Asks an AI what to buy
- → Answers and recommends She asks in full sentences What moves: her question
One shopper, one purchase, five parties. Follow it across the map and you will see what has to be true before an AI-mediated sale can happen at all, where Herm measures and decides, and the one join nobody has built yet.
States on this map reflect availability in our markets. In-chat checkout is live in one geography, with one AI provider, not ours. Everywhere else, the answer still has to send her somewhere.
Dashed lines mean Roadmap or Not generally available, and nothing else.
Brand Visibility runs repeated controlled observations across the five-model panel and records what came back: the brands named, their order, the sources cited and the explanation the answer itself gives.
Product Readiness runs representative shopping situations through controlled Shopping Tests against the brand's connected product information, and reports where they fail and why.
Customer Intelligence ranks the options a brand already has, for a shopper who chose to connect, and returns the ranking. The shopper's profile stays inside Herm. Early access.
Where consent and coverage exist, a covered purchase outcome confirms what was bought, where, when and for how much.
Herm cannot prove that one specific AI answer caused one specific cross-surface purchase, because the durable answer-to-transaction identifier required for that join is not generally available.
“fragrance-free vitamin C serum gentle enough for rosacea, under £40, ships to Germany”
Can a model name Velora in this category at all?
answersDoes the catalogue expose the form, not just the strength, where a machine can read it?
productIs the price current, and is there a reason to buy now?
product · rewardsCan the answer end somewhere she can actually buy?
buysFour clauses. Four different things that have to be true. Miss one and the sale dies quietly, and nobody ever learns it happened.
The purchase comes back verified. Which answer caused it does not.
Select any node on the map to see what moves through it.
the decision moves · the profile does not
no durable identifier exists to join on
The same map, one relationship at a time: who it runs between, what moves, what Herm does there, and whether it happens yet.
One shopper, one purchase: she asks, the AI answers, she arrives at the brand and buys. Herm is in none of it.
Asks an AI what to buy to Answers and recommends
Category, suitability, price and delivery in one line. Herm is not in this conversation and will not be. Brand Visibility observes questions like this one, put to the panel under controlled conditions; it never sees hers.
Answers and recommends to Checks specs, price, stock, delivery
The surface decides which brands are candidates and states why in the answer: which claim it points to, which source it cites. Most brands lose here, silently, with no impression and no line in anyone’s analytics.
Checks specs, price, stock, delivery to Product information the brand publishes
The AI surface reads what the brand itself publishes: its pages, or a feed it submitted. Herm is not in this path, does not proxy it and does not route it. Product Readiness tests the same information separately, in the loop beneath.
Checks specs, price, stock, delivery to Signs in to the brand
In our markets the answer routes her to the brand’s own store. An answer that cannot quote a price, confirm stock or reach a destination that ships to her recommends someone else.
Signs in to the brand to Brand-owned signed-in experience
From here she is on the brand’s own property: its product carousels, its site assistant, its offer slots. This is the only place on the map where Customer Intelligence operates.
Brand-owned signed-in experience to Merchant checkout, orders, payments
The merchant is the merchant in every version of this. However she arrived, the transaction happens on the brand’s own commerce stack.
Merchant checkout, orders, payments , going nowhere, by design
No version of this passes through Herm. Herm does not create carts, process payments or perform checkout. The gap is deliberate and permanent.
Two measurements, both beside the journey: what the answers say, and whether the catalogue works.
Answers and recommends to Brand Visibility
Repeated controlled observations across the supported model panel, up to weekly, signed out and personalisation-free. Herm records what came back. It does not read a model’s internal reasoning, it does not see this shopper’s conversation, and a cited source is not proof that the source caused the answer.
Product information the brand publishes to Product Readiness
The brand connects its product information to Herm for testing. This loop sits beside the journey, not inside it: it tells the brand whether it is ready for AI shopping, and it carries no real AI shopping interaction.
Product Readiness to Product information the brand publishes
Representative shopping situations run against the connected catalogue in a controlled environment. They are not the live consumer experience of ChatGPT, Claude or any other third-party assistant, and a result describes the brand’s readiness, not what an external system did.
