Most support dashboards show more numbers than anyone acts on. Handling time drops and nobody knows whether customers got what they needed. Satisfaction dips and the team debates whether the survey sample was too small. The numbers are accurate. They just aren't connected to a decision.
A useful service metric tells your team what happened and what to improve. That means defining each measure precisely, comparing like-for-like requests and tying every result to a specific change with an owner.
In this blog, we'll group metrics by the question they answer, define eight core customer service metrics with formulas and data sources, work through two calculations, show how to read results by channel and request type, and share a weekly review template you can reuse.
Key takeaways
Define the question before the metric
Before you choose a metric, decide what you're trying to learn. Most service questions fall into four groups.
Group | Question it answers | Suitable measures |
|---|---|---|
Response speed | How quickly do customers get a first reply? | First response time |
Resolution quality | Are requests resolved, and do they stay resolved? | Resolution time, first-contact resolution, repeat contact after resolution |
Employee workload | Where is our team's time going? | Average handling time, transfer rate, automated task completion |
Customer experience | How do customers feel about the result? | CSAT, survey comments, sentiment in conversations |
Eight customer service metrics with clear definitions
These eight metrics cover most of what a support leader needs. For each one, agree the formula, the events that start and stop the clock, where the data comes from and who owns it.
Metric | How to calculate it | Data source | Owner |
|---|---|---|---|
First response time | First response timestamp minus request received timestamp. Report automated and human first responses separately. | Help desk, chat and conversation logs | Team lead |
Resolution time | Resolution timestamp minus request received timestamp, shown as elapsed time. Report reopened cases separately. | Help desk | Team lead |
First-contact resolution | Requests resolved on first contact with no repeat contact for the same issue within seven days ÷ eligible requests with a complete seven-day window × 100 | Help desk, phone system and conversation logs | Support operations |
Repeat contact after resolution | Resolved requests followed by another contact about the same issue within seven days of resolution ÷ resolved requests with a complete seven-day window × 100. Count each original request once. | Help desk and CRM | Support operations |
Transfer rate | Requests transferred to a person ÷ requests entering the automated flow × 100 | AI conversation logs and phone system | Automation owner |
Automated task completion | Eligible requests reaching the defined completion event ÷ eligible requests entering that workflow × 100 | AI conversation logs and the system where the task completes | Automation owner |
CSAT | Satisfied responses ÷ valid survey responses × 100. State the scale and which scores count as satisfied. | Survey tool | Customer experience lead |
Average handling time | For a defined phone cohort: (talk time + hold time + after-call work) ÷ handled calls | Phone system or contact centre platform | Workforce manager |
Two rules keep these numbers honest. Match repeat contacts by case or request, using the same issue-matching rule every time, so one customer with two unrelated problems isn't counted as a repeat. And wait for the full seven-day window to close before reporting first-contact resolution or repeat contact, or this week's requests will look better than they are.
If you're measuring automated customer service workflows, report answered, completed and transferred requests as separate outcomes rather than one blended rate.
Work through an example
Here's how two of these metrics come together with real counts.
Worked example (illustrative)
Automated task completion. Of 100 eligible requests that entered an automated workflow, 72 reached the defined completion event. Automated task completion = 72 ÷ 100 × 100 = 72%. Review transferred and failed requests separately.
CSAT. The survey uses a five-point scale, and scores of 4 and 5 count as satisfied. Of 40 valid responses, 32 were satisfied. CSAT = 32 ÷ 40 × 100 = 80%, based on 40 responses.
Two things to notice. The 28 requests that didn't complete need splitting into those handed to a person on purpose and those that failed, because each needs a different fix. And CSAT should always travel with its response count: 80% from 40 responses and 80% from 4,000 responses support very different decisions.
Read results by channel and request
A single average rarely tells you where to act. Break results down before you draw conclusions.
Turn a result into a change
A metric matters only when someone acts on it. Here are three common results, what they usually point to and what to change.
Result | What to check | Change | Owner | Review again when |
|---|---|---|---|---|
Knowledge gaps rising on one topic | The questions the AI or agents couldn't answer, grouped by topic | Revise the knowledge source and review the answers it produces | Knowledge owner | The topic has a full week of new conversations |
Failed transfers | Which queue the transfer reached and what context arrived with it | Fix the queue rule or the handoff summary | Telephony or automation owner | A full week of transfers has run on the updated rule |
Repeat contact after resolution | Whether the task actually completed and whether the customer got a confirmation | Fix the completion step or add a confirmation message | Support operations | A seven-day window has closed on requests made after the change |
The first row is often the quickest win. When you identify knowledge gaps by topic, you know which source to fix before the same question comes back.

Build a weekly service review
A short, consistent weekly review is more useful than a long monthly report. Use the same layout every week so trends are easy to spot, and fill one row per request type and channel.
Request type | Channel | Sample | Metric and value | Failure cause | Change | Owner | Follow-up result |
|---|---|---|---|---|---|---|---|
Billing questions | Chat, English | 412 requests | Automated task completion 72% | No answer for prorated charges | Add the prorated-charge policy to knowledge | Knowledge owner | Check next week |
Billing questions | Chat, French | 96 requests | Automated task completion 61% | French billing article out of date | Update the French version of the article | Knowledge owner | Check next week |
For the AI side of the review, the Expertise Analytics Dashboard shows conversation counts and trends by day, week or month, positive, neutral and negative sentiment, knowledge gaps grouped by topic, and the custom filters your team sets up for themes it tracks. You can export the data for your own reporting.
How Expertise helps you measure customer service
Expertise is built so each conversation leaves a result you can review. Support Chat, AI IVR and Voice Agents hand off with the transcript and a summary, Agent Assist supports the employee who takes over, and knowledge gap analysis lists the questions your AI couldn't answer so you know which source to fix first.
Expertise is proudly Canadian. English and Canadian French are built in, Canadian data residency is available alongside US and European hosting, and the platform is SOC 2 Type II audited and HIPAA, GDPR and CCPA compliant. See our enterprise security and deployment details.
Teams using Expertise see measured results like these:
Explore service analytics or book a demo.
Frequently asked questions
Which customer service metrics should a team start with?
Begin with response time, resolution, repeat contact and satisfaction for a defined request type. Add workload and transfer measures when they help explain a service problem.
How do you measure automated task completion?
Divide eligible automated requests reaching the defined completion event by all eligible requests entering that workflow. Record failures and human transfers separately.
Can faster handling mean better service?
Faster handling is useful when the customer's issue is resolved. Review it alongside repeat contact, resolution and satisfaction to understand the outcome.
Should English and French requests be reported separately?
Yes. Report each language on its own so a smaller French queue isn't hidden inside a strong English average, and compare them using the same definitions and timing events.

