The agent is one campaign. Not the only one.
The AI agent answers out of your real catalogue — and the page underneath it does not stop personalizing while it is open. Banner, hero, recommendations and the agent’s own register all move together, including a complete switch to trade terms for a wholesale buyer. Switch the visitor below to see all four.
Made for weather that actually turns up
British wool, Portuguese knitwear, and a returns window that runs to sixty days.
An agent that knows the catalogue, the visitor, and the campaign.
Give it the catalogue
Products, articles and pages are fed in and re-read daily, so the agent answers with real items, real prices and real stock — and returns them as rows you can click, not a paragraph describing them.
Give it an audience
The agent is an ordinary campaign, so it targets like one. A wholesale visitor resolved by an ABM integration gets a different assistant, with different terms and a different opening line.
Let it open with an answer
Ask it a question of your own before the window opens and the answer becomes the greeting. One sentence covers every segment because the variables are filled per visitor, and the answer is cached, so a thousand visitors cost one call.
How Personyze personalizes this store — agent included.
The greeting is computed, not scripted
The first line is generated from what is known at that moment — the city and its forecast for a stranger, the saved basket for a returning shopper, the resolved company for a trade buyer. The visitor never sees the question that produced it.
It answers with rows from the catalogue, not prose
Ask about a product and the reply carries the real item: image, price, colours and stock. Those come from the same feed the recommendation engine uses, so the agent cannot invent a product or quote a price that is no longer true.
The banner and the hero are separate campaigns
The delivery-threshold strip, the early-access strip and the trade strip are three ordinary actions with three audiences. None of them is inside the agent, and each is measured on its own.
Recommendations change model with the visitor
Bestsellers for a stranger, bought-together for a basket, viewed-and-held for a member, and a trade catalogue with MOQs and margins for a retailer — the same engine, different inputs.
Offers can be suppressed as deliberately as they are shown
The VIP sees no discount, because discounting a customer who already buys at full price costs margin for nothing. Suppression is a targeting rule like any other.
A trade visitor gets a different assistant entirely
Register, catalogue, pricing and call to action all switch on firmographics. The consumer offers are not merely hidden — they are excluded from that audience.
Put an agent like this on your store.
It reads your catalogue, follows your audiences, and works alongside the banners and recommendations you already run — rather than replacing them with a chat window.