Where AI actually saves money: the boring work, not the chatbot

When a leader asks me where AI is going to save them money, they're usually picturing the chatbot. Something customer-facing, something with a demo. I get it — that's the version that makes the news. It's also, in my experience, rarely where the money is.

The money is in the boring work. The repetitive, well-defined, high-volume tasks that currently soak up expensive people's time and never make it into a press release.

Reset the expectation

Take a system migration — moving off an old accounting platform, say, or consolidating two acquisitions onto one ERP. The headline is the new software. The real cost is everything underneath: writing the scripts that transform old data into the new format, and testing that the whole thing still works once it's moved.

That work is repetitive and tightly scoped, which happens to be exactly what AI is good at. On the migration work we've put through this approach, the time to produce those transformation scripts dropped by roughly 60 to 67 percent. Not because the AI did anything clever, but because it took the first pass at work that's mostly pattern-matching, and a person corrected it from there.

Two workflows that pay

• Workflow 1

  AI-drafted transformation scripts

  Moving data between systems means writing rules: this field maps to that one, these values get reformatted, those records get merged. AI writes the first draft of those rules fast. An engineer reviews and adjusts. The work that used to take weeks takes days.

• Workflow 2

  UI-driven automated testing

  Testing a system by clicking through it manually is slow and tedious, so it usually gets shortchanged. AI paired with browser automation can drive the interface and check behavior across hundreds of cases — the kind of coverage no one budgets the hours for by hand.

time saved on test automation in controlled studies, climbing with well-scoped tasks

25–36%

TTC Global, 2025

more tasks completed by developers using AI assistance, across 4,867 engineers

26%

Microsoft Research, 2025

saved per user in a UK public-sector trial — about 28 working days a year

56 min/day

UK GDS, 2025

Why the audit trail survives

Here's the part that makes finance and compliance comfortable, and it's worth saying plainly: in these workflows, AI writes the code, but the code still runs as ordinary, deterministic code. It does the same thing every time. You can read it, test it, and keep it.

That's a very different risk profile from handing a live financial decision to a model and hoping it behaves. A human reviews the output, the logic is inspectable, and the audit trail stays intact. You get the speed of AI on the drafting without giving up the accountability that regulated work demands.

How to spot the work worth automating

You don't need a strategy deck to find these opportunities. You need three filters:

• Repetitive — the same shape of task done over and over.
• Rules-based — there's a right answer and you can describe how to get it.
• Currently expensive — skilled people are spending real hours on it today.

Skip the magic. The return on AI is hiding in the work you've stopped noticing because it's always been done by hand.