Case study · AI
Support FAQ Assistant.
A website assistant answering from an agreed FAQ set
We added a focused chat assistant to a company's site — answering the questions it was asked all day from an agreed set of FAQs, and handing off cleanly when it couldn't.
Start your project →- Industry
- AI & automation
- What we built
- 3 integrated components
- Key outcome
- Top 20 Questions answered automatically
Overview
Answering the same questions, automatically.
A small team was spending hours a week answering the same handful of questions over email and chat — hours, returns, delivery, sizing.
They wanted a website assistant to handle the repetitive ones, without a big data project or any access to their internal systems.
- The same questions asked over and over, all day
- Support time swallowed by repetitive, low-value replies
- No appetite for a complex data or systems integration
- A worry the bot would confidently make things up
The goal was simple: deflect the repetitive questions accurately, and hand off cleanly when unsure.
Experience flow
From a visitor's question to a reliable answer.
The solution
A focused assistant, grounded in your FAQs.
We embedded a chat assistant configured for one job — answering from an agreed FAQ set, in the company's voice, with a clean handoff when it couldn't.
Core components
Three small pieces: the content it answers from, the widget it lives in, and a safe way to say 'I'll get a human'.
-
01
Agreed FAQ set
Content
Curated the questions the team actually gets
Wrote clear, on-brand answers for each
Set the assistant's instructions and tone
-
02
Embedded chat
Widget
A chat widget embedded on the site
Answers in plain language from the FAQ set
Tested against a set of agreed example questions
-
03
Graceful handoff
Safety
Says clearly when the answer isn't in the FAQs
Hands off to email or a human instead of guessing
Model usage billed at cost
Stack used
- TypeScript widget
- Node.js API
- FAQ store
- Anthropic
- Cloudflare Turnstile
- Sentry
Implementation approach
Phased delivery.
-
01
Agree the FAQs
Gathered the questions the team gets most and wrote clear, approved answers for each.
-
02
Build and tune
Embedded the widget, configured the instructions and tone, and tuned it against agreed examples.
-
03
Test and launch
Checked answers against the agreed questions, set the handoff behavior, and went live.
Results
Outcomes at a glance.
- FAQ-grounded
- No systems access
- Safe handoff
Every morning it was the same three questions — where's my order, how do returns work, what size should I get. The assistant takes those now, and it answers only from our published help center: if the returns policy doesn't say it, the assistant doesn't say it either. Anything unusual comes to us with the whole conversation attached, and it has never once invented a discount. We were live in three weeks, and our order system never knew it happened.
- Took the repetitive questions off the team's plate
- Answered from approved content, not guesses
- Handed off cleanly instead of making things up
- Launched in weeks without touching internal systems
Impact & outcome
Small assistant, real time back.
We gave a small team back the hours they were losing to the same few questions — with an assistant that answered from approved content and knew when to step aside.
The discipline was the boundary: a focused FAQ assistant that says 'I don't know' is far more useful, and far safer, than a clever one that guesses.
Because it touched no internal systems, it launched in weeks and carried no data risk — a genuine first AI feature, not a project.
Answering the same questions all day?
We build focused AI assistants grounded in your own FAQs, with a safe handoff when they're unsure — live in weeks. We'll reply within one business day — no pressure, no sales script.
Start your project →