DEV Community: Rémi Henriot

ChatGPT Won't Capture Premium Brands — and They Know It

Rémi Henriot — Sat, 19 Sep 2026 12:15:08 +0000

ChatGPT Won't Capture Premium Brands, and They Know It

Hermès doesn't sell in supermarkets. Why would it answer through ChatGPT? Two markets: commodities in the generic assistant, story-driven brands in their own space.

Hermès doesn't sell in supermarkets. Why would it answer through ChatGPT? Two markets: commodities in the generic assistant, story-driven brands in their own space.

The tech consensus is settled: ChatGPT, Gemini, Perplexity are becoming the universal point of contact between consumers and brands. Search gives way to the assistant. The site gives way to chat. The brand becomes an answer in a stream of questions.

That consensus is half right. And the wrong half will cost the brands that believe it dearly.

For categories where brand barely matters, the thesis holds. "Best vacuum under $300?" "Cheap car insurance?" "A hotel in Chicago this weekend?" These queries migrate naturally to generic assistants. The consumer wants a fast, comparative, frictionless answer. A brand-specific interface adds nothing.

That market is real. It will grow. Commodity brands are right to adapt to it.

The consensus forgets that markets aren't homogeneous. There are categories where the brand is a commodity, and categories where the brand is the product.

A universal assistant applies, by definition, the same treatment to every brand. It can be competent, precise, useful. It cannot be the voice of a house. Its business model runs on indifferentiation, exactly what story-driven brands have spent decades fighting.

The conversational market will segment the way physical retail did. Generalist giants on one side: marketplaces, comparison engines, infinite shelves. Houses with a story on the other, the ones that always refused to reduce their experience to supermarket shelf space. They won't accept it in conversational space either.

And the clock is running. Every brand will soon talk directly with its customers, the question is where, and on whose terms. It's the mobile-app shift all over again: those who "thought mobile" early built their channel. The rest existed in other people's stores.

Brands with high image value, luxury, premium craft, B2B expertise, community brands, will choose to own their conversational space. For them, delegating the interface to a generalist wouldn't be a simplification. It would be a surrender.

Scenaro is built for these brands. Not because universal LLMs will fail, they'll thrive, in their segment. But because the most profitable part of commerce, in margin and loyalty, will never reduce to a query in a generic assistant.

Owning your space, concretely: a recognizable voice, memory that carries across sessions, versioned scenarios that hold the house's expertise, and multimodal staging, voice and interface in sync, the way Urbansider, a Paris travel concierge, runs it in production. Club Français du Vin gained a marked improvement in engagement. Lupi, in pet e-commerce, is building its customer relationship there.

Should your brand be visible in the LLMs? Probably, data, citations, discovery. Should your brand experience live there? If your value rests on that experience, certainly not.

Commerce will segment between the universal and the intimate. Premium brands have always known which side they're on.

The full article lives on Scenaro: ChatGPT Won't Capture Premium Brands, and They Know It

If you prefer the series by email: https://scenaro.substack.com/subscribe

The Infrastructure You Rent vs. the One You Build

Rémi Henriot — Thu, 17 Sep 2026 12:00:38 +0000

The Infrastructure You Rent vs. the One You Build

"Everything is managed" means your data, your costs, and your differentiation live with a third party. The real measure of infrastructure is the cost of leaving it.

"Everything is managed" means your data, your costs, and your differentiation live with a third party. The real measure of infrastructure is the cost of leaving it.

"Plug into our infrastructure. Everything is managed." It's exactly what a busy team wants to hear.

Because "everything is managed" means your conversation histories, your customer profiles, your negotiating power, and your technical room to maneuver all live with someone else. The brochure never asks the only question that matters: what happens the day you want to leave?

Renting conversational infrastructure means entrusting three assets you only measure the day you need them.

Your data. Conversations, profiles, recurring questions, friction points, the most valuable raw material your brand accumulates. It lives on the provider's servers. On departure day: in what format do you get it back? At what granularity? At what price?

