French food, AI tools, and the problem with too much choice
The companies learning fastest about AI aren't the ones doing the most research
Ollie on Work is a weekly newsletter about what I’m learning from building with AI, advising leadership teams, and trying to bridge the gap between what technology can do and how businesses actually work. If someone forwarded this to you, you can subscribe here:
I was in Paris last weekend with my wife. We spent most of the time doing what the French do best - sitting around eating. Slowly, properly.
Three courses at lunch because why not. A coffee that lasted an hour because nobody was in a rush.
It reminded me of a study I read a few years ago by a researcher called Ashley Whillans. She and her colleague Romain Cadario surveyed over 10,000 people about how they eat, and found something interesting, but probably not surprising for anyone who’s into national stereotypes.
The French spend more time than anyone eating their meals. By which I mean, sitting down, tasting the food. Being present for it.
Americans, on the other hand, spend more time than anyone choosing what to eat - researching restaurants, scrolling through delivery apps, reading reviews, debating menus, comparing options - but much less actually eating it.
Same amount of time committed to the overall activity. Completely different allocation.
And guess what - the French reported significantly more satisfaction from the experience and less stress around it.
The Americans weren’t making better choices with all that research. They were just postponing the moment of commitment.
I realise I’m about to do the thing where someone connects a holiday anecdote to a business insight, and I apologise in advance, but… I’ve been thinking about that study a lot since I got back, because I keep seeing the same pattern in a completely different context.
The menu problem
A leadership team decides AI is important, so they start evaluating.
They read reports. They book demos. They run a comparison spreadsheet. They attend a webinar. They set up a working group to assess the options. Someone writes a proposal. The proposal needs sign-off. The sign-off requires another meeting. Someone raises a procurement question. Legal gets involved.
Three months later, they haven’t deployed a single tool into a single workflow. But they’ve spent an enormous amount of time and energy on the project, so it feels like progress.
This is the menu problem. All the time goes into choosing. None of it goes into eating.
Meanwhile, down the corridor, someone on the team has been quietly using ChatGPT or Claude every day for the last year. They’ve figured out which tasks it’s brilliant at and which ones it’s useless for. They’ve built little workflows that save them hours a week. They’ve made mistakes and learned from them.
They have more practical knowledge about AI than the entire working group - not because they’re smarter, but because they started.
Why we get stuck
Because there are a few reasons companies get trapped in evaluation mode. The problem is that most of them feel perfectly rational from the inside.
Fear of picking wrong. The AI landscape is moving so fast that whatever you choose today might be obsolete in six months. So you wait. The thing is, waiting doesn’t reduce the risk - it just guarantees you learn nothing while you’re waiting.
The search for the perfect use case. Leaders want to find the highest-impact application before they commit. This sounds sensible but it’s backwards. You don’t know where AI will have the most impact in your business until you’ve used it in a few places and seen what happens. The best use cases reveal themselves through use, not through analysis.
Consensus as a delay mechanism. Getting everyone aligned before moving forward feels responsible. In practice, it means the most cautious person in the room sets the pace for the whole organisation. The French don’t hold a committee meeting before ordering lunch.
The Deliveroo data
Sticking with the food theme, there’s a nice piece of research that backs this up from a different angle.
A team at the Max Planck Institute partnered with Deliveroo and analysed 1.6 million food orders from 195,000 customers across 197 cities.
They found what you’d expect: when people tried a new restaurant, the average rating was lower than when they stuck with an old favourite. 4.26 versus 4.52. Exploration has a short-term cost.
But over time, the more people explored - the more new restaurants they tried - the higher their average satisfaction climbed. Because they kept discovering better options and dropping the duds.
The people who stuck exclusively with what they knew plateaued. The ones who kept trying new things kept improving.
And interestingly, people were actually more attracted to restaurants with fewer reviews. Not less. Fewer reviews meant more uncertainty - and uncertainty meant there might be an undiscovered gem that everyone else had missed.
The lesson isn’t “be reckless.” It’s that the discomfort of trying something new is the price of finding something better. And the longer you spend reading reviews instead of ordering, the longer you wait to start climbing.
Just pick one and start
If your team has been evaluating AI tools for more than a month without committing to one, stop.
Pick the best available option. Not the perfect one - the best available one. Give it to a small team. Point it at one real workflow - not a sandbox, not a test environment - a real thing with real stakes.
You’ll learn more in the first week of actual use than you learned in three months of demos. Some of what you learn will be that it doesn’t work the way you expected. Good. That’s information you couldn’t have got from a comparison spreadsheet.
Order something. Eat it. Then decide if you want to order it again.
The ROI isn’t in the selection. It’s in the savouring.
See you next week.
Ollie