Home/Services/AI Support Agent

Knowledge base · on-site AI support

Support that has actually read your material.

Generic chatbots can't answer because they don't know what you sell, how your shipping is calculated, or what your return policy says.

Your brand material, product catalog, policies and past conversations become a knowledge base first — then the widget answers from it, naming the actual size, material, lead time and tracking number instead of saying "please contact support."

Pre-sale guidance Post-sale answers Image replies Order lookup Multilingual
01

The knowledge base comes first

This is where the difference is made. The more complete the material, the more specific the answers — and this material already exists, scattered across your site, back end and support history.

Gather material

Brand background, product specs and selling points, size charts, materials, shipping and lead times, returns and privacy policies, existing FAQs — plus the questions your support history shows get asked most.

Build the knowledge base

Structured into knowledge entries, with products and policies kept in sync with your back end — change a price, add a product, adjust shipping, and support knows.

Set the answering rules

Tone matched to your brand, banned words written in, and the questions that must escalate to a human decided here. It won't promise anything you never wrote down.

Embed on the site

One snippet on your site — Shopify, WordPress, Squarespace or Shoplazza — styled to your brand colors, not an obvious third-party bubble.

Keep feeding it

Real conversations show what it answers poorly and what gets asked repeatedly; those go back into the knowledge base. The longer it runs, the sharper it gets.

02

What it takes off your hands

In real support volume, pre-sale questions and order lookups are the vast majority — and they're the most repetitive part, the part that least needs a person.

01
Pre-sale guidance

"I'm 165cm, 60kg — which size?" "Does this material pill?" "How is this different from the other one?" Answered from real product data, with the right product link handed over.

02
Post-sale answers

How returns work, when refunds land, what happens with a wrong item, what the warranty covers. The policy text is in the knowledge base, so answers match your policy pages exactly.

03
Image replies

Replies can include product photos, size charts and assembly diagrams. Customers can also send a photo asking "is this part broken?" and it reads the image before answering.

04
Order lookup

Connected to your store back end, an order number or email returns status, tracking number and expected delivery — no waiting for a person to go dig it out.

05
Multilingual

It answers in whatever language the customer writes in, so a multi-market store doesn't need separate support staff per language.

06
Conversation data

Every conversation is exportable. What customers care about and where they stall before buying feeds directly into product pages and content topics.

03

When it should still hand off to a person

It can take the everyday questions entirely. But I'd set the following to escalate by default — not because it can't handle them, but because these shouldn't be decided by AI alone.

Escalates to a human by default
  • Anything involving money as an exception: disputed refund amounts, requests for compensation, discounts beyond policy — concessions involving money are decided by a person.
  • Complaints and heated conversations: when a customer is clearly unhappy it hands off immediately, passing the context along so nobody has to repeat themselves.
  • Questions the knowledge base doesn't cover: better to say "let me check that with a colleague" than to guess an answer that merely sounds right — a wrong guess costs far more than a slow reply.
  • Large or custom orders: wholesale inquiries and custom requirements belong with a salesperson anyway.

Put another way: it replaces repetition, not judgment.

For most stores the repetitive questions are the bulk of the support load — hand that off, and your people can focus on the few conversations that genuinely need them.

Start with what material you have.

How far this can go depends on what's available now — how complete your product specs are, whether policies are written down, whether you have support conversation history. Tell me where things stand and I'll work out what needs filling in first.

Knowledge base before widget

Organizing the material is the bulk of the work and what decides answer quality — that step never gets skipped.

Platform agnostic

Embeds on Shopify, WordPress, Squarespace or Shoplazza, styled to match your brand.

Improved after launch

Real conversations surface weak answers, which go back into the knowledge base — it sharpens with use.