Your negotiating power. Day-1 pricing is guaranteed nowhere. When your entire customer experience runs on a third party's infrastructure, your leverage at renewal time approaches zero. The provider knows this. Their model depends on it.

Your differentiation. A better model ships, can you switch? A new channel opens, can you go there? On someone else's infrastructure, those are their decisions. Not yours.

Managed SaaS optimizes for retention, not portability. That's not a scandal. It's the business model. You just need to know it before you sign.

The alternative to "renting" isn't "building everything yourself." It's composing your conversational space on an open experience layer.

Your API keys are yours. Scenaro takes no margin on models. You plug in your own OpenAI, Anthropic, or Mistral keys, 40+ providers supported. A better model ships, you switch. Multi-model, no lock-in.

Your data belongs to you. Conversations, profiles, metrics, everything accessible via API, exportable at any time, deletable if needed.

Your scenarios are versioned and portable. What you build, personalities, scenarios, knowledge bases, are artifacts you own, with full traceability of what was deployed and when. And the layer is open: SDK, LiveKit, you compose, you're not boxed in. At Club Français du Vin, this architecture enabled fast iteration on the voice experience, a marked improvement in engagement, without rebuilding anything at each change.

Good infrastructure comes down to one question: if I want to migrate tomorrow, how long does it take, and what do I lose that's irreplaceable?

If the honest answer is "everything," you don't have a provider. You have a landlord.

Renting is a starting point. Building is a strategy.

The full article lives on Scenaro: The Infrastructure You Rent vs. the One You Build

If you prefer the series by email: https://scenaro.substack.com/subscribe

3 Conversions in 5 Days — Then What?

Rémi Henriot — Sat, 12 Sep 2026 12:12:45 +0000

3× Conversions in 5 Days, Then What?

Forced selling converts today and breeds resentment tomorrow. What you win in five days says nothing about what's left in month six.

Forced selling converts today and breeds resentment tomorrow. What you win in five days says nothing about what's left in month six.

"3× your conversions in 5 days." "+40% revenue in the first month." Conversational AI landing pages glow green.

Nobody asks the uncomfortable question: then what?

Optimizing a conversion and building a relationship are not the same objective. They frequently collide head-on. And miracle-ROI pitches are betting you'll only look at the first one.

The ROI-first logic is simple: the assistant is a sales lever. A/B-tested phrasing, automated follow-ups, every message optimized to maximize the click. The assistant becomes a funnel element. One that pushes.

Forced selling, the assistant that pushes too hard, follows up too fast, reduces every exchange to a call to action, converts today. And breeds resentment tomorrow.

Conversion metrics capture an outcome, not an experience. A rising conversion rate can mask a falling NPS. A growing basket can coexist with the word "pushy" in your reviews.

The right question isn't "did the assistant sell?" It's "will the customer come back?" Only the second one translates into lasting value, and no 5-day dashboard can answer it.

There's a paradox the best customer experiences keep confirming: conversations that don't try to sell outperform those optimized for selling.

At Club Français du Vin, the marked improvement in engagement Scenaro delivered didn't come from an assistant tuned to convert. It came from an experience that guides, advises, takes the time to explain why this wine fits this moment. Conversion is a consequence of a good conversation, not its objective.

The alternative to miracle metrics exists. It just requires more honesty.

Scenaro's conversation analytics look at what funnel metrics ignore: what customers actually ask, where they drop off, what keeps coming back conversation after conversation. A customer who asks three questions before buying understood what they bought. A customer who buys in 30 seconds after a nudge may never return.

And cross-session memory changes what measurement even means: profiles built over time track repeat behavior, loyalty, the arc of the relationship. The 10th exchange is worth more than the 1st, that's where the value lives, and it's precisely what 5-day promises will never measure.

Versioned scenarios let you iterate on tone and advisory depth, and verify the one thing the green dashboard won't show: do the customers who had the best conversations come back more often?

Conversion is a signal. Relationship is a value. One is won in five days. The other is built over time, and it's the one that pays.

The full article lives on Scenaro: 3× Conversions in 5 Days, Then What?

