An AI sales agent that diagnoses before it sells
Most store chatbots answer questions. A sales agent moves the customer forward. This one reads your catalog, prices and return policy, works out what the shopper is actually trying to solve, and puts two or three right products in front of them — then takes the order without handing off.
Why it converts where a chatbot stalls
It asks before it recommends
A shopper who types "recommend something" gets a diagnosing question and tappable options, not forty products. Narrowing to two or three is what turns browsing into buying.
It only says things it can prove
Prices, star ratings, review counts and return windows come from your own pages. If the crawler did not find it, the agent says so and offers to have someone confirm — it does not improvise.
It finishes the transaction
Cart with quantity controls, an auto-applied welcome code, a free-delivery progress bar and one-click checkout, all inside the conversation.
It recovers the ones about to leave
Price hesitation triggers the voucher card. Competitor mentions trigger an honest side-by-side. Frustration drops the sales motion entirely and offers a phone callback.
How it gets built
- 1
Paste your store URL into the Train studio.
- 2
The crawler reads up to 15 pages — product feeds, JSON-LD schema, policy pages, logo and brand colour.
- 3
A persona is written for your vertical, grounded strictly in what was found.
- 4
You get a live demo on a shareable link, in your brand colours.
- 5
Ship it with one script tag when you are happy with it.
Questions people actually ask
- How long does the build take?
- The crawl typically finishes in 20-60 seconds depending on how many pages your site exposes and how fast it responds.
- Will it invent products or prices?
- No. The system prompt restricts it to the extracted catalog and policy text, and the fallback engine reads directly from the same data. Anything it cannot verify, it declines to state.
- Does it work on non-Shopify sites?
- Yes. Shopify gets a clean product feed for free, but the crawler also reads schema.org JSON-LD, microdata and ordinary product-card markup, so WooCommerce, Webflow, WordPress and hand-built sites all work.
- What happens if it cannot answer?
- It escalates rather than guesses — a phone callback in under 30 seconds, a booking slot, or a message to your team, depending on what the customer needs.
See it on your own catalog
No signup, no sales call. Paste a URL and you will have a working demo on a link you can share.
Start the crawl