Fixing it costs less than ignoring it.
A made-up refund policy or a customer who gives up halfway will cost you more than fixing what caused it. Here’s what each piece of work involves and what it costs.
How a fix works.
Every fix goes through these four steps.
- 01
Read
I talk to your assistant the way your customers do, from friendly to awkward.
- 02
Diagnose
Every failure quoted word for word, and traced to the instructions, the content or a missing decision.
- 03
Rewrite
I rewrite its instructions and help content so it sounds like your brand and knows when to hand over.
- 04
Re-test
The whole thing again, so one fix doesn’t break another.
Prices.
Fixed prices, so you know before we start. Begin with the free first read and decide from there.
First read
I try your live assistant and tell you what I find.
- The three problems that matter most
- One real exchange, rewritten
- By email, no call
Diagnosis
A full test, friendly to hostile, written up so anyone can act on it.
- Every failure quoted word for word
- The cause of each one
- Fixes ranked by impact
- Hand it to your builder, or to me
The fix
I make the changes, then re-test everything.
- Instructions and persona, rewritten
- A voice guide for your assistant
- Fallbacks and handover to people
- Help content rewritten for chat
- Full re-test afterwards
Monthly
Assistants drift when your site, prices or model change.
- Re-test after every change
- Real conversations read
- New problems fixed
- Cancel any time
Works with Intercom Fin, Chatbase, Tidio, Voiceflow, ElevenLabs and phone agents, or anything built on GPT, Claude or Gemini.
Build assistants for clients?
I’ll do the voice and testing pass before you hand it over, under your name.
Questions.
What’s the free first read?
I talk to your assistant the way your customers do, send you the three problems that matter most, and rewrite one real exchange so you can see the difference. It costs nothing. If it’s in good shape, I’ll tell you so and leave you alone.
What platforms do you work with?
Website chatbots and phone agents built on tools like Intercom Fin, Chatbase, Tidio, Voiceflow and ElevenLabs, or built in-house on GPT, Claude or Gemini. If it talks to your customers, I can work on it.
Can’t I just set it to “friendly” and “concise”?
You can, and you should. But a dropdown isn’t a voice. It won’t decide what your assistant does when it doesn’t know, when it hands over, or how it says no without losing the customer.
We didn’t build it ourselves. Can you still fix it?
Yes. I can make the changes inside your account, or write them up so whoever built it can apply them.
Do you need access to our account?
Not for the first read or the diagnosis. For the fix, I work inside your account while we’re working and you remove my access when we’re done. I keep test conversations for your report and nothing else.
Does my assistant need to say it’s an AI?
If you have customers in the EU, the EU AI Act has required it since August 2026. It’s good practice anyway. I check for it in every diagnosis, though I’m not a lawyer, so check the detail with one.
Send me your assistant. I’ll try to break it, so we can mend it.
It’s free. You get the three problems that matter most and one exchange rewritten.