If you prefer the series by email: https://scenaro.substack.com/subscribe

"Connect Your Catalog" Is Not "Tell Your Story"

Rémi Henriot — Thu, 10 Sep 2026 12:01:38 +0000

"Connect Your Catalog" Is Not "Tell Your Story"

A catalog says what a product is. A scenario says why it matters. A brand isn't a database, it's your story.

A catalog says what a product is. A scenario says why it matters. A brand isn't a database, it's your story.

"Connect your catalog." It's step zero of nearly every conversational AI offering in commerce. And it's sold as if it were the finish line.

Let's be fair: plugging in your product feed, Shopify, PIM, CSV, has real value. The assistant gets price, stock, and shipping right, around the clock. It takes pressure off customer service. Fine.

But a search engine, even a conversational one, is still a search engine. And "connect your catalog" is not "tell your story."

The wine merchant who has guided customers for thirty years isn't reading a product sheet. When he says "this Burgundy will be perfect for your Thursday dinner," he's drawing on everything the catalog doesn't contain: taste, context, knowledge of the customer. He knows you prefer low-tannin wines, that there's a vegetarian at the table, that you want accessible but not obvious.

An assistant limited to catalog data responds: "here are 47 results for red wine." Technically correct. Practically useless.

Consumers have learned to distrust generic recommendations, those "you might also like" blocks that optimize for margin, not fit. A conversational experience that behaves like a slightly smarter search bar confirms that distrust. In categories where expertise is the product, wine, design, services, premium retail, the bar is higher.

Scenaro collections index meaning, not just product fields. Semantic search lets the assistant understand "something light for drinks with friends", a query no SKU filter will ever parse.

More importantly, scenarios carry the brand's character and goals. The persona colors every answer: the value system, the tone, the selection criteria, the pairings, the producers' stories. Two levels of knowledge coexist, structured data and editorial expertise, versioned and modifiable at any time.

At Club Français du Vin, the result speaks for itself: an assistant that says "this wine is perfect for your Thursday gathering" because it knows you often entertain on Thursdays and prefer light whites. Cross-session memory does the rest: over time, the assistant moves from "searching" to "advising." A marked improvement in engagement, not a better search engine, a better conversation.

A brand isn't a database. It's your story.

Your catalog is your inventory. Your story is your value. A platform that only asks for your product feed isn't selling storytelling. It's selling a comparison engine.

The full article lives on Scenaro: "Connect Your Catalog" Is Not "Tell Your Story"

If you prefer the series by email: https://scenaro.substack.com/subscribe

"Encode Your Brand" — Or How to Hand Over the Keys

Rémi Henriot — Sat, 05 Sep 2026 12:03:47 +0000

"Encode Your Brand", Or How to Hand Over the Keys

Encoding your brand into a universal LLM is handing over the keys. A brand isn't a database. It's your story.

Encoding your brand into a universal LLM is handing over the keys. A brand isn't a database. It's your story.

"Encode your brand identity into an AI agent." The vocabulary is clever. It sounds like control, consistency, presence.

Look at what it actually describes: reducing an identity built over years of editorial choices and tone decisions to vectorized text chunks, then hoping a model's statistical probability returns something that resembles the brand.

The technology is real and it works: embeddings, vector databases, RAG. Your guidelines, FAQs, and product sheets get ingested, and the agent "sounds like the brand." No smoke and mirrors there.

The problem isn't technical. It's structural.

The LLM hosting your data also hosts your competitors'. The same filter applies to everyone. Your voice is processed as text. Your character is diluted into the probability of the "correct tone." The same engine smooths every brand flat, that's what it's for.

What the agent produces is a correct response. Not a brand conversation. A correct response answers a question. A brand conversation creates a moment: it asks a follow-up, dares a recommendation, reveals something.

And your encoded data lives with a third party. Terms change, models update, platforms pivot, your "encoded identity" goes along for the ride. You handed over the keys.

Every brand will soon talk directly with its customers. The question is where, and on whose terms.

