Almost every support team tracks numbers. Far fewer know what those numbers really mean, which ones to trust, how to read them together, and how to stop them from quietly changing the behaviour of the people being measured. A dashboard full of green figures can sit on top of unhappy customers. A single red figure can send a good team chasing the wrong problem.
This guide explains customer support metrics from the ground up. You will learn what each metric is, the exact formula, when to use it, how many survey responses you need before a score means anything, what each metric hides, how metrics should be paired, how to set targets when you have no history, and how to diagnose problems from the pattern of your numbers. There are worked examples, charts, a calculator you can use with your own figures, and a glossary of every term at the end.
It is written for support managers, founders who now own a support team, and anyone who has to read a support report or sign a contract that includes service targets.
What Are Customer Support Metrics, and Why Do They Matter?
A metric is a measurement. "We answered 412 chats yesterday" is a number. It becomes useful when it is defined precisely, measured the same way every time, and compared with something: last month, a target, another team, another channel.
Support metrics matter for three reasons that pull in different directions, which is why measurement is harder than it looks.
- For customers, metrics are a proxy for whether they were helped quickly, correctly and kindly.
- For agents, metrics are a signal of what the organisation values. People do what is measured. If you measure speed, you get speed.
- For the business, metrics justify budgets, show where the product causes pain and decide whether to hire, train, automate or outsource.
Leading and Lagging Indicators
Some metrics tell you what has already happened. Others warn you about what is about to happen.
A healthy scorecard has both. If you only watch lagging indicators, you find out about trouble weeks too late. If you only watch leading ones, you may polish activity that does not help customers.
The Four Families of Support Metrics
Most customer support metrics fall into four families. Keeping them in mind stops you from collecting random numbers and helps you notice what you are not measuring.
| Family | Question it answers | Examples |
|---|---|---|
| Customer experience | How did customers feel? | CSAT, NPS, customer effort score, complaints |
| Resolution | Did we actually solve the problem? | First contact resolution, reopen rate, time to resolution, escalation rate |
| Speed | How quickly did we respond and finish? | First response time, handling time, wait time, service level, abandon rate |
| Quality and efficiency | Did we do it right, and at what cost? | QA score, occupancy, cost per resolved contact, backlog, attrition |
The most common mistake is to over-weight the speed family because it is the easiest to measure from system data. Speed matters, but it is the cheapest to improve in ways that hurt customers: answer faster with a useless reply, close tickets before they are solved, rush calls. The families keep each other honest.
Customer Experience Metrics
These come from asking customers. They are the closest you get to the truth about how support felt, and also the easiest to misread.
CSAT: Customer Satisfaction Score
CSAT measures how satisfied a customer was with a specific interaction. After a chat, call or ticket, you ask a question such as "How satisfied were you with the help you received today?" and offer a scale, most often 1 to 5.
A worked example. Suppose 200 customers answered your survey after a support interaction. Their scores were: 5 stars, 90 people; 4 stars, 74; 3 stars, 22; 2 stars, 8; 1 star, 6. The positive responses (4 or 5) number 90 + 74 = 164. So CSAT is 164 divided by 200, or 82%.
Choices you must make, and write down.
- The scale. Five points is common. Some teams use thumbs up or down, a 1 to 10 scale or smiley faces. Changing the scale changes the numbers, so keep it fixed or restart your trend line.
- What counts as positive. The usual choice is the top two boxes (4 and 5). Counting only 5 gives a much lower figure. Neither is wrong. Be consistent.
- When you ask. Immediately after the interaction captures a fresh memory but may catch customers before their problem has really been fixed. Asking a day later measures the outcome, but fewer people reply.
- What you ask about. "How satisfied were you with the agent?" measures the person. "How satisfied were you with the help you received?" measures the outcome. Customers often blame the agent for a policy they dislike. Decide which you want.
What CSAT hides. CSAT only counts people who answered. Very happy and very unhappy customers reply more often than people in the middle. A score based on a small, self-selected sample can be badly off. CSAT also says nothing about customers who gave up and never contacted you again. And it can be pulled around by things outside support's control, such as a price rise or a product outage.
NPS: Net Promoter Score
Net Promoter Score asks one question: "On a scale of 0 to 10, how likely are you to recommend us to a friend or colleague?" It was developed by Fred Reichheld with Bain & Company and Satmetrix, and it measures overall loyalty to a company rather than satisfaction with one interaction.