Two decisions, on surfaces built for them. The decision moves. The shopper profile does not.
Brand-owned signed-in experience to Customer Intelligence
Only inside an eligible signed-in session, and only for a shopper who chose to connect. The brand asks “which of these?”, never “who is this?”. There is no profile lookup, no attribute response, no batch mode and no offline channel.
Customer Intelligence to Brand-owned signed-in experience
Herm decides internally, using context the shopper permitted, and returns the decision the brand asked for. The brand acts on the order it gets back.
Customer Intelligence , going nowhere, by design
This is the design, not a limitation. The profile Herm holds never reaches the brand: not as attributes, not as segments, not as an export, and not as anything written into a CRM. What the brand keeps is the decision it was given.
Brand-owned signed-in experience to Offers
The brand states what it will give, and on what terms. Herm decides relevance internally while serving the shopper. This is not audience buying: the brand does not select, reach or receive individual shoppers.
Offers to Claims an offer in Herm
The shopper sees offers in Herm's own consumer experience and chooses whether to claim one. Herm genuinely carries this, because the surface is Herm's. The brand never receives the shopper's underlying profile.
Claims an offer in Herm to Merchant checkout, orders, payments
The claim happens on Herm. The purchase happens on the merchant’s own stack, like every other purchase on this map, and Herm is not in the money here either.
Offers to Brand-owned signed-in experience
Redemption, repeat purchase and verified revenue, in aggregate. A receipt-verified purchase outcome confirms a purchase happened under the applicable consent rules. It does not establish which AI answer, or which offer impression, caused it.
The purchase comes back verified. Which answer caused it does not.
Merchant checkout, orders, payments to Covered purchase outcome
Consented, itemised, revocable, and true across every retailer she uses rather than one store alone. This is the only part of the map that comes back verified, and it is live.
Covered purchase outcome to Offers
Covered outcomes are what make Be Preferred measurable: what share of a category’s buyers also buy the brand, and whether they buy it again. Where receipt panel coverage is too thin, the dimension is reported unavailable, never as zero.
Covered purchase outcome to Answer-level cause
What was bought, where, when and for how much is verified. The next step, establishing which observed answer produced it, is where the map stops.
Answer-level cause to Answers and recommends
The line out of this node is dashed and visibly does not arrive. That is missing infrastructure, not a roadmap item, and the two are not the same.
More AI shopping surfaces to Merchant checkout, orders, payments
Roadmap. As more AI shopping surfaces open, structured product and commercial experiences could be distributed into them, and the merchant would still be the merchant. Nothing here runs today, and AI Distribution is the destination of the framework, not a fifth readiness rung.
One provider, in one geography, can already tell a merchant that an order came through its own chat. That is the channel, not the answer, and it stops at the edge of that provider. It says nothing about the shopper who asked on Tuesday and bought on Velora.com on Friday, which is most of them. Until an identifier survives from the answer to the transaction, a verified purchase is exactly that: verified, with its cause unknown.
Consumer buying is not a requirements document. People buy a perfume because they liked it, not because it scored well against criteria. Five general assistants can serve the rational half of that. They are much weaker on taste, on wardrobe, on occasion, and on the hundred niches underneath, and the apps that are good at those things are already appearing. Stylists that hold your wardrobe and tell you what’s missing exist today.
So we think AI shopping spreads across thousands of them, each good at one segment. We can't prove it. It's why AI Distribution is drawn on this map as Roadmap rather than as a fact, and we'd rather say so.
“Herm-side” means we have built and implemented our half. It runs the moment a counterparty implements the other half. Technical details are available to connected brands and partners inside Herm rather than as public standalone specifications. There is more than one agentic checkout protocol and they are not converging yet, which is the whole argument for a layer that sits above all of them.
Four dimensions, measured separately: whether AI represents you, whether your product information works, whether your signed-in experience can choose better, and whether your category's buyers choose you. Where a dimension can't be measured yet, you'll see that, not a zero.