When mobile arrived, the brands that mattered built their own apps. They didn't "encode their identity" into someone else's. Encoding your brand into a universal LLM is making the opposite choice, and calling it strategy.

Scenaro starts from the opposite premise: a brand is not data to inject into an existing system. It's the starting point for a space to build.

In a third-party LLM, a model speaks, and your brand partially informs it. In your own conversational space, your scenario, your personality, your logic define every exchange. You're the narrator, not a source.

Concretely: Scenaro scenarios define the assistant's personality, values, and limits. They're versioned, cleanly published, modifiable at any time, and they're yours. Your API keys are yours: Scenaro takes no margin on models. You pick your provider from 40+ options and switch without losing your experience. Your data exports via API.

At Club Français du Vin, this approach drove a marked improvement in engagement. Not because the brand was "encoded" somewhere, because it operates its own space, with scenarios that carry its expertise and tone.

A brand isn't a database. It's your story. It doesn't encode. It gets told.

The question to ask any conversational AI platform: if you leave tomorrow, what do you take with you?

The full article lives on Scenaro: "Encode Your Brand", Or How to Hand Over the Keys

If you prefer the series by email: https://scenaro.substack.com/subscribe

Everywhere and Nowhere: Why Universal Presence Isn't a Brand Strategy

Rémi Henriot — Thu, 03 Sep 2026 12:01:07 +0000

Everywhere and Nowhere: Why Universal Presence Isn't a Brand Strategy

In a universal assistant, you're never the one speaking. Unforgettable somewhere beats invisible everywhere.

In a universal assistant, you're never the one speaking. Unforgettable somewhere beats invisible everywhere.

The pitch is everywhere: be "present" in ChatGPT, Gemini, Perplexity. Hundreds of millions of people ask them questions every day. You'd be foolish not to show up.

Except one question never gets asked: show up how?

In a universal assistant, you're not the one speaking. The assistant is. With its tone, its filter, its rules. Your brand is one source among thousands, processed by the same algorithm as every competitor you have.

The offer is well-rehearsed: connect your product data, FAQs, spec sheets, and your brand becomes a source the big assistants can cite. Fine, for awareness.

But watch what the assistant does with your data. It compares. It synthesizes. It standardizes. That's its job. Its business model runs on indifferentiation.

It can relay your price. It cannot relay your voice. It can mention your brand. It cannot make it exist.

You can be the brand that best answers a generic question, and remain perfectly interchangeable in the customer's mind. Anyone who lived through the social media land grab knows the pattern: presence everywhere, identity nowhere.

Every brand will soon talk directly with its customers. That's no longer the question. The question is where, and on whose terms.

Mobile apps forced the exact same choice. Some brands built their own app, their own experience, their own relationship. Others settled for existing in someone else's search results.

There's a difference in kind between being a source in a system you don't control and being the narrator of your own experience.

Scenaro starts from that premise: every brand deserves its own conversational space. A space where the brand speaks, with its voice, its values, its memory. An assistant that remembers a customer is back for the third time, that they already bought, that they hesitated once. That cross-session memory turns exchanges into a relationship.

At Club Français du Vin, our first customer, the Scenaro voice experience drove a marked improvement in engagement. Not because it was everywhere, because it was exactly where it needed to be, with a versioned scenario and a tone you couldn't mistake for anyone else's.

Unforgettable somewhere beats invisible everywhere.

The question isn't "how do we show up everywhere?" It's "where do we want to be unforgettable?"

The full article lives on Scenaro: Everywhere and Nowhere: Why Universal Presence Isn't a Brand Strategy

If you prefer the series by email: https://scenaro.substack.com/subscribe

Your Brand Is Not a Database

Rémi Henriot — Sat, 29 Aug 2026 12:12:50 +0000

Your Brand Is Not a Database

A new category of tools promises to put brands inside ChatGPT. Encoding your brand into a universal LLM means handing it the keys. Your brand is not a database. It's your story.

A new category of tools promises to put brands inside ChatGPT. Encoding your brand into a universal LLM means handing it the keys. Your brand is not a database. It's your story.

