Most teams try to run a contact centre on a single number, usually average handle time. But a call center KPI set works as a whole: push one metric and another quietly breaks. Below is what each metric measures, how they pull against each other, and what changes once an AI worker joins the operation.
Accessibility: missed calls and first response time
Missed call rate shows how much demand was never met at all. This is not a technical number, it is lost business. A healthy overall rate can hide large gaps in specific hours, so look at the spread across hours and outside working time.
First response time is how long it takes for a customer to be met for the first time. On the phone it is measured in seconds, on WhatsApp and email in minutes or hours, so measure it per channel rather than merging everything into one average.
Together these two answer whether people can reach you at all. Everything else only matters after that.
Resolution: first contact resolution and repeat contact
First contact resolution (FCR) means the issue ends without the customer coming back on the same subject. It is among the most valuable metrics because it speaks to satisfaction and cost at once.
What to watch:
- If agents mark "resolved" themselves, the metric skews optimistic. Validate it by checking for repeat contact within a defined window.
- Do not lose customers who switch channels: a phone case that returns via WhatsApp the next day is a repeat contact.
- Break it down by topic. Overall FCR says little; which requests fail the first time says a lot.
Efficiency: handle time and after-call work
Average handle time (AHT) is an efficiency indicator, not a quality one. Targeted on its own, teams learn to shorten conversations rather than resolve problems.
After-call work (ACW) is missing from most reports but consumes a serious share of real capacity: notes, tagging and CRM records. This is usually where automation pays off fastest.
Escalation rate to a human shows where the automated flow falls short. Lowering it in isolation is dangerous: pushing customers into a corridor where they cannot reach a person damages satisfaction where your metrics cannot see it. The useful read is the distribution of escalation reasons.
Experience metrics and measurement traps
Satisfaction is usually measured with short post-conversation surveys and periodic loyalty questions. They do not measure the same thing: one describes the interaction, the other the relationship.
Points to watch:
- Survey respondents skew to the extremes, delighted and furious. Report the response rate alongside the score.
- Keep the question short and focused on one thing.
- If you never tag free-text answers and group them by topic, the most valuable data evaporates.
- Turn scores into a disciplinary tool and you stop measuring, you start managing the score.
How metrics break each other
Targeting one metric is the fastest route to improving it while degrading the system:
- AHT drops, FCR drops: conversations end early and the customer returns.
- Missed calls drop, quality drops: everyone is answered quickly but unprepared.
- Escalation drops, satisfaction drops: customers cannot reach a person.
- ACW drops, data quality drops: notes go unwritten and the next agent starts from zero.
So track metrics in pairs: put a quality metric beside every speed metric, and when setting a target, answer in writing which metric it might break.
What changes once an AI worker is live
Automation across voice and messaging changes what some metrics mean:
- Missed calls become a question of coverage, not capacity: which topics does the agent handle?
- AHT stops being comparable. Do not pool automated and human conversations into one average; track them separately.
- Escalation rate becomes the primary quality indicator. Classify the reasons: missing information, an action needing authority, an explicit customer request, or something not understood.
- Automated completion rate joins as a new metric: how many bookings, record updates and information requests finish end to end?
- ACW shifts significantly on the human side, since summarising and tagging can be automated.
- Channel mix moves; track volume shifting from phone to messaging as its own line.
When comparing periods, mark the go-live date on the report; before and after on one chart with no cut line is misleading.
Common reporting mistakes
- Looking only at averages. An average hides the bad experience; look at percentiles.
- Merging every channel into one table.
- Mixing out-of-hours traffic into working hours.
- Comparing rates without accounting for volume changes.
- Excluding abandoned contacts from the dataset.
- Reporting a metric without writing down its definition; "resolved" must mean one thing.
- Mistaking weekly noise for a trend.
When setting team targets, use a small set rather than one number: one accessibility metric, one resolution metric, one experience metric. Give targets to the team rather than to individuals. The right level varies from business to business depending on sector, request complexity and channel mix, so the healthiest baseline is your own last three months.
Checklist
- Every metric has a written definition and the team uses the same one.
- Metrics are reported per channel.
- A quality metric sits beside every speed metric.
- FCR is validated against repeat-contact data.
- Escalation reasons are classified.
- Percentiles are reported alongside averages.
- Working hours and out-of-hours are visible separately.
- Automated and human conversations sit in separate segments.
- Survey response rate is reported too.
- System changes are marked as dates on the report.
- Targets are set at team level and based on your own history.
- Someone reads the report each month and turns it into an action.
A good metric set exists to show where things get stuck, not to police the team. Pick a small set, fix the definitions, and read speed and quality side by side; once those are in place, it becomes obvious which improvements actually work. For channel-level reporting and how measurement is set up after automation, see /analytics-insights and /ai-call-center, and reach us via /contact to review your flows together. More at newads.ai