Customer support starts before the ticket

Ask a team how they want to automate support and you will hear about the inbox. Triage, tagging, suggested replies. All of it starts after the customer has already written in.

By then you have paid twice. Once for the ticket. Once for the sale that did not happen while they waited.

The question is a purchase in progress

Someone asking whether a product fits, what it is made of, whether it works with what they own, is not a support case. They are deciding. Answer in seconds and they buy. Send them to a contact form and they leave.

So the first rule is about place, not technology. Put the answer where the question is asked. On the page. Not behind a form, not buried in a help centre.

Ground it, or do not ship it

An assistant that answers from general knowledge will eventually invent something. On a physical product, one confident wrong answer about compatibility produces a return, a bad review, and a ticket that costs more than the sale was worth.

The fix is a constraint, not a better model. The assistant answers from your product data, and says what the catalogue says. When the catalogue is silent, it says it does not know. That sentence is not a failure. It is the system working.

If you cannot ground it in your own data yet, you are not ready to put it in front of customers. Fix the data first.

Read the transcripts, all of them

The answers are half the value. The other half is the questions.

Every conversation is a customer telling you what your page forgot to say. Export the full history and read it. Be careful here: many tools show only the opening message of each thread in their own interface, which makes the volume look far smaller than it is and hides where conversations actually went.

Sort the questions by product. The list that matters most is the one people skip: the questions the assistant could not answer. That is your missing content, written by your customers, in their words.

Then make the assistant less necessary

Turn the recurring questions into the page itself. A product FAQ, a clearer spec, a better photo. Each one you fix is a question nobody has to ask again.

This is the part most teams miss. When one product draws far more questions than the others, the customer is not the problem. The page is. Support data is a product roadmap that someone has been filing under "tickets".

What this looks like when it works

Fewer tickets, because the question was answered before it became one. Better pages, because you finally read what people asked. And an assistant that does less every month, which is the point.

We built this for an online store. The figures are in the product page AI case study.

If questions are reaching your inbox that your pages should have answered, bring us one page.