Most AI products are built for people who already understand AI. That's not a market — that's a mirror.
I spent the last few months building an AI-powered business operating system for UK tradespeople. Plumbers, electricians, roofers, gas engineers. People who fix real things with real hands and have approximately zero patience for onboarding flows.
It broke every assumption I had about building software.
The Problem Nobody's Solving
Here's a day in the life of a self-employed plumber:
- Wake up. Check WhatsApp. There are 14 messages from customers, 3 of which are "are you still coming Tuesday?"
- Drive to first job. Quote something on the back of a receipt. Lose the receipt.
- Second job runs long. Forget to invoice the first one.
- Customer from last week calls asking for a VAT receipt. What's a VAT receipt? (He's not VAT registered. She doesn't know that. Neither does he, technically.)
- End of day: exhausted, behind on paperwork, dreading the tax return that's due in four months.
This person doesn't need a "dashboard." They need someone to handle the admin while they handle the pipes.
Why WhatsApp?
The first thing we got right: meet them where they already are.
Every tradesperson in the UK lives on WhatsApp. Their customers message them there. Their suppliers message them there. Their mates send memes there. Building a native app and expecting them to switch is fantasy.
So we built the entire system as a WhatsApp conversation. Quote a job? Send a WhatsApp message. Generate an invoice? WhatsApp. Check your schedule? WhatsApp. Track expenses? Take a photo of the receipt and send it on WhatsApp.
No app to download. No login to remember. No UI to learn.
The AI handles natural language understanding, figures out what they're trying to do, and does it. "Quote Mrs Patterson £450 for the boiler service, parts included" becomes a formatted quote sent to the customer's email. Done.
What Tradespeople Actually Need
We assumed wrong about almost everything at first. Here's what we learned:
They don't want CRM. They want to not forget people. The word "CRM" makes their eyes glaze over. But "remind me to check in with Dave about that leak in January" — that they understand. Same feature, completely different framing.
Quoting speed is everything. A tradesperson who sends a quote within an hour of visiting a property closes at 3x the rate of one who takes a week. Our fastest-growing feature is instant quote generation from a voice note description of the job.
Tax is terrifying. Most sole traders in the UK are running on vibes and a carrier bag full of receipts. Building in automated expense tracking and tax estimates made grown adults visibly relieved. One user said "I might actually sleep in January for once."
They want to look professional but can't design. Branded PDF quotes with their logo and terms? They love it. They were previously sending quotes as text messages. Their customers now think they hired an office manager.
The Technical Bits
For the builders in the audience:
- WhatsApp Cloud API for messaging. Meta's API is adequate. Not great, not terrible. The webhook reliability is better than it was.
- Intent classification for understanding what the user wants. Not an LLM call for every message — too slow, too expensive. A lightweight classifier for routing, LLM for the heavy thinking.
- Prisma + PostgreSQL for the data layer. Boring. Reliable. Correct choice.
- Next.js dashboard for the occasional desktop user who wants to see their numbers. Most never log in.
The AI layer is thinner than you'd expect. The hard work is in the business logic — UK tax rules, VAT thresholds, CIS deductions, Making Tax Digital compliance. The LLM is the interface. The domain knowledge is the product.
The Uncomfortable Lessons
1. AI is the wrong selling point.
We stopped saying "AI-powered" in any customer-facing material. Tradespeople don't care about AI. They care about getting paid faster and doing less paperwork. The AI is invisible, as it should be.
2. Voice notes are the killer input method.
Typing on a phone in a van with plaster on your hands is miserable. Voice notes are natural, fast, and surprisingly information-dense. Our best user interactions start with a 30-second voice note that contains everything we need to generate a quote, create a job record, and update the schedule.
3. Trust takes forever, switching takes seconds.
One bad experience — a wrong invoice, a missed reminder — and they're gone. These are people who've been burned by software before (looking at you, every field service management tool with a £40/month subscription and a three-month minimum). Trust is earned in millimetres.
4. The business model is simpler than I expected.
No freemium. No complicated tiers. Flat monthly fee, cancel anytime. Tradespeople understand subscriptions (they pay for van insurance, tool insurance, gas safe registration). They don't understand "seats" or "API calls" or "premium features."
Where It's Going
We're not building a tool. We're building a business partner that lives in their pocket. The roadmap isn't "add more features" — it's "understand more of what they say and handle more of what they hate."
Receipt scanning. Automatic follow-ups with customers who haven't responded to quotes. Integration with accounting software their accountant already uses. Seasonal reminders to existing customers ("Your boiler service is due — shall I book you in?").
None of this is technically revolutionary. It's all been done before, in various forms, for various audiences. The difference is doing it through a channel they already use, in language they already speak, without making them feel like they need a computer science degree.
The Point
The best AI products in 2026 won't be the most technically impressive. They'll be the ones built for people who don't know what RAG stands for and never will. People who just want their invoices sent, their taxes estimated, and their Tuesdays to stop disappearing into admin.
Build for the plumber, not the programmer.
