If you run a company with a large customer base — a bank, an insurer, a telco, a utility, an energy retailer — you already own the most valuable dataset in your industry. Every call your contact centre has ever taken. Every email a relationship manager has sent. Every chatbot and voicebot session. Every question a customer has typed into an AI feature in your app. Every complaint, every hesitation, every “I might also be interested in…”.

You own all of it. And until very recently, you could not read it.

What we did instead

Not for lack of trying. I spent years inside one of Europe’s largest banking groups, responsible for parts of its digital transformation, and I can tell you exactly what a large company does when it wants to know what its customers actually experience. It commissions research. It runs focus groups. It pays external agencies. It pulls a sample of call transcripts and has consultants read them. It looks at dashboards that show what customers did — clicks, churn, product holdings — and tries to infer from that what they said.

All of this was, and is, an elaborate way of guessing. Expensive, slow, and built on samples. The decision usually got made anyway, on a fraction of the evidence, because the quarter would not wait.

What changed

AI agents can now read every single interaction. Not a sample. All of it. You put every bit of communication between your company and your customers onto one pile — transcripts, emails, chat logs, voicebot sessions, in-app AI conversations — and you pose a business question in plain language.

The agents work through the entire pile. They find the relevant conversations, count them, group them, quote them, cross-check them, and come back with an answer and the evidence behind it. If you want, they hand you a board-ready report. It behaves like an analytical strike team — a consulting project that starts the moment you finish typing and finishes before your next meeting.

Why this is now an obligation, not an option

Consider the decisions a C-level team makes in a normal year. Launch or kill a product. Reprice a segment. Redesign an onboarding flow. Approve a multi-million investment in technology “to serve clients better”. Open a new business line to upsell an existing base.

Each of these decisions is now made in one of two ways: with one hundred percent of the evidence, or with a guess dressed up as research. There is no longer a respectable middle ground, because the first option exists and is cheap.

The competitive logic is uncomfortable but simple. Strategic decisions compound. A company that makes each of them on complete evidence, and a company that makes each of them on a slice, do not end up a few percentage points apart. They end up travelling in different directions. You cannot win simply by having this capability — soon every serious competitor will. But you will lose without it, one slightly worse decision at a time.

Make it a ritual

The practical implication is a new habit at the top of the organisation. Before any product decision, ask the pile. Before any pricing decision, ask the pile. Before signing off any investment meant to improve customer service or customer value, ask the pile — and ask it what customers are actually struggling with, not what the vendor’s slide deck assumes.

Because asking is cheap, you can afford to ask questions that would never have justified a research budget. The follow-up question. The second-order question. The question that sounds naïve. That is where most of the value hides, and it has been hiding there for years.

One condition

This pile is also the most sensitive data your company holds — every conversation a customer has ever had with you. The way to do this responsibly is to run it on infrastructure you control, with open-weight models, so the data never leaves your perimeter and the cost of a question is compute time you already own rather than a per-token bill someone else sets. That is a deliberate architectural choice, and it is the subject of the third article in this series.

Your customers have already told you what they think, what they want and what would make them buy more. They told your people, your bots and your app. The only thing that has changed is that you can finally listen — all of it, at once, one prompt away.