Hola!
This issue is written from northern Spain where I was working from a co-working space inside a Santander bank branch, open not just to customers but to anyone who wants good coffee, a quiet desk, and solid Wi-Fi, all complimentary.
A bank is making its physical infrastructure work harder. Fitting, given what this issue is about.
But getting back to the question in title - the AI ROI in the financial services.
The headline numbers on AI in financial services look promising.
CEOs are optimistic. Investment is increasing. Announcements are being made.
But one figure sits underneath all of that, which does not make it into the press releases.
According to PwC's 2026 AI Performance Study, based on over a thousand senior executives across twenty-five sectors, 74% of AI's economic value is being captured by just 20% of organisations. Across banks, insurers, asset managers, and the consulting firms advising them, the majority are investing in AI and not seeing returns that survive a board-level conversation.
Most institutions can point to productivity gains somewhere in the business. The problem is translating those gains into outcomes anyone can point to on a P&L.
MIT's NANDA initiative found that 95% of enterprise AI initiatives fail to show measurable P&L impact. The era of productivity metrics as an AI business case is over. A presentation showing employees save four hours a week is no longer a business case. It is a description of activity. Boards and CFOs across regulated financial institutions are now asking a harder question: where is the financial return, and when does it arrive?
Most institutions do not have a good answer yet.
The gap is not where most people think it is
The instinct, when returns disappoint, is to look at the technology. Wrong model. Wrong vendor. Wrong use case. Not enough data.
That instinct is understandable. It is also wrong.
Microsoft's 2026 Work Trend Index surveyed twenty thousand AI users across ten countries and found that sixty-seven percent of AI's reported impact comes from organisational factors such as culture, manager support, workflow redesign, and how performance is measured. Thirty-two percent comes from the technology and the people using it.
PwC's 2026 Study referenced above found that the organisations capturing the most value are not simply deploying more AI tools. They are building strong organisational foundations before scaling. The top performers are nearly three times more likely than their peers to have those foundations in place.
The pattern across all studies is the same. The organisations getting returns built the infrastructure first. The ones stuck in pilot mode are still treating AI as a technology deployment rather than an organisational transformation.
What that infrastructure actually means
It is not a data lake or a centre of excellence or a responsible AI policy.
It is clarity on three things that most regulated firms have not yet resolved: who owns AI decisions, how those decisions are governed across the full lifecycle, and how accountability is maintained when a human process becomes an automated one.
Those are not technology questions. They are organisational design questions. And across banks, insurers, asset managers, and the firms advising them, they are going largely unanswered at the point in the process where they would actually change what gets built.
That is the execution gap. Not the model. Not the vendor. The architecture of ownership and accountability that should exist before deployment, in most institutions, does not.
In the next issue, I will look at what regulators are now finding when they go inside regulated institutions and examine this directly.
And now back to the coffee.
Hasta luego, as they say in Spain (Until soon).
Gaziza