The pitch is everywhere: "encode your brand into LLMs." Structure your catalog, format your data, and when a consumer asks for "the best natural wine under $25," you'll get cited.

When ChatGPT answers, ChatGPT is speaking. Neutral tone, drive for comprehensiveness, zero point of view. Your century-old wine house and the discounter next door come out with the same voice, the same format, the same indifference.

That's the very principle of a generalist assistant: useful about everything, therefore nobody in particular. Your brand isn't the narrator there. It's an entry the narrator consults.

A catalog says what a product is. A story says why it matters. Thirty years of tone, universe, and knowing how to welcome a hesitant newcomer or a loyal regular, the universal LLM compresses all of it into metadata.

Think of Hermès in a supermarket aisle: the product is there, the price too. Everything that makes someone choose Hermès is gone. Encoding into a universal LLM is the supermarket of conversational commerce.

Encoding your brand into a universal LLM means handing it the keys. The tone? The model decides. The recommendation? Its logic, not yours. The customer's memory? Its property. The relationship? No longer yours.

Every brand will soon talk directly with its customers, that shift is underway, just as mobile was. The question isn't whether to go. It's whether you go in your own home or in someone else's.

Scenaro builds the space that belongs to the brand. The name isn't accidental: a stage, with its sets, its characters, its narrative, owned, controlled, and evolved by the brand.

In practice, every scenario carries a persona: a character and objectives of its own, configured by your teams. And that persona colors everything, the tone of each answer, the way it guides, all the way down to the conversation summaries. What the house says, what it never says, how it treats a first-timer versus a regular: written by you, not inferred by a generalist model.

The experience remembers, too. The returning customer is recognized; their preferences shape what comes next. That's the difference between a database and a relationship.

At Club Français du Vin, our first client, that's what made the difference: not a "smarter" assistant, an assistant that knew the subscribers, their cellar, their tastes, and spoke like the brand.

Being visible in ChatGPT can be worth the investment, like being indexed by Google. But that's a visibility decision, not a relationship decision. Brands that navigated digital, mobile, and social all ended up making the same call: the strongest existed on other people's platforms and built their own home. They never confused the two.

Your brand is not a database. It's your story. And your story deserves its own stage.

The full article lives on Scenaro: Your Brand Is Not a Database

If you prefer the series by email: https://scenaro.substack.com/subscribe

Scenaro, the CMS of Tomorrow: Reinventing the Web Through Conversation

Rémi Henriot — Thu, 27 Aug 2026 12:00:34 +0000

Scenaro, the CMS of Tomorrow: Reinventing the Web Through Conversation

Every brand will soon talk directly with its customers. The static web got WordPress. The conversational web will get Scenaro.

Every brand will soon talk directly with its customers. The static web got WordPress. The conversational web will get Scenaro.

In 2003, WordPress didn't invent HTML. It changed who gets to speak on the internet. Twenty years later: 500 million websites, a third of the web.

In 2026, the conversational web stands at the same turning point. And it's repeating the pre-2003 mistake: everything goes through developers.

Look at what it takes to launch a conversational experience today: custom RAG pipelines, agent orchestration, voice infrastructure. Powerful tools, for developers.

The people who should steer the relationship, marketing, content, brand, are spectators. They're being talked at about embeddings and multi-agent orchestration. An engineer's vocabulary for what is, at its core, a communication problem: the brand wants to talk to its customers, and customers want to be heard.

Between them sits a layer of complexity that excludes exactly the people who should be driving. Marketing is still waiting in the hallway.

An assistant demo comes together in days. Then what?

The content lead wants to adjust the tone for the summer campaign: ticket, developer, delay. The product team adds a collection: ticket. The brand manager wants to test two journeys: ticket. The experience freezes, ships once, and ends up in a corner of the site.

The web already solved this problem, for editorial content, twenty years ago. Pages, media library, plugins, preview, publish. Conversational content still has no such foundation. We know how to build. We don't know how to run.

