Fixing a tedious expenses process with AI

Our month-end expenses process was manual, slow and error-prone – a necessary job that everyone dreaded. It was eating up time that nobody had to spare and leaving finance to untangle the mess at month end.

So we practised what we preach and put technology to work. Our team built a bespoke, AI-assisted expenses tool that handles the heavy lifting end to end. Submissions are faster, data is cleaner, and everyone gets back to the work they're good at.

The problem in a bit more detail

It wasn't one big problem. It was a dozen small annoying ones, strung together.

• Manual data entry. Consultants photographed receipts, typed values into the ERP, and matched each transaction to a card line.
• Context chasing. Working out the right project and cost centre meant trawling Outlook calendars, recalling which customer was visited that day, and picking from around 100 cost categories. Not exactly a good use of anyone's time.
• A knock-on effect. Claims came in late, queries bounced back and forth, and month-end processing was harder than it needed to be.

Why an add-on, not an overhaul

AI wasn't the only way to solve this, and it wasn't our first assumption either. There are off-the-shelf systems out there built specifically for expenses management, and for some businesses that's the right call. But our ERP system does its job well; the expenses process was just one area where it fell short. It's a familiar position for most businesses: large, well-established systems that work fine day to day but aren't perfect for every task. Replacing it, or introducing an expensive third-party system, wasn't right for us, so instead, we built a small, targeted add-on that solved our specific problem.

What we built

An AI-assisted expenses tool that automates the complex steps while keeping humans in control.

• Scans receipts and extracts the relevant details automatically.
• Matches each receipt to the correct card transaction.
• Uses calendar entries, client meetings, and active ERP projects to infer the right project and cost centre, with a confidence score.
• Presents a pre-filled claim to consultants for an accuracy check, and any manual corrections, then submits to the ERP and notifies finance once approved.

Here's a good example of it in action. A "Train to Kings Cross" receipt gets matched to a calendar entry for an onsite client visit in London, then cross-checked against an active ERP project for that client. Right project, right cost code, no manual effort required.

A bespoke web application wraps the AI to provide an intuitive interface, secure integrations with the ERP and Outlook, and an audit trail that finance can rely on.

What changed

It might have been a small project – but what an impact it’s had.

• Consultants cut expense admin from hours per month to minutes.
• Finance receive cleaner data with far fewer queries bouncing back.
• Faster submissions mean timely month-end processing.
• Nobody has to build a claim from scratch. They just check, correct if needed, and submit.

What's next

Building the tool solved the immediate problem, but it also opened up possibilities we didn't have before. We're looking at policy nudges to flag out-of-policy spend, auto-splitting complex receipts across multiple cost centres, and spend analytics that could shape how we think about supplier negotiations and travel policies. Mobile-first capture with real-time validation is on the roadmap too.

Have you got a process that's driving your team up the wall? Or a system that doesn’t quite work like it should? Get in touch and let's talk about what we could do with it. Info@waterstons.com