Why Median Beats Average for Support Metrics
When reporting on first reply time or resolution time, you have to choose: average or median? This post explains why median is usually the better choice for support metrics.
The problem with averages
Averages are sensitive to outliers. One ticket that took 72 hours to resolve can skew your “average resolution time” by hours, even if 99 other tickets resolved in 4 hours.
Example:
| Tickets | Resolution time |
|---|---|
| 99 tickets | 4 hours |
| 1 ticket | 72 hours |
| Average | 4.68 hours |
| Median | 4 hours |
The average suggests your typical ticket takes 4.68 hours. The median shows that most tickets take 4 hours—and you have one outlier. Which number tells you more?
What median shows
Median is the middle value when all tickets are sorted. Half of tickets are faster; half are slower. It represents the “typical” experience better than the average.
For customer-facing metrics, median often matches reality:
- “Half of our tickets get a reply in under 2 hours” (median FRT = 2 hours)
- “Our average FRT is 4 hours” (because of a few slow outliers)
Customers care about what they’re likely to experience, not what happens in rare edge cases.
When average makes sense
Averages aren’t always wrong. Use average when:
- You’re measuring total load — Total handle time for capacity planning.
- You want outliers to matter — If every slow ticket is a problem, average shows the impact.
- You’re comparing cost — Cost-per-ticket is typically an average.
But for SLA-style metrics (FRT, resolution), median is more representative.
Percentiles: The next level
If median is good, percentiles are better. Percentiles show distribution:
| Percentile | What it means |
|---|---|
| p50 (median) | 50% of tickets are faster |
| p75 | 75% of tickets are faster; 25% are slower |
| p90 | 90% of tickets are faster; 10% are slower |
| p99 | 99% of tickets are faster |
Tracking p90 shows how your slowest tickets perform—without a single outlier destroying the metric.
Example target: “90% of tickets get a first reply in under 4 hours (p90 FRT < 4 hours).”
How to choose
| Use case | Recommendation |
|---|---|
| SLA targets | Median or p90 |
| Customer communication | Median |
| Internal ops | Median (easier to interpret) |
| Capacity planning | Average (total load matters) |
| Outlier detection | Compare median vs average (big gap = outliers) |
Tip: Compare both
If your average FRT is 6 hours and your median is 2 hours, you have a long tail of slow tickets. That gap is a diagnostic signal—it tells you to investigate outliers, not to report the average as your target.
For more on metrics, see support KPIs and first reply time in Zendesk.
FAQ
Do SLAs use average or median?
Most SLAs are phrased as “X% of tickets within Y hours”—that’s a percentile, not an average. Align your internal metrics with how your SLA is defined.
Should I report both?
You can, but for simplicity, pick one (median) and explain it. Adding average confuses most audiences.
What if my tool only shows average?
Consider switching tools or exporting data to calculate median. Averages can mislead.