AI performance monitoring: see early, act early

The fastest way to make performance monitoring fail is to measure attendance: hours connected, keystrokes, time in meetings. It produces compliance behaviour within a fortnight and tells leadership nothing about whether the work is good.

What is worth measuring in professional services is narrower and harder to game: work delivered against what was committed, rework caused, and whether commitments made to clients were met on the date given. Those three describe performance in terms a consultant can recognise and argue with, which is what makes a review conversation productive.

Most performance problems in consulting firms are structural. A consultant who misses deadlines on three projects is often a consultant assigned to three projects at eighty percent each, and no amount of individual coaching fixes arithmetic.

Before a performance conversation it is worth checking allocation, the number of context switches per week, and how often their planned work was interrupted by unplanned client requests. If those explain the pattern, the conversation belongs with whoever plans the work, not with the person delivering it.

Monitoring that only points upward is surveillance; monitoring that a person can read about their own work is a tool. The practical difference is that in the second case the correction usually happens before a manager has to raise it.

The rule that keeps this credible is that individuals see their own data in full, teams see aggregates, and nobody browses individual detail about someone else without a stated reason. Writing that down and honouring it is what determines whether people trust the numbers or work around them.