How to set up your store's customer service with AI

TLDR: The bulk of a store’s queries are five repeat questions whose answers are already written on your site. Tidio puts an agent that resolves them by reading your documentation and escalates the rest to a person, from $24.17 a month. What decides the result is not the tool, it is rewriting those pages properly before connecting it, and reviewing weekly what the agent could not answer.

This guide is for anyone selling online watching customer service eat their evenings: small shops without a support team, direct-to-consumer brands, businesses where the owner answers messages at eleven at night. The goal is not removing human contact, it is having people handle what deserves a person.

The warning upfront: an agent connected over bad documentation answers badly at scale, which is worse than not answering. The real work sits in steps one and two.

1. Sort the questions you actually receive

Before touching any tool, open your inbox from the last two months and count. Do not estimate: count. The real list almost always looks like this:

Note next to each what you normally answer. That document is the agent’s syllabus and, incidentally, reveals something uncomfortable but useful: most of those questions exist because your site does not answer them clearly.

Questions that appear once in a blue moon do not belong on this list. Automating the exception is the mistake that makes an agent look stupid exactly when the customer genuinely needs help.

2. Rewrite the documentation the agent will read

This is the step that decides accuracy. An agent answers with what it finds written, so if your returns page says “we accept returns within a reasonable period”, the agent will serve that vagueness to everyone.

Rewrite your three or four key pages with concrete answers: exact return window in days, who pays return shipping, what happens if the product arrives defective, delivery times by zone, what to do if the order does not arrive. Every concrete fact you write is a query you stop receiving, whether the agent answers it or the customer reads it directly.

To speed up the rewrite, Claude turns your notes into clear answers, from $17 a month and with a permanent free tier. Precision beats variety here: every sentence you write is a commitment a customer can hold you to. The review is yours, because you are writing commercial commitments and approximation has no place there.

3. Build the agent with Tidio

Tidio is the piece the customer sees: store chat, tickets and an AI agent that answers by reading your documentation. It connects with the usual commerce platforms, which lets it answer about specific orders rather than only general policy. Free plan and a card-free trial, with paid plans from $24.17 a month.

A customer would rather wait ten minutes for a human than receive three answers that do not help.

4. Review weekly what it could not answer

Here is the habit that turns a chatbot into a system that improves. Every week, a short pass over the conversations the agent escalated:

That cycle has a valuable side effect: your public documentation improves, so queries also drop among people who never open the chat.

If your product is technical and you need answers over manuals, specifications or extensive documentation, FastGPT builds a knowledge base on your own documents with more control at low cost, from 99 yuan a month, billed in its own currency. It sits behind the customer-facing agent as the layer that knows the product.

With support solved, the freed time usually goes to acquisition: the product photos and catalog descriptions guides cover that. The whole sector lives in AI for e-commerce.

Common mistakes

Connecting the agent over vague documentation. It multiplies your ambiguity across every customer at once. Write first, automate second.

Configuring the agent never to escalate. It saves billable conversations and destroys trust. Early escalation is a feature, not a failure.

Not reviewing the conversations. Without the weekly cycle, the agent stays frozen on what it knew on day one while your catalog and policies change.

Automating serious incidents. A lost order or a duplicate charge gets a person. The saving there costs a fortune in reputation.

Frequently asked questions

How many queries can the agent resolve?

With concrete, well-written documentation, the bulk of repeat questions. The proportion depends far more on your shipping and returns pages than on whichever AI model sits underneath.

How much does it cost?

Tidio has a free plan and a card-free trial, with paid from $24.17 a month. The important warning is that it bills by conversations, so measure your real volume during the trial before choosing a plan.

Do customers reject talking to an AI?

What they reject is not getting their problem solved. An agent that answers quickly and well, and hands off to a person as soon as it stalls, generates no complaints. One that insists with generic answers does.

Do I need to integrate it with my store platform?

That is what separates answering general policy from answering about the specific order in question. It works without integration, but at half its usefulness.