That's exactly what Scenaro is. Not a metaphor, an architecture.

Scenarios are the pages. A page defines what the visitor sees; a scenario defines what the experience says, how it guides, when it adapts its messaging. Marketing builds them, tests them, evolves them, no code.

Collections are the media library. Where WordPress stores images and PDFs, Scenaro connects product catalogs, spec sheets, editorial content, natively wired to Shopify, Magento, WooCommerce.

Features are the plugins: multimodal product display, comparison, forms, suggested replies. Activate, configure, deploy. A React SDK for full custom.

Publications bring to conversational content what software has known for decades: staging, production, rollback. Every version traceable, every rollback possible.

And conversational analytics close the loop: what users ask, where they drop off, what converts. Marketing steers with facts.

One line of script to embed on any website. An interface non-technical teams own from day one.

When mobile apps arrived, everyone had to think mobile, everyone resisted, then everyone caved. Now that conversation is arriving, you have to think conversation-first. This isn't a distant prediction: it's the next design reflex, and it starts now.

WordPress took a decade to become indispensable, not because it was the best technical tool, but because it was the most accessible. The conversational web will follow the same path: not through experiences built once and abandoned, but through platforms marketing can drive.

The real question is no longer "how do we build?". It's "how do we...

The full article lives on Scenaro: Scenaro, the CMS of Tomorrow: Reinventing the Web Through Conversation

If you prefer the series by email: https://scenaro.substack.com/subscribe

Voice, Data, GDPR: What Brands Need to Clarify Before Going Live

Rémi Henriot — Sat, 22 Aug 2026 12:00:51 +0000

Voice, Data, GDPR: What Brands Need to Clarify Before Going Live

Voice is intimate data, and a generation of assistants treats it like a cookie. Brands that don't ask the data question before the POC will face it after.

Voice is intimate data, and a generation of assistants treats it like a cookie. Brands that don't ask the data question before the POC will face it after.

A voice is not a click. It's a timbre, a hesitation, a purchase intent spoken out loud, sometimes a health detail slipped into a sentence. It's what a customer entrusts to you when they speak naturally.

An entire generation of assistants, designed in the US, for the US market, treats this material like one more cookie: collected by default, sent who knows where, governed by terms nobody reads.

The scenario repeats itself: a team builds a voice POC in three weeks, the demo impresses, the project accelerates. Then the DPO asks their questions. Where are conversations hosted? Who has access? Can they be deleted? Which sub-processor, in which jurisdiction?

Silence. The vendor answers "we're GDPR-compliant", no detail, no contract, no architecture to show. The project dies there. Not because of the DPO: because of a tool choice made without asking the data question.

The order is the problem. Compliance can't be retrofitted, it's chosen upfront, contractually, in the very architecture of the platform.

Treating voice as intimate data isn't a legal constraint. It's a brand position. The customer who speaks to your conversational experience gives you a trust no form ever earns. That trust is protected, or lost all at once.

Brands deploying voice in Europe need three simple answers before signing anything: where the data lives, who can erase it, and what a visitor has to give up in order to speak.

Scenaro's infrastructure is hosted in Europe, in Paris. Conversations don't have to cross the Atlantic to exist.

Conversations are deletable. Not a manual process triggered by a support ticket, a capability of the platform.

And visitor identity stays anonymous: no mandatory login to speak. The returning customer is recognized without ever having handed over their identity. You can remember someone without knowing who they are.

Transparency can go all the way to the user themselves. At Urbansider, in production, a history panel gives travelers direct access to what the experience holds about them, conversations, memories, saved plans, with the ability to delete each item. Trust isn't declared in terms and conditions: it's shown on screen.

We don't sell turnkey compliance, no serious vendor can; every brand remains responsible for its own processing. What we offer is an architecture that makes the conversation with your DPO possible, early, with concrete answers.

Voice is coming to every customer relationship. Brands that treat this data with the respect it deserves will turn it into an advantage. The others will discover the topic during an audit.

