Customer service metrics are the measurements a support team uses to track speed, quality, customer sentiment, workload, and cost. Common examples include first response time, first contact resolution, CSAT, NPS, and customer effort score. Measuring them well depends on consistent definitions, the right data source for each metric, and enough responses to trust the result.
What are customer service metrics?
Customer service metrics are quantifiable measures of how a support operation performs, drawn from the systems the team already uses: the helpdesk, the phone system, survey tools, and the logs of any AI agent handling conversations. They answer practical questions about how fast customers hear back, how often a problem is fixed the first time, and how customers feel afterward.
Most metrics fall into two families. Operational metrics (O-data) describe what happened inside your systems: response times, resolution rates, volume, and handle time. Experience metrics (X-data) describe how customers perceived it: satisfaction, effort, and loyalty, usually captured by survey. Operational data shows where the process broke. Experience data shows whether the customer noticed. Read both together.
One distinction matters first. A metric is anything you can measure. A KPI is a metric you have tied to a goal, with a target and an owner. This page is the glossary: every metric, its formula, and how to measure it reliably. For choosing which metrics to elevate, setting targets, and examples by role, read the guide to customer service KPIs.
Why customer service metrics matter
They turn service quality into something you can manage. Without numbers, "good support" is an opinion. With consistent metrics you can see which channels, topics, and hours produce the worst experiences and fix those first.
They connect support to retention. Experience metrics like CSAT, CES, and NPS move before churn does, giving an earlier warning than renewal rates can.
They justify investment. In the 2026 Customer Service Transformation Report, 87% of teams at a mature stage of AI deployment reported improved metrics since adopting AI, compared with 62% of teams overall. That gap is only visible to teams that measured before and after.
They keep commitments honest. Service level agreements are defined in metrics. You cannot know whether you are keeping your promises without measuring SLA compliance consistently.
20 customer service metrics to measure
The table below is the full glossary. Each metric is then defined in its group with its formula, data source, and most common measurement pitfall.
| # | Metric | Category | Formula | Data source | Measured how often |
|---|---|---|---|---|---|
| 1 | First response time (FRT) | Speed | Sum of first response times ÷ conversations | Helpdesk | Daily |
| 2 | Average resolution time (ART) | Speed | Sum of resolution times ÷ resolved conversations | Helpdesk | Weekly |
| 3 | Average handle time (AHT) | Speed | (Talk + hold + wrap-up time) ÷ contacts | Phone system, helpdesk | Daily |
| 4 | Average wait time | Speed | Sum of wait times ÷ answered contacts | Phone system, chat queue | Daily, by hour |
| 5 | SLA compliance rate | Speed | Within-SLA conversations ÷ conversations with an SLA × 100 | Helpdesk | Daily |
| 6 | First contact resolution (FCR) | Resolution and quality | First-contact resolutions ÷ total issues × 100 | Helpdesk | Weekly |
| 7 | Resolution rate | Resolution and quality | Resolved ÷ total conversations × 100 | Helpdesk, AI agent logs | Weekly |
| 8 | Reopen rate | Resolution and quality | Reopened ÷ resolved conversations × 100 | Helpdesk | Weekly |
| 9 | Escalation / transfer rate | Resolution and quality | Escalated ÷ total conversations × 100 | Helpdesk, phone system | Weekly |
| 10 | CSAT | Customer experience | Positive ratings ÷ total ratings × 100 | Post-conversation survey | Continuous |
| 11 | NPS | Customer experience | % promoters (9–10) minus % detractors (0–6) | Relationship survey | Quarterly |
| 12 | Customer effort score (CES) | Customer experience | Sum of effort ratings ÷ responses | Post-conversation survey | Continuous |
| 13 | Ticket volume | Volume and workload | Count of new conversations | Helpdesk | Daily |
| 14 | Backlog (open tickets) | Volume and workload | Count of open conversations at period end | Helpdesk | Daily snapshot |
| 15 | Contact rate | Volume and workload | Contacts ÷ active customers × 100 | Helpdesk plus CRM or orders | Monthly |
| 16 | Self-service rate | Self-service and AI | Sessions without a conversation ÷ help center sessions × 100 | Help center analytics | Monthly |
| 17 | AI resolution rate | Self-service and AI | AI-resolved ÷ AI-handled conversations × 100 | AI agent logs | Weekly |
| 18 | Involvement / handoff rate | Self-service and AI | Handoffs ÷ AI-handled conversations × 100 | AI agent logs, helpdesk | Weekly |
| 19 | Cost per resolution | Cost and efficiency | Total support cost ÷ resolved conversations | Finance, helpdesk | Monthly |
| 20 | Agent utilization (occupancy) | Cost and efficiency | Handle time ÷ available time × 100 | Workforce management, phone system | Weekly |
Speed metrics
Speed metrics are the operational measures contact centers have used for decades, so a search for "call center metrics" turns up the same names.
