What an AI sales transformation actually runs on

Every management team we speak to is being pitched an AI transformation. The demos are impressive, the tools are cheap, and the pressure to be seen doing something is real.

Here is what we have learned from doing this work inside PE-backed industrial and B2B businesses. The AI is rarely the hard part. The hard part is underneath it, in the segmentation and the data.

Industrial equipment, 125 agents across 50 markets. Before any model could score customer potential, the records had to be cleaned and enriched market by market, because half the fields the scoring needed were missing or wrong. Once that was done, the potential-based segmentation redirected agent effort to must-win accounts and produced seven million euros of additional sales in six months. The insight came from the basics. Any AI model needs accurate, clean data to work from.

German precision parts, entering Medtech self-injection. A strong technical reputation, hundreds of reachable decision makers, and no system to work them. Account intelligence lived in slide decks and individual heads. Rather than hire an account manager and wait a year for impact, we built the system an account manager would need. Twelve weeks later there were 250 named contacts in the CRM, 15 of the 16 relevant platform makers engaged, and a weekly steering view ranking accounts by potential and probability. Targeted outreach had already converted into a new customer win with a multi-year sales tail.

None of the layers in that stack is novel. What is new is the price. Capability that used to need a data team and a board approval now costs a small monthly subscription, which means a mid-cap industrial business can afford to run its segmentation continuously rather than once.

That is the quiet shift worth acting on. Segmentation used to be a project. A strategy phase produced a map, the map went into a drawer, and eighteen months later it described a market that no longer existed. With current tools the map can rebuild itself as the data moves. The exercise stops being a one-off and becomes part of how the commercial team runs the week.

There is a step after that, and it is where most of the money goes missing. A map that is current still pays nothing until it reaches the person who has to act on it.

Most segmentation work stops at the map. It produces an accurate picture, everyone agrees the picture is accurate, and then it sits there. What a seller actually needs is a routed opportunity. A named account, someone who owns it, a reason to call this quarter, and something to open with.

Four things go wrong on the way there, and none of them is a modelling problem. The analysis lands as a deck nobody has a reason to open. Nobody decides who calls whom, so the list stays a suggestion. The seller gets a name but no reason, so the call never gets made. And one out-of-date list is enough for a team to stop opening the map at all.

So the fix is routing. Every priority account gets an owner, a reason to call, and a weekly review that is honest about what has not been touched. Start with one product line or one region rather than a complete map nobody acts on.

Before funding any AI initiative in the commercial function, spend a week auditing the two basics it will stand on. Score your customer data for completeness on the fields that drive potential, and check the date on your last segmentation. In our experience that week tells you more about the likely return than any vendor demo will.

Both engagements in full: the agent segmentation across 125 agents and the AI growth engine in Medtech.

This piece is the third issue of the Hirondl newsletter, one idea a month from inside the businesses we work in. Subscribe here.