Ten years ago I was working in one of the largest banking groups in Europe, responsible for parts of its digital transformation. Every significant decision we faced — whether to launch a new offering, whether to lower the price of a product for a particular segment, whether to rebuild a customer journey — started with the same question. What do our customers actually experience with us?

And every time, the same machinery started up.

The machinery

We ran joint projects with Big Four consultancies, working in collaborative mode. We collected feedback notes and customer messages and had people read through them. We pulled samples of call transcripts. We sat and listened to recorded conversations between operators and clients. We asked colleagues in the branches and the contact centre what they were hearing. We organised focus groups. We paid external agencies for research on top of all that.

Then we tried to compile everything into one coherent insight that could answer the original question well enough for a decision.

It was a priority, so it got resourced accordingly: hundreds of consultant man-days, dozens of my own, and a real budget. The bank could afford it. But looking back, the cost that mattered most was not the money.

The real cost

It was time. A proper answer took months. By the time it arrived, the market had moved and half the question with it.

It was sampling. However many transcripts we read, they were a fraction of a fraction of what customers had told us. We were extrapolating from a keyhole.

And it was selection. Because each question was so expensive to answer, we only asked the big ones. The hundreds of smaller, sharper questions — the follow-ups, the “wait, why did they say that?” — never got asked at all. Curiosity had a price, and the price was too high to indulge.

I want to be fair to everyone involved: the people were excellent and the method was the best available. It was the tools that were inadequate.

The same question, today

Take the same question — should we lower the price of this product for this segment? — and ask it now. Every call transcript, every email between a representative and a client, every voicebot and chatbot session, every conversation customers have had with AI features in the app: all of it sits on one pile. AI agents read the entire pile. Not a summary of a sample; the whole thing.

They come back with what customers have actually said about that product, that price and that segment, how often, in what tone, and with the evidence attached. If you ask for it, you get a structured report you could take to the board. What used to be a project is now a query.

What changes for a manager

The obvious change is speed. The deeper change is that curiosity becomes affordable. You ask the second question. You ask the third. You ask the question that sounds naïve in a steering committee and turns out to be the one that matters. You ask at four in the afternoon and read the answer with your morning coffee.

That is not a productivity gain. It is a different way of running a company: every decision touching the customer made on complete evidence, as a matter of routine.

The catch, and the way through

There is one thing the old way had going for it: the cost was predictable. A consulting project had a price. An AI agent that reads everything and keeps digging does not — the same question can cost a euro or half a million in tokens, and you cannot know in advance how valuable the answer will be. That tension is real, and it is why we believe this belongs on infrastructure you own, running open-weight models, where the cost of thoroughness is compute time rather than a meter. I take that up in the next article.

I wish I had had this ten years ago. The leaders who have it today will make decisions I could only have guessed at — and they will make them before lunch.