Work out the question first.

I spent this morning at Vena’s “How AI is Influencing Finance” event at County Hall, and the first session was from Dr Nicola Millard - futurologist and author. Her talk was called “Thinking outside the bots: staying human in an AI world”, and it was one of the better AI sessions I’ve sat through this year (and I’ve sat through a few!).

She opened with The Hitchhiker’s Guide to the Galaxy. The great computer spends seven and a half million years working out that the answer to the ultimate question of life, the universe and everything is 42 - at which point everyone realises they don’t actually know what the question is.

Which is a pretty fair description of a lot of AI selling right now. The answer is AI. What was the question again?

When MIT Technology Review asked what stops people embracing AI in their roles, the top answer wasn’t fear of losing their job (40%). It was lack of clarity about value, at 60%, with lack of training next at 55%. People aren’t so much scared of it as unsure what it’s for.

The slide that’s stuck with me, though, was a study from METR on experienced software developers. Before the study, they expected AI tools to make them roughly a quarter faster. Afterwards, they estimated it had made them about 20% faster. When the researchers actually measured it, they’d been 19% slower.

Now, developers aren’t finance teams, and writing code isn’t month-end. But I’d be surprised if finance were immune. AI takes away the blank page, and the blank page is the bit that feels slow. What it adds - checking, re-prompting, tidying up, working out why the third paragraph has invented a number - doesn’t feel like work in the same way, so you don’t count it. If you’re going to tell the CFO that AI has saved the team a day a month, measure it first, properly.

Then came what Nicola called our lazy brains. We anchor on the first answer we see, so once the AI has drafted something, the reviewer edits it rather than questioning it. We get lazy about checking our own thinking. We make judgement calls without the experience that used to come from doing the groundwork. And we offload - why remember it, or work it out, when the machine will? She had a lovely line from Daniel Kahneman for it: thinking is to humans as swimming is to cats - they can do it, but they’d prefer not to.

None of this is a reason not to use AI. I’ve written before that accountants aren’t trained to try things twice, and that the value comes from keeping going. But keeping going with a tool is different from nodding along with it. She argued for designing in verification rather than blind acceptance - which, if you’ve read my post on validation, is a drum I’ll happily bang.

Her next point was one I hadn’t really thought about. If AI takes all the simple, repetitive work, what’s left is the complex, the ambiguous and the emotional - all day, every day. She called it the intensification of work. The simple stuff was never just simple stuff; it was also the breather between the hard bits. Anyone who’s done a pile of grunt work with a podcast on will know what I mean.

Which brings me to her key message: be in the lead, not in the loop. “Human in the loop” sounds reassuring, but it can easily mean a person whose job is to click approve on whatever the machine produces. In the lead means people decide what the AI is for, and where it isn’t wanted.

She made that point with airlines. A pilot is what she called a proficiency role - you need them to be very good, but past a certain standard, a better pilot doesn’t add much more value. A flight attendant is a pivotal role - the gap between a good one and a great one is huge, and it’s exactly what passengers remember. So the two get AI in different ways. The pilot gets autopilot, with the pilot watching it and ready to take over - the human is in the loop. The flight attendant gets a button that plays the in-flight announcements, so they can spend their time on the passengers - the AI is in the loop, handling the routine so the person can do the part that makes the difference.

There was a chart from Harvard Business Review comparing an augmentation path with an automation path over time. Automation looks better at first - you’ve cut cost, job done. Augmentation dips first, while people learn, then pulls ahead and keeps compounding, through wellbeing, keeping and attracting talent, and eventually the leadership pipeline. Automation drifts down.

She had a nice example of the augmentation path in practice. IKEA built a bot that dealt with 47% of customer contacts. Rather than trying to get rid of the other 53%, they looked at what those customers wanted, found a lot of them were after design advice, retrained the advisers as interior designers, and built a €1 billion business out of it.

I like the IKEA one a lot. Most teams would count the 47% and call it a win, but the real money was in the 53%.

My notes from the session finish with the tortoise and the hare. AI is the hare - very fast, and quite capable of running confidently in the wrong direction.

So by all means use it. Just work out the question first!