By Mo Hussein

2 min read

Start with the paperwork

When people ask me about AI, they usually picture a chatbot on the website. In a service business, that's rarely where the time goes. The time goes on reading things and typing them somewhere else: emails, PDFs, photos of forms, notes scribbled after a site visit.

That's what today's AI models are genuinely good at. They read messy, unstructured information and turn it into something tidy. That's a far less glamorous job than a chatbot, and far more useful.

Five jobs worth handing over

  • Sorting incoming enquiries and drafting a reply for a person to check and send.
  • Pulling the details out of supplier invoices and delivery notes, ready for approval.
  • Turning call notes or site-visit notes into a proper job record.
  • Writing first drafts of quotes from your standard price list, for someone to adjust.
  • Summarising a long customer email thread before a colleague picks it up.

Three to keep human

Notice that every job above ends with a person checking the result. These three should stay with people entirely:

  • Anything that commits you to money: the final price, a refund, a credit.
  • Anything you say to an upset customer.
  • Decisions about people, such as hiring, performance or pay.

Ask where your data goes

Before you switch anything on, find out which company runs the model, whether it keeps a copy of what you send, and whether it trains on your data. You should get a clear answer to all three. On our own website, the estimate assistant only uses model providers that keep no copy of your messages and don't train on them, and we hold the tools we build for clients to the same standard.

Measure it honestly

Pick one job and run it for a month. Count the time it saves, then take off the time spent checking its work and the running cost. If what's left is worth having, do the next job. If not, you've lost a month, not a year.

If you'd like a second opinion on which job to start with, that's a good use of a discovery call.