A vendor scorecard can be green every week while your customers get steadily unhappier. That happens when the metrics measure effort instead of outcomes. These seven, read together, are hard to fake.
1. First contact resolution
The share of contacts fully resolved without a follow-up. This is the single best proxy for whether agents are empowered and trained. Measure it from the customer's side — no new contact on the same issue within 72 hours — not from an agent-selected disposition code.
2. Repeat contact rate
The mirror of the metric above, and the one that exposes fast closures. A queue with excellent handle time and a rising repeat rate is not efficient; it is deferring work to next week.
3. CSAT with comment coverage
A score without comments tells you nothing actionable. Require a minimum comment volume in your reporting and read a sample yourself every week. Patterns show up in language long before they show up in averages.
4. Quality assurance score against a shared rubric
QA only means something when you and the vendor calibrate on the same rubric and score a shared sample. Ask for the rubric, the sample rate, and a monthly calibration session where you both grade the same ten interactions.
5. Response and resolution time by percentile
Averages hide your worst experiences. Track the 90th percentile too: an average first response of four minutes with a 90th percentile of nine hours means a slice of your customers is being abandoned.
6. Agent attrition and tenure on your account
This is the leading indicator everyone forgets. Quality follows tenure, so a pod losing people quarterly will never reach the depth you were promised. Put named-team stability and a backfill retraining obligation in the contract.
7. Escalation rate and escalation aging
How often work leaves the pod, and how long it sits once it does. A rising escalation rate points to a knowledge gap; aging escalations point to a broken handoff between your team and the vendor's. Both are fixable once visible.
Metrics to treat with suspicion
Occupancy, tickets per hour and average handle time are capacity-planning inputs, not quality measures. They are trivially gamed and should never carry a bonus on their own. Pair every efficiency metric with an outcome metric in the same SLA table.
Key takeaways
- Read first contact resolution and repeat contact rate together.
- Calibrate QA on a shared rubric or the score is decorative.
- Track 90th-percentile response times, not just averages.
- Tenure on your account predicts quality better than any weekly score.
Frequently asked questions
How many KPIs belong in an SLA?
Five to seven. Beyond that nobody optimises anything, and the vendor manages the report rather than the work.
How often should we review performance?
A weekly operational review on trends and a monthly business review on root causes and quality calibration.
Want this handled for you?
Our team builds support, marketing and payment operations for growing businesses. Book a free consultation and we'll map the fastest path for yours.