blog

What can we learn from Nationwide?

Written by Adam | Sep 16, 2026, 11:39:54 AM

A story I've come across on multiple occasions in tech news this week focuses on a recent move by Nationwide. The Building Society has collapsed 19 data stores into one platform, retired a 45-year-old customer system and moved 15 million accounts off its old core banking engine.

Its Chief Data and Analytics Officer, Sri Kanisapakkam, put the reason plainly:

"AI is not magic. It needs to understand how your data is organised and how your data is structured. The data has to be trusted first before you can say your AI is trusted."

He's absolutely right and I'm delighted to come across that sentence in multiple publications this week. That said, I think the public sector could use a word of caution when taking onboard these wise words, given his advice is already being repackaged as the case for public bodies to do the same: one place, one technology, one platform.

Ultimately, Nationwide could take this approach because it is one institution. It owns all 19 stores, answers to one board, and works under one legal basis. Their approach here might seem appealing for public sector data too, but unfortunately, their 'one institution' context describes almost nobody in health, government or research.

An integrated care board, a national statistics office, a research consortium or a cross-departmental programme is a federation of data controllers with different obligations, different sensitivities and different suppliers. For them, "consolidate to one platform" isn't a viable strategy. It's a decade-long procurement fight followed by the lock-in problem the NHS is currently arguing about.

Look at what actually made Nationwide's AI trustworthy. Not the storage. The work of knowing what each of those 19 stores held, what the fields meant, who owned them and how they mapped onto each other. That layer — description, discovery, governance — is the part that travels. And it's the one part of the story that hardly ever gets mentioned because, sadly for us, its not particularly glamourous.

Trusted data does come before trusted AI. But for the public sector at least the trust lives in the metadata, not the platform. We can federate the description and leave the data where it belongs.