First response time (FRT) is the time between a customer's first message and the first reply, whether from a person or an AI agent.
FRT = sum of first response times ÷ number of conversations
Report the median as well as the mean, because a few conversations that sat overnight drag the average up. The pitfall is the clock: calendar hours versus business hours can change FRT threefold with no change in behavior. Pick one definition and apply it to every channel.
Average resolution time (ART) is the elapsed time from a conversation opening to the issue being resolved.
ART = sum of resolution times ÷ number of resolved conversations
It comes from the helpdesk and is most useful segmented by topic, because a password reset and a billing dispute should never share a target. The pitfall is what counts as "resolved." Auto-closing after silence improves ART while hiding issues that were never fixed. Measure to a resolution the customer confirmed or your team marked.
Average handle time (AHT) is the active time an agent spends on one contact, including talk time, hold time, and wrap-up work.
AHT = (talk or chat time + hold time + wrap-up time) ÷ number of contacts
Phone systems capture it natively. For messaging, the helpdesk must record active rather than elapsed time. AHT is a staffing input, not a quality target. The pitfall is managing agents to a low AHT, which pushes them to rush and surfaces later as a worse reopen rate and lower FCR.
Average wait time, or average speed of answer, is how long a customer sits in a queue before a live agent engages.
Average wait time = sum of queue wait times ÷ number of answered contacts
It comes from the phone system or chat queue and is best read by hour and day, since weekly averages hide the Monday spike. The pitfall is excluding abandoned contacts: the customers who gave up waited longest.
SLA compliance rate is the share of conversations that met an agreed response or resolution target.
SLA compliance rate = conversations within SLA ÷ conversations with an SLA × 100
The classic contact center version is the "80/20" service level: 80% of calls answered within 20 seconds. The pitfall is pausing the SLA clock too liberally, so compliance stays high while customers actually wait. Decide which states pause the clock and keep that rule stable.
Resolution and quality metrics
These metrics tell you whether the problem was fixed, and fixed well. Many teams also use a QA score, where a reviewer rates a random sample of conversations against a rubric, as a cross-check on all four.
First contact resolution (FCR) is the share of issues fully resolved in a single interaction, with no follow-up and no transfer.
FCR = issues resolved on first contact ÷ total issues × 100
Measure it in the helpdesk by checking whether the same customer raised the same topic again within a set window, such as seven days. The pitfall is relying on the agent's judgment: agent-reported FCR runs higher than customer-confirmed FCR. Decide which you report and say so.
Resolution rate is the share of conversations that ended with the customer's issue genuinely solved.
Resolution rate = resolved conversations ÷ total conversations × 100
It comes from the helpdesk and from AI agent logs. Resolution measures outcomes, not activity. The pitfall is counting closures as resolutions. Define a resolution event (customer confirmation, a positive rating, or no reopen within a window) and measure against that.
Reopen rate is the share of resolved conversations the customer reopened.
Reopen rate = reopened conversations ÷ resolved conversations × 100
The helpdesk records reopens automatically, and they check your other metrics: a low ART with a rising reopen rate means the team is closing conversations, not resolving them. The pitfall is a short reopen window. If conversations lock after 24 hours, reopens appear as new volume instead.
Escalation rate, or transfer rate, is the share of conversations moved to another agent, team, or tier.
Escalation rate = escalated or transferred conversations ÷ total conversations × 100
It comes from the helpdesk's assignment history or the phone system's transfer log. Some escalation is healthy. The pitfall is reading a low rate as inherently good when it may mean frontline agents are guessing at answers they should have passed on.
Customer experience metrics
These X-data metrics come from asking customers, so response rates and timing shape the numbers as much as the service does. AI analysis of conversation sentiment is a useful complement, because it covers every conversation rather than the minority who answer a survey.
CSAT, or customer satisfaction score, is the share of customers who rated a specific interaction positively.
CSAT = positive ratings ÷ total ratings × 100
It comes from a survey at the end of a conversation, usually a 1-to-5 scale where 4 and 5 count as positive. The pitfall is a low response rate, which skews toward customers who were delighted or furious. A 92% CSAT on a 6% response rate says less than an 85% on a 40% response rate.
NPS, the Net Promoter Score, measures how likely customers are to recommend you on a 0-to-10 scale.
NPS = % promoters (9 to 10) minus % detractors (0 to 6)
It comes from a periodic relationship survey, because it measures loyalty to the company rather than satisfaction with one interaction. The pitfall is using it as a support metric. Pricing, product quality, and marketing all move NPS, so a change cannot be attributed to support alone.