Want to discuss Scenaro's data architecture? [Contact us]

The full article lives on Scenaro: Voice, Data, GDPR: What Brands Need to Clarify Before Going Live

If you prefer the series by email: https://scenaro.substack.com/subscribe

From Conversations to Personas: The Hidden Intelligence in Your Conversational Data

Rémi Henriot — Thu, 20 Aug 2026 12:02:12 +0000

From Conversations to Personas: The Hidden Intelligence in Your Conversational Data

Your web analytics measure clicks. Your customers, meanwhile, voice objections, criteria, and doubts, out loud. What your users say to your assistant is worth more than any consumer panel.

Your web analytics measure clicks. Your customers, meanwhile, voice objections, criteria, and doubts, out loud. What your users say to your assistant is worth more than any consumer panel.

You know the user left the product page after 45 seconds. You don't know they were hesitating between two bottles and couldn't find an answer to their food-pairing question.

The first fact costs you a dashboard. The second is worth a catalog decision. And your current tools will never see it.

Bounce rates, time on page, conversion funnels: twenty years of web analytics have perfected the measurement of behavior. And plateaued there.

A click doesn't say why. A session recording shows hesitant scrolls, not the question that went unanswered. A consumer panel? It captures what people think they should answer to your question. Not what they say spontaneously when nobody's asking.

A conversation does. The real objection to a price. The decision criterion no form ever captures. The moment of doubt right before add-to-cart. It's all there, in plain words. Brands that talk with their customers, and soon they all will, collect this material with every exchange.

A conversational experience in production generates thousands of conversations. Qualitative data at scale: the holy grail of marketing research, free, continuous.

But nobody reads ten thousand transcripts. And patterns don't emerge from linear reading. That's the paradox: AI produces data that only AI can process at scale.

On Scenaro, every conversation is stored and summarized automatically. Daily and weekly summaries are produced per scenario: recurring topics, questions that went unanswered, the signal of a gap in your collection, drop-off moments, requests your offering doesn't address yet.

And the analysis isn't generic: the scenario's persona, its character, its objectives, structures the reading. A scenario built to advise isn't analyzed like a scenario built to sell. What you receive isn't a keyword cloud: it's a synthesis shaped by what that scenario was meant to accomplish.

All of it aggregates trends, never individuals. The summaries speak about your customers in the plural, never about one customer in particular.

Your analytics will always tell you what happened. Your conversations tell you why. The question is no longer collecting this material, it arrives on its own. It's reading it. That's what Scenaro does for you, every week.

Want to see what a Scenaro weekly summary looks like for your sector? [Request an example]

The full article lives on Scenaro: From Conversations to Personas: The Hidden Intelligence in Your Conversational Data

If you prefer the series by email: https://scenaro.substack.com/subscribe

Memory Between Sessions: What Your AI Assistant Should Actually Remember

Rémi Henriot — Sat, 15 Aug 2026 12:12:40 +0000

Memory Between Sessions: What Your AI Assistant Should Actually Remember

Your customer spent twenty minutes opening up. Eight days later, your assistant asks "How can I help you?". That's not a technical limit. It's a choice, and it's costing you your loyal customers.

Your customer spent twenty minutes opening up. Eight days later, your assistant asks "How can I help you?". That's not a technical limit. It's a choice, and it's costing you your loyal customers.

The user gave everything: their tastes, Burgundy preferred, €30 budget, family dinners on Sundays. The assistant answered well, advised well. Then it threw it all away.

The customer comes back. "Hello! How can I help you today?"

That's not a bug. It's the market norm: amnesiac, stateless bots that reset to zero every session. The problem isn't GPT-4. It's zero context, zero scenario, zero memory.

Look at the demos from conversational platforms: they all show the first conversation. Never the second.

Because there is no second. An amnesiac bot treats your loyal customer exactly like a stranger, a salesperson who never recognizes anyone. Nobody comes back to talk to someone who forgot them.

Meanwhile, the rest of the digital world remembers. Amazon knows your history. Netflix knows your genres. Spotify knows your weeks. Your brand's assistant forgets. That asymmetry is untenable, and your users feel it: "I would have had to explain everything again" is the number one reason they don't return.

