What is a capability audit?

A capability audit tests whether an organisation still possesses the human knowledge, judgement and practical ability its operations depend on. The distinguishing move is that it removes the assistance and looks, rather than asking people how capable they feel.

One thing belongs at the top rather than in a caveat at the bottom. This is a method proposed by this research, drawing on proficiency testing, business continuity and the organisational capability literature. It is not a standard, no study validates it, and nobody should present its output as a measurement. What follows is a way of asking, offered because the questions are not being asked at all.

Why the usual instruments fail

Two proxies are in general use and both are broken in the same direction.

Confidence fails because the faculty that would report the loss is the one impaired. Sankaranarayanan's unrestricted group failed at 77 per cent against 39 for the scaffolded group once the tool was gone, and nobody in the weaker group knew they were in it. Graded entry. A survey asking whether AI has affected capability is using the broken instrument to measure the breakage.

Output fails because assisted output is the thing that stays high. That is the finding, not a confound: capability can fall while production rises, and a dashboard tracking production will show improvement throughout.

Which leaves one option. Remove the assistance on a real task and see what happens, which is as uncomfortable as it sounds.

What the audit asks

Four questions, in order of how much discomfort they cause.

What can people still do unaided? On a real task, at real difficulty, without the tool. Not a quiz about the tool.

Where does the expertise actually reside? Usually in fewer people than the process map implies. Nelson and Winter's test applies: a capability that cannot survive the loss of any single individual was never organisational. An Evolutionary Theory of Economic Change.

Which capabilities have no redundancy? Not which are important, which have exactly one path.

Which would be hard to rebuild? Relearning is generally faster than learning, which is the good news. The evidence is here. It has never been tested on professional judgement, which is the bad news, and rebuilding a routine that several people used to run together is a different problem from an individual relearning a skill.

How it differs from an adoption review

An adoption review measures use: seats, prompts, hours saved, satisfaction scores. A capability audit measures what remains without the tool.

The two can point in opposite directions, and an organisation scoring well on adoption while failing a capability audit is the case the method exists to find. Nothing in an adoption programme would surface it, because adoption programmes are designed to demonstrate that adoption happened.

Argyris and Schon explain the persistence. Organisations correct errors inside their existing assumptions and rarely question the assumptions, because organisational defences prevent it. Organizational Learning (1978). A capability audit is a double-loop question asked of a single-loop programme. That makes it unwelcome rather than merely unusual.

Minimum viable human capability

The natural follow-on question is how much is enough. What can be given is a shape rather than a number.

Enough to recognise when an automated system is failing, to keep critical operations within acceptable limits while it is unavailable, to decide the cases outside the system's competence, and to recover control when necessary.

The level depends on consequence, recoverability and acceptable downtime, which is business continuity reasoning rather than anything specific to AI. There is no universal percentage, and any figure presented as one should be treated as invented until its derivation is shown.

Weick and Sutcliffe's high-reliability principles describe the organisational conditions this depends on, particularly deference to expertise rather than to rank. Managing the Unexpected. Capability that exists in a person without the authority to use it does not count for this purpose.

What this page does not establish

The method is a synthesis, not a finding. No trial shows that organisations running such an audit perform better, and none has been run.

Its components have separate support of varying strength. That confidence and output are poor proxies rests on the illusion-of-competence literature and on two AI studies that removed the tool afterwards. That capability lives in routines rests on theoretical work. That relearning beats learning rests on skill-decay research never conducted on professional judgement. Those are three different evidential standards inside one method.

Removing assistance is also not free. It costs time, it produces worse output on the day, and in some settings it would be unsafe. Any organisation applying this has to decide where an unaided test is appropriate, and that decision is not one this research can make for them.

Key sources

• Sankaranarayanan, S. et al. (2026). Mitigating "Epistemic Debt" in Generative AI-Scaffolded Novice Programming using Metacognitive Scripts. Graded entry.
• Nelson, R. R. and Winter, S. G. (1982). An Evolutionary Theory of Economic Change. Belknap Press. In the essential works.
• Weick, K. E. and Sutcliffe, K. M. (2001). Managing the Unexpected. Jossey-Bass. In the essential works.
• Argyris, C. and Schon, D. A. (1978). Organizational Learning: A Theory of Action Perspective. Addison-Wesley. In the essential works.

Related SuperSkills research

On the measurement problem, assessing capability rather than output and the illusion of competence. On what is being audited, organisational capability. On what accumulates without it, capability debt. On adoption metrics that measure the wrong thing, measuring AI adoption properly and usage theatre. On recovery, can you regain a skill you have lost.

About this research

Rahim Hirji is the author of SuperSkills (Kogan Page, 2026), keynote speaker on AI and human capability, and founder of The SuperSkills Intelligence Company. The capability audit is a method proposed by this research and is not an established construct, a standard, or a validated instrument. The literature it draws on is cited above so that the proposal can be judged against it.

How this research works  ·  Reviewed quarterly  ·  Found an error? Tell me and it is corrected on the page.