You’re in a loft with a torch in your teeth when the phone goes. It’s someone selling “AI for tradespeople”, and they promise it’ll run the office while you work. Maybe some of it would. But you’ve been in the trade long enough to know that anything promising to do everything usually does most of it badly. So here’s a straight answer, from people who sell one of these tools: what AI can do for a one-to-five van gas or plumbing business today, what it can’t, where it trips up, and how to test it before you let it near a customer.
What it can do today
Your trade is safe. Nothing is coming to change a diverter valve or find the leak under a floor. But running the business around the trade is mostly listening, writing, remembering and following rules you already know. That part, today’s tools do reasonably well:
Notice the label on that card. Every one of those works best as a draft or a summary that you glance at, not a decision made without you. A quote drafted in the van from your price list and approved with one tap is a real saving. A quote sent without you seeing it is a risk you’ve chosen to take, and you should only choose it for work you price the same way every time.
What it can’t do, and shouldn’t
It can’t diagnose a boiler. It can’t hear what you hear, smell what you smell or see the scorch mark. Over the phone it can take a fault code and a description and pass them on. Anything that tells a caller what’s wrong, or what’s safe to do, is overstepping.
It can’t sign a gas safety record. HSE’s guidance says a landlord’s gas safety record must include the name, registration number and signature of the person who did the check. Software can fill in the landlord, address, tenant and date, and type up readings you give it. The inspection, the findings and the signature are yours, every time. See why you sign the paperwork.
It can’t replace your judgement. Whether a landlord is worth chasing or worth dropping. Whether that customer’s “it’s probably nothing” means it’s nothing. Whether to squeeze in the pensioner on Friday at six. You make a hundred of these a week without noticing. A tool can follow your rules. It can’t make new ones well.
A good tool does the typing and asks you about the thinking. A bad one does both and tells you afterwards.
Where it goes wrong
These tools fail in a few predictable ways. Knowing them is most of the protection.
It sounds sure when it shouldn’t be. Language tools write confidently whether they’re right or wrong. Ask one a price you haven’t given it and a badly set up one will give a plausible number. That’s the most dangerous habit, because it sounds exactly like a right answer.
It mishears the details. Postcodes, house numbers, surnames, a boiler model read out over a bad signal. Worth checking the address on any booking before you set off, at least in the first few weeks.
It picks the wrong record. Two customers called Patel on the same street, or a landlord with eight properties. A careful tool asks which one. A careless one guesses.
It says “done” when it isn’t. The invoice was drafted but not sent. The booking was written but didn’t save to the calendar. If a tool can’t show you that the other system confirmed it, treat “done” as “probably”.
It promises on your behalf. “Someone will be with you within the hour” is a lovely thing to tell a cold customer, and a terrible thing to say if it isn’t true. What gets said to a customer about your service can become part of the deal under consumer law, so the tool should only say what you’ve told it to.
How to test any tool before you trust it
Don’t judge it on the demo. Demos are rehearsed. Judge it on your own awkward Tuesday. Before you let any tool talk to customers, ours included, do this:
Then ask the supplier four plain questions. What exactly will it send to customers without asking me? Where are my customers’ details and call recordings kept, and what are callers told? How do I get my data out if I leave? What does it do when a connection fails? Vague answers are an answer.
Start small. Let it draft for a fortnight while you send everything yourself. Then let it handle one routine thing, like next-day visit reminders, under a written rule. Add the next thing only when you’ve stopped checking the last one. See write your rules down first and why trust is a dial.
Where we stand
Since we’ve asked you to be sceptical, here’s Datify held to the same standard.
Datify answers business calls, takes the details and sends you a summary on WhatsApp. It books visits only when they fit the hours, area and rules you set and your calendar is free, and asks you otherwise. It drafts quotes, invoices and job paperwork, including the admin side of a gas safety record, for you to check, complete and sign. It never signs or issues a certificate, whatever rules are set. By default it asks you before anything goes to a customer, anything touches money, or any gas safety record moves past draft. The exceptions are only the ones you’ve set as rules. It doesn’t tell you something is done until the connected system has confirmed it, and if something fails, it says so.
What it doesn’t do yet matters as much. We’re working with a first group of trade businesses and set each one up by hand. Call handling is set up business by business and only covers your number after a real test call. It connects to Google Calendar, Xero and ServiceM8 today. The ServiceM8 connection and sending PDFs over WhatsApp are still being tested in the pilot, and gas paperwork forms are set up with each business. If you use other software, it doesn’t connect to it yet.
It’s a 14-day free trial, then £37 a month. Run every test on this page against it. If it fails one, tell us, and don’t switch that part on. You can read exactly what needs your approval, or book a demo and bring your worst Tuesday.
AI is useful for the business around the trade: answering and logging calls, drafting quotes, invoices and paperwork, reminders, chasing and daily summaries, all with you checking. It can’t diagnose, sign a gas safety record or replace your judgement. It goes wrong by sounding sure, mishearing, picking the wrong record, claiming “done” too early and promising for you. Test any tool on your own awkward calls, start with drafts only, and hand over one routine job at a time.
Writes about what actually happens when you put an AI in charge of real business operations — including the parts that go wrong.