The KPI That Drove the Wrong Behaviour
A software support team is measured on average ticket resolution time. To hit the target, agents have begun closing difficult tickets quickly and asking customers to reopen them if the issue persists, which resets the clock. Average resolution time has fallen 30% quarter on quarter - leadership is delighted - but reopened-ticket volume has doubled, first-contact resolution has dropped, and a customer survey shows falling satisfaction. The support manager wants to keep the current KPI because 'the numbers look great'.
1. Explain what is happening to the metric and why it no longer measures what leadership thinks.
Read the worked answer
Agents are gaming the metric: closing and reopening resets the resolution clock, so measured time falls without any real improvement in how fast problems are solved. The KPI now measures 'time to a first close', not 'time to resolution'. Because the clock resets, the hardest tickets - the ones that matter most - become invisible in the average. This is Goodhart's law in action: once a measure becomes a target, it stops being a good measure. Falling resolution time alongside rising reopenings and falling satisfaction is the tell.
2. Propose a replacement measurement set and the decision rule leadership should adopt.
Read the worked answer
Replace the single average with a small balanced set: (1) time to true resolution measured from first contact to final close, counting reopenings; (2) first-contact resolution rate; (3) reopen rate; (4) customer satisfaction. Report them together so no single number can be optimised in isolation. The decision rule: a KPI is only 'improving' if the customer-outcome measures (satisfaction, FCR) hold or rise; a fall in resolution time accompanied by rising reopenings is treated as a red flag, not a win. Include a median and a tail measure (e.g., 90th percentile) so hard tickets aren't hidden.
3. Why is an average alone a poor summary here, and what should be shown instead?
Read the worked answer
An average hides the distribution: a few very long, complex tickets can be masked, and closing-and-reopening shifts mass out of the long tail artificially. The analyst should show the distribution - median plus 90th percentile, and a breakdown by ticket complexity or category - so leadership can see whether the easy tickets got faster while the hard ones were pushed out of the count. A single average invites exactly the behaviour observed.