"Our Data Isn't Ready for AI"

I hear it at least twice a month. A director at a brokerage, somewhere halfway through an introductory meeting, says: "Sounds great, but our data isn't ready for AI." And then the conversation stops. Not formally, but in practice it does. An internal project is launched to "get the data in order first." That project takes a year. Sometimes two. And then nobody calls back.

I understand the logic. It feels responsible. Foundation first, then the building. But it is a fallacy.

Perfect data does not exist

We now work with dozens of brokerages. Not one of them had "perfect data" when we started. What they did have: client records in a broker management system, policy data from insurers, and some claims history. Messy? Sometimes. Incomplete? Often. Unusable? Almost never.

We built our first model for a client on data that the IT manager himself called "a disaster." Three months later, that model predicted with 78% accuracy which clients would cancel within six months. Not perfect. But considerably better than the existing method, which was: nothing.

The real problem lies elsewhere

Let me be honest: the data excuse is often a proxy for something else. Uncertainty about what AI will concretely deliver. Fear that advisers will not use it. Or simply no owner who says: we are doing this.

That is understandable. Most AI pilots in the industry do not fail because of technology. They fail because there is no clear business question underneath. "We want to do something with AI" is not a business question. "We want to know which of our 3,000 clients are at risk of cancelling in the coming months" -- that is one.

Start small. Start imperfectly.

The advice I give to every leadership team that hesitates: start. Not with a large transformation programme. Not with a data warehouse project. Start with a question you already have and data you already have.

Our approach is deliberately modular. No big bang, no complete IT replacement. We connect to whatever is already in place -- ANVA, CCS, MSS, it does not matter -- and build a predictive layer on top. Within weeks, not months.

The biggest risk is not that your data is imperfect. The biggest risk is that your competitor has already started while you are still tidying up.

On adoption

There is one more thing I want to address. The best AI in the world is worthless if advisers do not use it. I have seen enough dashboards that were opened exactly zero times after launch.

That is why we do not build dashboards. We build triggers. An adviser opens their system in the morning and sees: these five clients deserve attention today, for this reason, with this conversation point. That is not extra work. That is better work.

The technology is the easy part. The hard part is behaviour change. And that does not start with a perfect data model. It starts with one adviser calling a client they would otherwise have forgotten.