A skills audit is a structured review of the skills people in a team or organisation actually have, measured against the skills their roles require. Before or during an AI agent deployment, it answers a precise question: who can safely review, approve or build for the agents we are running, and who cannot yet?
Why does a skills audit matter?
Organisations usually assume the people who did the work by hand can oversee the agent that now does it. That assumption is untested. A skills audit tests it, and often finds the gap is not where managers expect. Experienced staff may know the process well but not understand why an agent holds an action, or that a document's source class decides whether it may support a payment.
It also gives evidence. When a supervisor or internal auditor asks how you know your reviewers are competent, "we trained them" is weaker than "we audited their skills against the role and here are the results".
How is a skills audit run?
A practical audit for agent oversight follows five steps:
- Define the roles: reviewer, team lead, builder, owner, executive sponsor.
- List the skills per role: for a reviewer, reading verdicts, judging sources, checking limits, knowing red lines, escalating.
- Gather evidence: tests set in the team's own cases, observed reviews, the agent's record of past approvals and send-backs.
- Rate each person: against a simple scale, by skill, not overall.
- Record the result: in a skills matrix, with the date and the evidence used.
The third step is what separates an audit from a survey. Asking people to rate themselves measures confidence. Testing them, and looking at what they actually approved, measures skill.
An example
An internal audit function is asked to review an agent that drafts expense approvals. Before auditing the agent, the head of audit runs a skills audit on her own team. Three auditors can read the agent's record and trace a verdict to the bound that decided it. Two cannot. She pairs them for the engagement and books training. The guides on training an internal audit team to review agent decisions and the reviewer onboarding checklist for internal audit cover what came next.
How is a skills audit different from auditing the agent?
A skills audit looks at people. An audit of the agent looks at the system: its certificates, its bounds, its record, whether held actions were approved by the right people. The two meet at one point. An audit of the agent may find that approvals were given by people whose skills were never checked, and that finding leads straight to a skills audit. For the system side, see how to prepare agents for an audit.
A skills audit also differs from a skills matrix. The audit is the exercise; the matrix is where its results live.
Which terms are related?
- Skills matrix: the grid of people against skills.
- Skills gap: the difference the audit reveals.
- Skills passport: a portable record of a person's verified skills; see what a skills passport is.
For the full approach, read training teams for AI agents.
Questions people ask
How often should a skills audit be repeated?
Before go-live, before raising an agent's depth, and at least yearly. People move roles and agents gain capabilities, so an old audit soon describes a different team.
Who should run the audit?
Someone independent of the team being audited, using tests and records rather than opinion. The team lead should contribute but not mark their own team.