How Do We Recognise Employees Applying AI Well?

Most financial services firms have now given staff access to some form of AI tool, whether for drafting client communications, summarising documents or supporting analysis.

Far fewer firms have worked out how to tell when someone is using that tool well.

It is tempting to assume that the most frequent or confident users are the strongest performers. In practice, the opposite can be true. An employee who challenges an AI output, spots an error or decides not to use AI at all for a particular task may be demonstrating far better judgement than a colleague who simply accepts whatever the tool produces.

Recognising good AI use matters because it shapes behaviour. If firms reward speed and volume, they will get more of it, often at the expense of accuracy and accountability. If they recognise verification, scepticism and sound judgement, they build a workforce that uses AI safely and effectively.

Why Recognising Good AI Use Is Harder Than It Looks

AI tools present their outputs with the same confident tone whether they are right or wrong.

A drafted client summary, a suggested policy wording or a generated risk note all read fluently regardless of whether the underlying content is accurate. This makes it difficult for a manager glancing at finished work to tell how much genuine checking and thinking went into it.

Two employees might produce near-identical outputs. One may have accepted the AI's first draft without question. The other may have queried the tool, checked it against source documents and corrected several errors before finalising the same piece of work. From the finished product alone, a manager cannot easily tell them apart.

This is the core difficulty: good AI use is largely invisible unless someone specifically looks for it. Without a deliberate approach to recognition, firms are left assessing outputs rather than the judgement that produced them.

How Firms Have Traditionally Recognised Good Practice

Financial services firms have long-established methods for recognising skilled practice that has nothing to do with AI.

Quality audits sample completed work against defined standards. Peer review has partners, underwriters or senior handlers check each other's output before it goes to a client or into a system of record. Appraisal criteria set out the behaviours expected at each level of seniority. Spot checks by team leaders catch errors before they compound.

These methods share a common feature: they focus on process and behaviour, not just the finished item. A quality audit does not simply check whether a document is correct. It checks whether the person followed appropriate steps to make sure it was correct.

This same principle applies directly to AI-assisted work. The question is not only whether the output was right, but whether the employee took reasonable steps to establish that it was right, and knew what to do if it was not.

Where AI Can Help Surface Recognition Signals

AI-assisted analysis can help managers see patterns that would otherwise take considerable manual effort to identify.

For example, version comparison tools can highlight where an employee has made substantial edits to an AI-generated draft, suggesting active review rather than passive acceptance. Workflow data might show which staff routinely flag AI outputs for a second opinion, or which consistently escalate uncertain cases rather than proceeding regardless.

These signals are useful starting points for a conversation, not a verdict. A high edit rate could indicate careful checking, or it could indicate the AI tool is poorly suited to that task. A low edit rate could mean the AI got it right, or it could mean the employee did not look closely enough.

Interpreting what these patterns actually mean, and deciding whether they reflect good judgement, remains a decision for the manager. AI can narrow down where to look. It cannot decide what counts as good practice.

Building AI Judgement Into Existing Recognition Processes

Most firms do not need a new recognition scheme. They need to extend the criteria already used in appraisals, one-to-ones and team reviews to explicitly cover AI-related behaviour.

Practical steps include:

• Adding a small number of AI-specific behavioural criteria to existing appraisal templates, such as "identifies and corrects AI errors" or "applies AI outputs appropriately to the task at hand".
• Asking employees, as part of regular one-to-ones, to describe a recent example where they checked or challenged an AI output.
• Using team reviews to share specific examples of good AI judgement, rather than general reminders to "use AI responsibly".
• Making the criteria visible in advance, so staff understand what good use looks like before they are assessed against it.

The aim is to make AI judgement a normal, visible part of how competence is already discussed and rewarded, rather than creating a separate and burdensome process that sits alongside existing appraisal work.

Example

A mid-sized London insurance broker introduces an AI tool to help account handlers draft renewal summaries for commercial clients. Two handlers use the tool at similar rates, but one consistently flags where the AI has misread policy wording or omitted a material change, while the other passes drafts through largely unchecked. Their team leader is reviewing quarterly performance and needs to identify which behaviour reflects genuinely good AI use.

The team leader introduces a simple review step where handlers note any corrections made to AI-drafted summaries. The handler who consistently catches and corrects errors is recognised in the review as demonstrating strong AI judgement, and this practice is shared with the wider team as an example of good use, rather than treating tool usage volume as a proxy for competence.

FAQs

• Is using AI more often a sign of good AI literacy?

  Not on its own. Frequent use can reflect genuine efficiency, but it can equally reflect over-reliance on a tool without proper checking. What matters is the quality of judgement applied around that use, such as whether errors are caught and whether the employee recognises when AI is not the right tool for the task.

• How can a manager assess AI use if they don't understand the tool themselves?

  Managers need a working level of AI literacy to recognise good judgement when they see it. Without this, they risk rewarding confident-looking output rather than sound practice. Building manager AI literacy alongside frontline training is an important first step.

• Should recognising good AI use be a formal part of appraisals?

  It can be incorporated into existing appraisal criteria rather than requiring a wholly new scheme. What matters most is that the criteria are clear and shared with staff in advance, so recognition feels consistent rather than arbitrary.