the needle. For advertisers, the takeaway is simple: Rideshare creatives should be purpose-built for the context riders experience — and the data shows that when you do, performance lifts are measurable and meaningful. Huge shout-out to our collaborators across Data Science, Creative Teams, UX, Product Marketing, and Sales for pushing this work forward. Francisco Romaldo M. Will Eagle Shash Mehrotra Jenny Liu-Li Kat Murray If you’d like to learn more or apply these findings to improve your campaigns, reach out — we’re excited to share. Onward. 🚘📊✨

34

If we want trustworthy AI experiences, someone has to measure what is actually meaningful. Fortune really does favor the prepared. As I keep building out my Measuring What’s Meaningful work, I’m tracking examples of how teams define meaningful success in ways that shape real decisions. Design and research teams are built for this because we sit closest to the people who feel the impact. Core to my recent talk at the Fed’s Design Summit are three practices and one mindset that help teams anchor meaning inside metrics. This post covers the first practice, Define Meaningful Success, which centers on developing clear design principles and metrics that support teams in tracking experience quality and human impact. This stood out clearly in the opening panel of the Dscout Co-Lab Continued replay, moderated by Julie Marie Norvaisas. Eleanor Sandford from Disney talked about how their Experience Quality Framework guides teams: start with fundamentals like predictable, intuitive, efficient, streamlined before reaching for “magic.” Those basic principles protect trust, especially as AI takes over more of the experience. A few highlights I pulled from her talk: • Trust breakers reveal when teams are slipping on their own principles, often before any dashboard spikes. • One emotional clip can surface a principle gap faster than any automated summary. • If the foundation slips, any AI layered on top erodes trust fast. • Strong cross functional relationships help teams uphold these principles in practice, which keeps experiments honest and prevents shortcuts. All of this reinforces a simple point: Design and research teams must take their seat at the success-definition table. Principles are only the first step. Translating them into metrics is how we protect the human impact inside experience quality. Grateful to Dscout for sharing the session with the broader community. Define success. Build real principles. Translate them into metrics that keep human outcomes visible, accountable, and impossible for anyone to sideline. #DesignLeadership #HumanCenteredAI #ExperienceQuality #MetricsThatMatter #UXResearch #TrustbyDesign

18

I read a lot of news about AI Agents failing in production. You’ve probably heard the stat that less than 5% of them ever make it live. People imagine a digital employee that figures things out, navigates ambiguity, and executes 20 steps perfectly. If you strip this down to first principles (specifically probability theory), you see exactly why these agents fail. It’s called Error Propagation. If an AI model is 95% accurate at a single task, that sounds great. But an AI Agent by definition usually requires a chain of sequential steps. Let's say 10 steps. The math is simple: 0.95 ^ 10 = 0.59 Your agent now has a 59% success rate. It is effectively a coin flip. This is why we tell our clients: Let's not build General Agents. Let's focus on Narrow Chains. When we reduce the scope, we reduce the steps. When we reduce the steps, we fight the entropy. We never sell "magic digital employees". We sell high-probability agentic systems, constrained by their environment to solve very specific problems. And that actually works. The market is obsessed with Intelligence. We are obsessed with Reliability. Physics favors the latter. Grateful for the team helping make this possible :)

25

1 Comment

We built an AI system that outperformed Claude. On evaluating creative portfolios, which is what Mela's vetting layer runs on. 48% exact match rate against human evaluation ratings as ground truth. MAE of 0.52 versus Claude's 0.81 on the same portfolios. Every score within one point of the real answer. This matters because evaluating creative work is one of the hardest things to automate. Design and creativity don't have right answers. They have judgment calls. Getting AI to that standard is a different problem than evaluating code or summarizing text. We're rolling out Mela's beta in two weeks. The vetting layer is what the whole trust model sits on, and this is what makes it work. The reason it beats Claude isn't that it's smarter. It does one thing. Built for one task, calibrated against real human judgment, with no generalist noise getting in the way. Last week I wrote that generalist loses to specialist. We just proved it on ourselves. If you're a freelancer, brand, or agency interested in being part of the beta, DM me.

9

This morning at the #AdobeAISummit, one number set the tone, 50% of organic site traffic will disappear by 2028 (Gartner) Kévin Bourlier, from Adobe, chosed to open his conference with it. We’ve been using this exact figure in our GEO presentations at Capgemini for about a year now. A year ago, the reaction was: “That’s impossible.” Six months ago: “Maybe not impossible… but not that fast.” Today? The consensus in the room was strikingly different. Not only does the 50% figure feel robust, it already sounds conservative. What will we be saying in six months? Because this isn’t just about traffic decline. It’s about structural migration. As Kevin put it: “We are moving from search engine optimization to the famous GEO […] your content can be directly consumed and interpreted by AIs.” The traditional web visit is no longer the default decision space. Conversational interfaces and LLMs increasingly are. At the same time, we are moving: • From software as a tool to software as a collaborator • From content production to content supply chains • From experimentation with AI to organizational dependence on AI And as multiple speakers reminded us, 85% of AI failures still stem from data silos and poor data quality. Agentic ambition without data discipline will collapse. The strategic takeaway is simple: GEO, LLM-optimized content, and AI-native content supply chains are not tactical upgrades. They are structural shifts. If 50% sounded impossible a year ago, and conservative today the real question is not whether the curve will bend. It’s how prepared we are when it does. #AdobeAISummit #GEO #AgenticAI #AIStrategy #ContentSupplyChain Pierre Bastien Heemesh PUTTY Mathilde Arai Anne-Sophie Vilcot Rob Pillar Jens Jacobsen Robin van den Hoven

145

3 Comments

https://lnkd.in/g9jAu7p4 Anthropic coding market share increased to 54%, from 42% just six months ago! Build/buy preferences shifted from ~50/50 -> ~25/75 Open-source market share decreased 19% -> 11%. Within remaining usage, Llama models surprisingly still ~70% of share

28