In 2006, British mathematician Clive Humby said five words that would shape a decade of technology strategy: "Data is the new oil."1 The analogy had genuine insight. Like oil, raw data needs refining before it yields value. Like oil, whoever controlled the largest reserves seemed to win. For a while, that was true enough.
I encountered this argument from the inside. In 2016 I joined RGA in Hong Kong as Head of Strategic Innovation for Asia-Pacific, and a year later I wrote about what data actually means for insurers from inside one of the world's largest life reinsurers. My argument then was simple: data without the ability to analyse it, without the right people to act on it, is worth very little. Storage and computing were commoditising. The real costs were people and software. The real constraint was judgment.
Eight years on, I stand by that. But the oil metaphor has not aged well, and not for the reasons most people cite.
The standard critique is that data, unlike oil, is not depleted when used. True, but not the most important flaw. The more damaging problem is what the metaphor implied about strategy: that the resource is the value. Drill, refine, distribute — and you win.
It encouraged boards to count data lakes as strategic assets. CDOs were hired to "unlock the value of data." Billions were invested in storage and processing infrastructure on the assumption that the data itself was worth something. And it led, consistently, to programmes that stalled not because the ambition was wrong but because no one had mapped what was actually there before announcing what would be done with it.
In 2024, the British Computer Society proposed uranium yellowcake as a better analogy: data as a raw material with significant potential energy requiring careful handling, capable of enriching or destroying depending on how it is processed.2 More sophisticated. Still a material metaphor. And swapping one material metaphor for another does not resolve the underlying error.
Here is a pattern I have encountered more than once, in more than one institution. A transformation programme is underway. The first question seems simple: how much does each application cost to run? Headcount, software licences, infrastructure, maintenance.
The answer is that no one knows. Not approximately, precisely. Not at all. Invoices are not itemised. Costs are not coded to the system or the function they support. The organisation is paying for things it cannot name, maintaining things it cannot price, and making investment decisions about technology on a foundation of complete financial opacity.
Ask the technology function for the group's technical architecture. What comes back is partial, a view of some systems, some markets, some layers. Not a complete picture. Not a group-wide agreed architecture. An assemblage of partial maps that together do not constitute a map.
Ask for the data strategy. There is a document. It describes ambitions. It does not describe decisions: which data, governed by whom, accessible to whom, under which regulatory constraints, integrated at which layer of the stack.
And ask for the cost model underpinning the offshore capability hubs (the so-called low-cost locations). The assumption is that they are low-cost. The actual cost structure, what resources genuinely cost once you account for management overhead, quality control, capability limitations, turnover, and the hidden costs of distance, has never been honestly modelled. The headcount arbitrage has been assumed and sold upward. It has not been verified.
This is not a story about one organisation. It describes a pattern across financial services. And it matters because every AI programme built on top of this foundation is not transforming the organisation. It is digitising its opacity.
Those who argue AI will be "the new electricity" are reaching for the right analogy but drawing the wrong lesson from it. Thomas Edison's first commercial power station opened in Manhattan in 1882. By 1935, fifty years later, only 10% of American farms had electricity, compared with 90% in Germany and Japan.3 The electricity was available. The infrastructure to deliver it was not.
AI in financial services is in the same position. The capability exists. The organisational architecture to absorb it (clean data pipelines, governed access, integrated systems, honest cost models, capable people in the right roles) largely does not. Announcing an AI strategy while the data architecture underneath it remains unmapped is not ambition. It is expensive wishful thinking.
Not: how much data do you have. Not: which AI platform should you buy. But: does your current architecture actually support the questions your strategy requires. Have you ever honestly found out?
Most organisations have not. The oil is in the ground. No one is entirely sure where. And the drill has never been properly tested.