Those measures were not wrong. While the question was whether people were using AI they were the right things to count. They are the wrong measures if they are now read as evidence that the organisation is becoming more capable. They tell you how many people can open the tool, and nothing about whether anyone has become better at the work they are accountable for.
The Financial Services Skills Commission has made the point more precisely than the sector has absorbed it. In A Workforce Transformed it concludes that access to the technology itself is universal and not a differentiator. If that is right, and we believe it is, the sector has built its adoption narrative around the one variable that will not distinguish anybody from anybody.
The same report sets out five challenges. Three turn on a question almost no firm can answer: upskilling the workforce, redesigning routes in for early careers and closing the time to competence gap all require a firm to know whether its people are becoming more capable.
The Commission has already given firms a framework for identifying future skills and proficiency, but it is largely role agnostic. In an AI-enabled organisation, the capability required should reflect an individual’s role, accountability and the judgement they are expected to exercise.
The sector may need to recruit and train 450,000 highly skilled people over the next decade to replace those leaving or retiring, while AI is changing many of the tasks through which people historically became highly skilled.
The Commission has done the skills diagnosis, and done it well. This paper asks the question behind it. A skills gap concerns what people know and can often be addressed through training. A judgement gap concerns how experience and independent view are formed in the first place, and training alone does not close it.
Can we evidence that our people are becoming more capable as AI becomes more capable, and that the capability they develop is sufficient for the decisions, controls and accountability they carry? If AI performs more of the work through which professional judgement was traditionally developed, how will firms build the experience and judgement the next generation will need?
There is also a precedent worth remembering. Previous financial crises exposed what can happen when accountability outpaces the capability to understand and challenge the financial instruments, models and risk processes on which firms rely. AI risks creating a similar governance gap, but one that extends far beyond the Board.
We make four arguments.
Capability, not access, is the thing to measure, and almost no firm can evidence it: Firms need to know whether people are becoming more capable, where capability is developing unevenly and whether the work is becoming better, not simply faster.
AI is creating an Accountability–Capability Gap: The level of AI awareness and capability required is different for a Board member, accountable executive, control owner and practitioner. It should be proportionate to the accountability held; the consequence of the decision and the degree of reliance placed on AI. Senior leaders are being asked to govern and take accountability for a technology they may have limited understanding and practical experience of using.
AI can create capability debt by removing some of the experiences through which professional judgement develops: The productivity benefit is immediate, while the cost may surface later through longer time to competence, weaker independent challenge and a reduced pipeline of experienced judgement.
The destination is human elevation, not human preservation: The question is not how much work AI should do. It will do more, and then more again. The question is what humans must continue to learn, know, decide and own, and how firms deliberately develop and assure that capability.
WHAT SHOULD FIRMS DO
We propose four actions. Define accountable AI capability by role from Board level to practitioner and test whether it exists. Change what you measure, from access and activity towards capability, quality and distribution. Give AI the institutional and individual context that makes its output useful and ownable. And deliberately rebuild the learning loop as AI removes some of the work through which judgement was previously formed, assuring the human, the AI and the process together.