ai reads. a person decides.

Every evening, someone in a trade business sits down with a pile of supplier invoices and types them into the accounts. Line by line: supplier, parts, quantities, VAT, totals. It’s slow, it’s boring, and it’s exactly the kind of job AI is good at.

So when we built gaffer, the office platform now running at MW Plumbing & Heating, invoice reading was one of the first things we automated. Invoices arrive by email to the office inbox, or as a photo from the staff app on site. The AI reads each one in seconds and pulls out the supplier, the line items, the VAT and the totals.

And then it stops.

the golden rule

Nothing the AI reads goes into the accounts on its own. Every invoice lands in a review queue. Someone in the office checks the read, fixes anything it got wrong and picks a category for each line. When they approve it, the items go into stock and any price rises get flagged. It only goes to KashFlow when a person taps send.

We call this the golden rule, and it’s written into the code: the AI does the reading, a person makes the call.

why not let it run?

Because the cost of being wrong isn’t the same on both sides. A missed invoice is annoying. A wrong figure pushed silently into the accounts is a problem you find at year end. Keeping a person on the final step costs a few seconds per invoice and removes that risk entirely.

It also builds trust. Owners who’ve never used AI before are happy to let it do the typing when they can see every result before it counts.

what it changes

The office still owns the numbers. It just doesn’t spend its evenings typing them any more. That’s the kind of AI we like building: useful, boring in the right way, and never out of sight.