I've been deep in a rabbit hole this week.
Building tools and calculators in AI, experimenting with what's actually useful and what just feels clever.
It's far too easy to spend hours building something that looks impressive and adds nothing.
We switched from ChatGPT to Claude a little while back. I won't pretend I've done a proper head-to-head comparison, but the difference in how it handles complex work has been noticeable.
If you're thinking about how to measure whether your AI investment is actually working, there's one metric I'd point you to: revenue per employee. Total revenue divided by headcount.
In a service business it tells you, bluntly, how efficient you are. If you're sitting somewhere around £60k to £80k per person today, that number should start moving.
I'd expect businesses that genuinely embed AI to double it over the next couple of years. Not by cutting people. By letting them do more of the work that actually matters.
That's the opportunity. But here's the warning I'd attach to it.
This week I built a tool to convert a net bonus into a gross figure through payroll. Basic calculation. It was wrong. Not slightly off, structurally wrong. It took three versions to get to something I was happy with.
I only caught it because I know the numbers. Someone without that background would have taken the output at face value and run with it.
And that's the thing about AI right now. It is completely confident whether it is right or wrong. It doesn't hesitate. It doesn't caveat. It just gives you the answer.
In finance, that is genuinely dangerous.
Build, automate, experiment, yes. But if you're building anything involving numbers, check the maths yourself before you trust it with anyone else.
Until next week,
Luke