Support Analytics

Support analytics

Support metrics drive behaviour hard, which makes the wrong metric actively harmful: measuring tickets closed per hour produces fast, bad answers and repeat contacts. The useful metrics describe whether the customer's problem went away.

Method

1. Measure resolution, not closure. A ticket closed that generates a follow-up two days later was not resolved, and repeat contact rate exposes it.
2. Track time to first meaningful response. An automated acknowledgement is not a response, and measuring it flatters the number while the customer waits.
3. Watch the distribution, not the average. A good average with a long tail means some customers wait days, and they are the ones who leave.
4. Attribute volume to cause. Which product areas generate tickets is the number that reduces future volume, rather than describing current load (see customer-feedback-loop).
5. Measure deflection honestly. Self-serve success means the customer solved it, not that they gave up and did not write in.
6. Interpret satisfaction scores cautiously. They measure the interaction more than the outcome, and a polite refusal can score well while the customer churns.
7. Review the metrics for perverse incentives quarterly. Any metric agents are judged on will be optimised, including in ways nobody intended.

Boundaries

Metrics describe support; they cannot fix a product that generates the tickets. Satisfaction surveys have severe response bias. Agent-level metrics need care, since they affect people's livelihoods and reward gaming (see agent-people-ops-desk).