Customer effort score (CES) measures how easy the customer found it to get their issue resolved.
CES = sum of effort ratings ÷ number of responses
A post-conversation survey asks customers to rate a statement such as "The company made it easy to handle my issue" on a 1-to-7 scale. The pitfall is asking too early. Send it after resolution, not after the first reply, or you measure the effort of reaching you rather than the effort of getting help.
Volume and workload metrics
Workload metrics explain the denominators behind everything else. A spike in FRT is often a spike in volume.
Ticket volume is the number of new conversations opened in a period.
Ticket volume = count of new conversations
It comes from the helpdesk. Segment it by channel, topic, and AI versus human. The pitfall is counting messages or automated replies as tickets, which inflates volume and hides the real trend.
Backlog, or open tickets, is the number of conversations waiting for action at a given moment.
Backlog = count of open conversations at period end
Take a helpdesk snapshot at the same time each day. Backlog growing faster than volume means capacity is short. The pitfall is counting conversations waiting on the customer. Separate "waiting on us" from "waiting on them."
Contact rate is the number of support contacts relative to your customer base or order volume.
Contact rate = contacts ÷ active customers (or orders) × 100
It needs the helpdesk for contacts and a CRM or order system for the denominator, and it tells you whether your product is generating avoidable work. The pitfall is comparing periods without adjusting for growth: volume can rise while contact rate falls, which is a success.
Self-service and AI metrics
Once help content and an AI agent answer part of your volume, you need metrics for the part no human touched. These definitions are the newest and worth revisiting as tooling changes.
Self-service rate is the share of customer questions resolved by help content without a conversation starting.
Self-service rate = help center sessions without a conversation ÷ total help center sessions × 100
It comes from help center analytics, ideally joined to the helpdesk. The pitfall is counting a visit as a resolution. A customer who read three articles and gave up is a failure, so pair the rate with article feedback and failed searches.
AI resolution rate is the share of conversations an AI agent resolved without handing off to a human.
AI resolution rate = AI-resolved conversations ÷ conversations the AI agent handled × 100
It comes from the AI agent's logs. The pitfall is the definition of "resolved." Some tools count any conversation that never reached a human, which is deflection, not resolution. Insist on a definition where the customer confirmed the answer or did not return. Fin AI Agent resolves 76% of queries on average, including complex, multi-step work, and reports resolution separately from deflection.
Involvement rate, and its complement handoff rate, describe how much of your volume the AI agent engaged with and how much it passed to a person.
Handoff rate = conversations handed to a human ÷ conversations the AI agent handled × 100
The data comes from AI agent logs joined to the helpdesk, and together with AI resolution rate it shows how volume divides: AI-resolved, handed off, or abandoned. The pitfall is reading a falling handoff rate as pure progress. If the AI is keeping conversations it should have escalated, CSAT on AI conversations and reopen rate will show it.
Cost and efficiency metrics
Cost metrics connect everything above to the budget conversation and need data from outside the helpdesk.
Cost per resolution is the fully loaded cost of your support operation divided by the number of issues it resolved.
Cost per resolution = total support cost ÷ resolved conversations
Total cost comes from finance and should include salaries, tooling, and overhead. The pitfall is a changing denominator. Once an AI agent resolves part of the volume, calculate the figure for AI and human conversations separately, because the blended number falls while the human number rises. Fin AI Agent charges $0.99 per outcome, making the AI cost component predictable.
Agent utilization, or occupancy, is the share of agents' available logged-in time spent actively handling contacts.
Agent utilization = handle time ÷ available time × 100
It comes from a workforce management tool or the phone system, and it is a staffing health check, not a target to maximize. The pitfall is pushing occupancy toward 100%: teams with no buffer cannot absorb spikes, and burnout follows.
See what Intercom costs for your team
Helpdesk seats plus a per-resolution price for Fin, published in full so you can model it before you talk to anyone.
How to measure customer service metrics
Every page on this subject lists metrics. Fewer explain why two teams can run the same report and get different answers. The gap is usually in one of five places.
Choose the data source deliberately
Speed and volume metrics live in the helpdesk. Phone-specific measures live in the phone system. Experience metrics come from survey tools. AI metrics come from the AI agent's logs. When a metric needs two systems, decide which is the source of truth for each half and when the join happens.
