Getting to know the assistant

The two halves (the question tree and the chat), the ways visitors get there, and how to put it to work.

Search serves whoever already knows what they want. The assistant serves the rest: whoever arrives with a problem ("a gift under €50", "which of these is for me") and not with a term.

It lives in Search → Assistant and has two halves, which are independent plan features:

• The question tree: a path written by you, with closed questions that lead to a result (a selection of products and categories). Deterministic, predictable, and with no cost per visit.
• The chat: the customer writes in their own words and the assistant answers, searching the catalogue. It only recommends what the search returns: it doesn't invent products or prices.

The Questions + chat mode combines the two: the tree handles the common cases, the chat covers the rest. If the plan only includes one of the halves, the screen says which one is missing and from which plan it exists.

While it's still taking shape The assistant screen is up and working, but it isn't announced in the panel sidebar yet: you get there by the direct address. It will be announced when the path is finished end to end. If you're building one now, talk to us and we'll pass you the way in.

Putting it to work

With no assistant created, the screen only asks for the store and a name. After that, the settings:

• Text of the button in the store: it's the first thing the customer reads, and it influences the start rate more than the whole tree does.
• How it moves on: questions only, questions + chat, or chat only (the list only shows the modes included in the plan).
• Status: Draft (invisible in the shop, so you can build calmly) or Active.
• Handoffs from search: the doors through which search hands somebody over to the assistant, coming up next.
• Notes for the assistant: free text that goes into the context of every conversation. It's for the way of answering and for half a dozen short facts; anything that's real information belongs in the knowledge base, which is only read when somebody asks.

For the button in the shop there are two ways: mark any button in the theme with data-buskara-quiz, or use the direct link https://your-shop/?bk_quiz=1, handy in campaigns and newsletters because it opens the assistant on arrival.

The ways visitors get there

Besides the button, three doors lead out of search itself, and they're switched on one by one:

• The box opened with nothing typed. Whoever opens the search and doesn't type doesn't know where to start, and that's exactly what the assistant is for.
• A search with no results. It avoids losing someone who was already buying: instead of the dead end, a question.
• A search typed as a question. "Is it suitable for sensitive skin?" isn't a catalogue term, it's a conversation starting in the wrong box.

The "Where they came in" table, at the top of the screen, splits the starts by these doors. It's worth looking at it before touching the tree: if nobody comes in, the problem isn't the questions.

The chat, from the customer's side

It survives navigation: if the customer changes page or reloads halfway through, on reopening they find what they'd already said and the products they were shown, and carry on from where they were. Reviewing it doesn't count as a new message nor spend from the plan.

And it's managed like the chats they already know: they can start a new conversation without losing the previous one, go back to the session's conversations through the list (the label of each one is the first question) and delete the ones they don't want to see again. Deleting is from their own screen: deleted conversations stop showing in the customer's browser, but they stay in the platform records, where the spend is measured and a disputed answer can be audited.

The assistant also sees the cart as it is: every message carries a snapshot of the basket (product and quantity), including what the customer added through the shop's buttons outside the conversation. That's what lets it answer "yes, it's already there" instead of arguing, and pick the purchase up halfway without repeating anything. The snapshot is just that; it carries no personal data at all.

With AI-assisted purchase in the plan, the chat closes the loop all the way to checkout. See AI-assisted purchase.

Switching the AI on and off

In Store → AI in the store are the switches, one per feature: the chat, the AI-assisted purchase, the AI search on dead ends, the generative suggestions while typing, and the tone of voice it uses with visitors.

The plan decides which ones exist; the switches decide which ones show up in your shop, and each one turns off one thing without touching the others. Everything starts switched on, and a choice made on a plan that has since changed is kept: on an upgrade it applies again without you having to remember it.

Usage limits

The chat has fair use limits, designed never to show up to a real customer: a ceiling on messages per session and per day, and another per IP address. They're there so that a script or a curious visitor hammering away all night doesn't eat your plan or our models.

When a ceiling is reached, the assistant says goodbye politely and the question tree carries on working; at midnight everything goes back to normal. If your shop has an out of the ordinary usage pattern, talk to us: the ceilings are adjusted per account.

The spend itself is measured against the plan's AI allowance, and it's explained in plan, requests and payments.

Measuring

The top of the screen shows the funnel for the last 30 days: how many started, how many reached the end, the clicks on products, the attributed revenue and, with the chat switched on, the conversations and the spend for the period.

Further down, two tables do the fine diagnosis: "Where they give up" shows the drop-off per question (above 40% it goes red, usually a sign of one question too many or a badly worded one) and "Results that work" shows views, clicks and thumb feedback. Each question and result also shows its own statistics inside the editor itself, in the place where you fix them.