At Scenaro, memory is architected in three levels.

First level: visitor identity persists. The returning customer is recognized, no mandatory login, no friction. The relationship has a thread.

Second level: memories configurable per scenario. You decide, scenario by scenario, what deserves to be remembered, tastes, budget, occasions, and those memories are injected into the next conversation. After each exchange, a structured summary is generated: not a raw log, a usable synthesis.

Third level: session resumption with the state of the interface. The assistant doesn't just remember what was said, it remembers what was seen. "You were looking at the Bordeaux selection": the conversation picks up exactly where it stopped, screen included.

That third level is live in production at Urbansider: a traveler who closes the tab mid-planning finds, on return, a "Resume" button, and the conversation restarts on the exact screen where they left it, brief, map, and timeline included.

Good memory isn't memory that keeps everything. It's memory that keeps what serves the relationship, and stays governable: every element is deletable, and the brand decides what gets remembered.

Governance extends to the user too. At Urbansider, a history panel gives them access to everything the experience holds about them: past conversations, memories, saved plans, all viewable and deletable. That transparency isn't a regulatory burden to endure. It's what makes memory acceptable: people trust an advisor who remembers, not a black box that records.

An assistant that remembers is the difference between a tool people try once and a relationship they maintain. Without a memory architecture designed from day one, you're building loyalty on sand.

Want to understand how to configure memory on Scenaro? [Let's talk]

The full article lives on Scenaro: Memory Between Sessions: What Your AI Assistant Should Actually Remember

If you prefer the series by email: https://scenaro.substack.com/subscribe

Connecting Your Product Catalog to an AI Assistant: It's More Than an API Call

Rémi Henriot — Thu, 13 Aug 2026 12:00:43 +0000

Connecting Your Product Catalog to an AI Assistant: It's More Than an API Call

Plugging Shopify into an AI assistant takes an afternoon. What you get is a search engine in disguise, not an advisor. The difference is architecture, not the API.

Plugging Shopify into an AI assistant takes an afternoon. What you get is a search engine in disguise, not an advisor. The difference is architecture, not the API.

The pitch is everywhere: "connect your catalog in a few hours." A webhook, two endpoints, a demo that handles "show me your running shoes." Sold.

Except the demo lies by omission. What you just wired up is keyword search with a voice. A duct-taped RAG pipeline that matches strings and recites product sheets.

The assistant can answer. It can't advise. It lists. It doesn't recommend. And that difference shows up straight in your conversion rates.

A product API gives you access to data. Not to meaning. Ten thousand SKUs loaded raw into a language model are ten thousand rows of metadata.

Ask that system for "a light wine for a summer dinner" and it greps your descriptions for "light" and "summer." It doesn't understand you mean a low-tannin red, served cool, under €20.

The user asks a customer's question. They get a database's answer. They wanted advice; they got a list. That's where the conversational promise dies, politely, factually, one exchange at a time.

At Scenaro, a connected catalog is called a collection. Shopify, Magento: synchronization is native, the catalog stays current. But the connection is only the starting point.

Every field in a collection can be indexed three ways: full-text for exact terms, semantic for meaning, not words, phonetic for what the ear actually hears. And it's configurable collection by collection, to fit your business.

The result: the agent doesn't match strings. It understands "a light wine for a summer dinner" and goes looking for what that means in your cellar. That's what runs in production at Club Français du Vin, our first client.

And when the request outgrows a product query, advanced search profiles take over. At Urbansider, the Paris travel concierge, the conversational experience doesn't return a list of addresses, it orchestrates the generation of an actual itinerary.

Anyone can connect a catalog. Making an agent understand it, field by field, intent by intent, is architecture. That's what Scenaro builds.

Want to see how catalog indexing works on Scenaro? [Request a demo]

The full article lives on Scenaro: Connecting Your Product Catalog to an AI Assistant: It's More Than an API Call

If you prefer the series by email: https://scenaro.substack.com/subscribe