I built my dad's painting business an AI that books jobs while he's up a ladder
My dad runs a decorating business in Hove. I gave his website an AI assistant that qualifies visitors, books free quotes, and texts him the lead. Here is how it works.
Key takeaways
- A small trades website is usually a brochure. Most visitors leave without ever making contact.
- An AI assistant catches that visitor, qualifies the job, and books a free site visit.
- It never quotes a price. It captures the lead and lets my dad do the quoting.
- Every chat lead flows into the same pipeline as the contact form, so my dad gets a text within seconds.
- I built it config-driven, so the same assistant clones onto any trade in a config change, not a rebuild.
My dad runs a painting and decorating business in Hove. Last week I gave his website an AI assistant that books jobs for him while he is up a ladder.
Her name is Helen. She sits in the corner of hovedecor.co.uk, and when someone lands on the site she does the thing my dad cannot do when he has a roller in his hand: she talks to them.
What it actually does
Helen turns a silent website visitor into a booked quote. She has a short, friendly conversation, works out the job, the area and the timeline, takes a name and number, and hands the lead straight to my dad.
Most small trades sites are a brochure. Someone visits at 9pm, reads a bit, means to call tomorrow, and forgets. That visitor is gone, and my dad never knew they existed.
Helen catches that visitor. She works through four things:
- What is the job (interior, exterior, a whole house, a communal block)
- Where it is (she confirms the area my dad covers and politely bows out if it is too far)
- Roughly when they want it done
- A name and a phone number
That is it. She is not trying to be clever. She is trying to turn a silent website visitor into a booked quote.
How it works under the bonnet
The widget is a small React component on the site. The brain is a single API route running Claude Haiku 4.5 through the Vercel AI SDK. The responses stream in as she types, so it feels like a conversation rather than a form.
The important part is what happens at the end. When Helen has qualified someone, she calls a tool that fires the exact same lead pipeline the website contact form already uses. So a chat lead flows straight into the system I built for my dad earlier: it lands on a sheet, an auto-reply goes out, and my dad gets a text on his phone with the lead.
The chatbot is just a second front door to an engine that was already running.
Speed matters more than people think. The longer a lead sits, the colder it gets. Helen exists so the gap between "someone was interested" and "my dad is calling them back" is as short as possible.
What I deliberately did not let it do
Helen never gives a price and never leaves her lane. She books the free site visit, refuses off-topic and jailbreak attempts, and tells anyone who asks that she is the online assistant, not a person.
This is the part I care about most.
Helen never gives a price. Decorating quotes depend on the job, and a bot guessing a number is worse than no number at all. So she books the free site visit and lets my dad do the quoting.
She also stays in her lane. She only talks about Hove Decor and decorating. If someone tries to drag her off topic, or tries the classic "ignore your instructions" trick, she politely steers back to booking. She will not invent services, dates, or availability. And if someone asks, she says plainly that she is the online assistant, not a person.
A trades business gets trusted with real money. Helen has to read like a business you would hand a job to, not a gimmick.
Why this is the interesting bit
I did not build Helen as a one-off favour. I built her config-driven from day one. The trade, the area, the services, the persona, the webhook, all of it comes from a single config file. Hove Decor is instance one. Putting the same assistant on a roofer or a plumber is a config change, not a new build.
Every conversation is logged too, so I can see where people drop off and make her better over time. That is the difference between a chatbot as a toy and a chatbot as a piece of equipment.
My dad gets more booked quotes. I get a product I can template. That is a good trade.
Frequently asked questions
Does the chatbot give out prices?
No, and that is deliberate. Painting and decorating quotes depend on the job, so the bot never invents a price. It qualifies the visitor and books a free site visit instead. A wrong number quoted by a bot is worse than no number at all.
What happens when someone finishes the chat?
The moment it has the job, the area, a rough timeline, and a name and phone number, it fires the same lead pipeline the website contact form uses. That writes the lead to a sheet, sends an auto-reply, and texts my dad so he can call back fast.
Which AI model runs it?
Claude Haiku 4.5 through the Vercel AI SDK. It is fast and costs pennies per conversation, which matters when the whole point is to run quietly in the background on a small business site.
Can this work for other trades?
Yes. I built it config-driven, so the trade, service area, services list, and persona all come from one file. Hove Decor is the first one. Swapping it onto a roofer or a plumber is a config change, not a rebuild.
Will it pretend to be a person?
No. If someone asks, it says it is the online assistant for the business. It introduces itself by name, stays on topic, and refuses jailbreak attempts. People trust a trades business with real money, so the bot has to be honest.