TLDR Tech
Your Data Moat Matters More Than Your Model
The race to deploy AI in financial services has mostly been a race to adopt the same tools. GPT wrappers, third-party fraud models, bought-in credit scoring engines. Everyone's running on similar infrastructure, and the differentiation has been thin.
What this piece gets right is that the real advantage was accumulating quietly in the background the whole time. Transaction history. Behavioural patterns. The messy, proprietary signal buried in years of customer activity. Revolut and Stripe aren't winning because they picked a better foundation model. They're winning because they have training data that nobody else can buy.
For UK consumer credit brokers and lenders, this is a genuinely uncomfortable read. We have data too, but most of us have fragmented it badly. Origination systems that don't talk to servicing platforms. Affordability checks that live in spreadsheets. Fraud signals that never fed back into underwriting. We built point solutions for point problems, and now we're sitting on a pile of disconnected signals rather than a coherent dataset.
The shift from hundreds of specialist models to a single foundation model trained on unified financial behaviour is the part worth paying attention to. The firms doing this are collapsing the wall between fraud, credit risk, and product performance into one learning system. That's not just an efficiency gain. It changes what questions you can ask.
The FCA's increasing focus on affordability and persistent debt makes this more urgent, not less. A model that sees the full behavioural picture, across channels and over time, is a better tool for identifying genuine financial stress than a point-in-time credit score ever was.
The question for anyone running a lending or broking operation in the UK: do you actually know what data you're sitting on, and is it in a shape where it could do that kind of work?
- fintech
- underwriting
- AI