Write down the definitions
The table below lists the traps that most often make numbers incomparable.
| Metric | Common inconsistency | Recommended definition |
|---|---|---|
| First response time | Business vs calendar hours; first AI reply vs first human reply | Calendar hours for customer-facing reporting, business hours for team targets; label the reply type |
| Resolution | Auto-closed after inactivity vs customer-confirmed | Customer confirmed, positive rating, or no reopen within seven days |
| First contact resolution | Agent-reported vs no repeat contact | No repeat contact on the same topic within seven days |
| CSAT | 4 and 5 positive vs 5 only; every conversation vs a sample | 4 and 5 positive, surveyed after every resolved conversation, response rate shown |
| SLA compliance | Clock pauses when waiting on customer vs never | Pause only on "waiting on customer"; document it |
| AI resolution rate | Any AI conversation without handoff vs confirmed resolution | Customer confirmed or no return within the window |
Mind survey timing and response rates
Send CSAT and CES after resolution, not after the first reply. Report the response rate with every score, and set a minimum sample before reading anything into a segment.
Watch for sampling bias
Low response rates skew toward the extremes. Non-random QA sampling describes the sample, not the whole. Surveys sent on one channel describe that channel. Where you cannot remove bias, state it next to the number.
Segment by channel, topic, and AI vs human
AI changes the denominators. If the AI agent resolves the simple questions, the conversations that reach people are harder on average, so human FRT, ART, and AHT drift upward with no drop in performance. Report each metric three ways: AI, human, and blended. The guide to customer service KPIs explains this complexity shift and how to set targets around it.
How to build a customer service metrics dashboard
A dashboard is a chosen view, not the whole glossary. Six to eight metrics per audience is enough, refreshed as often as that person can act on them.
Support lead dashboard (refreshed daily)
- Ticket volume by channel, today vs the same day last week
- Backlog, split into waiting on us and waiting on customer
- First response time, median and 90th percentile
- SLA compliance rate
- AI resolution rate and handoff rate
- CSAT for the last seven days, with response rate
VP of CX dashboard (refreshed monthly)
- Resolution rate, split AI, human, and blended
- Contact rate, trended against customer growth
- CSAT and CES, with response rates, by quarter
- NPS from the latest relationship survey
- Cost per resolution, AI vs human
- Top conversation topics by volume and negative sentiment
Put the definitions from the traps table beside the charts when sharing a template. Intercom reporting includes prebuilt reports covering these metrics for AI and human conversations, plus a custom report builder.
Are there SaaS-specific customer service metrics?
SaaS teams often add churn rate, customer lifetime value, monthly recurring revenue, onboarding completion, and feature adoption to their support reporting. Those are customer success metrics rather than customer service metrics: they measure the health of the account, not the quality of a support interaction. The guide to customer success metrics covers each one.
Customer service metrics FAQ
What are the main customer service metrics?
The main customer service metrics fall into six groups: speed (first response time, resolution time, SLA compliance), resolution and quality (first contact resolution, reopen rate), customer experience (CSAT, NPS, customer effort score), volume (ticket volume, backlog, contact rate), self-service and AI (self-service rate, AI resolution rate), and cost (cost per resolution, utilization).
What are the 4 or 5 most important customer service metrics?
Most teams start with first response time, first contact resolution, CSAT, and customer effort score, then add AI resolution rate once an AI agent handles part of the volume. Which five matter most depends on your goals and channel mix.
What is the difference between a customer service metric and a KPI?
A metric is anything you can measure. A KPI is a metric you have tied to a goal, with a target and an owner. Every KPI is a metric, but most metrics are not KPIs. This page defines the full set of metrics. The guide to customer service KPIs covers choosing, targeting, and owning them.
How do you measure customer service quality?
Combine an operational measure with an experience measure. First contact resolution and reopen rate show whether problems were actually fixed. QA scores show whether conversations met your standards. CSAT and customer effort score show how the customer felt. Any one alone can be gamed. Together they are hard to fool.
How do you measure AI customer service performance?
Measure AI resolution rate with a strict definition, where the customer confirmed the answer or did not return, rather than counting every conversation the AI touched. Pair it with handoff rate, CSAT on AI-handled conversations, and reopen rate. Report human metrics separately, because conversations that reach people get harder once an AI agent takes the simple ones.
What is a good CSAT score?
There is no universal benchmark. A good CSAT depends on industry, channel, and how the survey is worded. The more useful questions are whether your score is improving, whether your response rate is high enough to trust it, and how AI-handled and human-handled conversations compare.
Keep track of your customer service metrics with Intercom
Intercom is the only helpdesk with a natively integrated AI agent, Fin, so every metric on this page is reported for AI and human conversations in one place. Fin AI Agent averages a 76% resolution rate across thousands of customers, with the rate improving approximately 1% per month over the past 24 months. CX Score evaluates every conversation automatically, providing five times more coverage than CSAT surveys. Topics Explorer groups conversations by subject so you can see which topics drive volume and which score poorly.
See Intercom reporting or explore Fin AI Agent.
Measure AI and human support in one place
Intercom reports on Fin and your team side by side, with resolution rates, CSAT and conversation topics, so you know what to fix next.