A worked example. Out of 200 answers, 100 are promoters, 60 are passives and 40 are detractors. Promoters are 50% of the total, detractors 20%. NPS is 50 minus 20, which is +30. Note that NPS is a number between minus 100 and plus 100, not a percentage, and the "+" sign is part of the convention.
Why NPS is easy to misuse in support.
- It measures the whole relationship (product, price, brand, support), so a support team cannot fully control it.
- It is often collected on a different schedule (quarterly, after purchase) from support contacts, which makes it a poor measure of one agent or one week.
- It is very sensitive to sample size. Moving a few customers from passive to detractor swings the score noticeably.
- Scores differ greatly between industries and countries because of cultural habits in how people use the scale. Comparing yourself with an average from another sector can mislead.
Some companies send a "transactional" NPS survey after support contacts to look at support's effect on loyalty. That is reasonable, as long as you label it as such and do not compare it with a company-wide figure.
CES: Customer Effort Score
Customer Effort Score asks how easy it was for the customer to get their problem solved. A typical form is a statement, "The company made it easy for me to handle my issue", with agreement on a seven-point scale from "strongly disagree" to "strongly agree".
CES rests on an idea supported by customer service research: customers care less about being delighted than about not having to work hard. Repeating information, being transferred, switching channels and waiting for callbacks all increase effort. A support team can lower effort directly, by fixing handovers, improving self-service and solving problems on the first contact.
Use CES after interactions where the customer had a task to complete, such as fixing a problem or changing an account. It is less useful for conversations that are only informational.
CSAT, NPS and CES Compared
| CSAT | NPS | CES | |
|---|---|---|---|
| Asks about | One interaction | The whole relationship | The effort of one task |
| Typical scale | 1 to 5 | 0 to 10 | 1 to 7 agreement |
| Answer to | "Were you happy with this help?" | "Would you recommend us?" | "Was it easy?" |
| Best for | Judging contacts, coaching, quality | Tracking loyalty over time | Finding friction in processes |
| Responds | Fast | Slowly | Fast |
| Main risk | Small, self-selected samples | Blamed on support for things it cannot control | Measuring only simple tasks |
If you can only run one survey, run CSAT after the contact and add one optional free-text question: "What could we have done better?" The comments will teach you more than the number.
How Many Responses Do You Need?
This is the question that most guides skip, and it matters more than almost anything else on a survey. A score from a handful of responses is noise.
When you survey a sample of customers and compute a percentage, the true figure for all customers could be a little higher or lower. The size of that uncertainty is the margin of error. For a percentage, a standard approximation at 95% confidence is:
Two lessons follow. First, do not celebrate or panic over small movements in survey scores unless you have many responses. Second, never rank individual agents on a handful of surveys. An agent with 12 responses and an agent with 90 are not comparable. If you must show agent-level scores, show the number of responses beside them and withhold the score below a minimum.
Survey Bias and Response Rates
- Non-response bias. People who reply differ from those who do not. Encourage replies from everyone: short surveys, one tap, sent promptly.
- Selection by the agent. If agents decide when a survey is sent, they will send it after the good conversations. Send it automatically, to a random or complete set.
- Leading wording. "How great was your experience?" pushes answers up. Use neutral wording.
- Survey fatigue. Asking after every contact annoys loyal customers. Cap how often one person is surveyed.
- Coaching to the score. Agents who ask "please give me five stars" are measuring their charm, not service. Make this a coaching point.
Try It: Three Core Metrics Calculator
The numbers below are examples. Replace them with your own and the results update as you type. Nothing you enter is sent anywhere.
Resolution and Effectiveness Metrics
Experience scores tell you how customers felt. Resolution metrics tell you whether the problem actually got solved, which is the strongest driver of how customers feel.
First Contact Resolution (FCR)
FCR sounds simple, and it is the metric most often measured badly, because "first contact", "resolved" and "same problem" all need definitions. Here are the four common ways to measure it, from easiest to most accurate.
| Method | How it works | Weakness |
|---|---|---|
| Agent marks it | The agent ticks "resolved on first contact" | Agents are measured on it, so it drifts upwards |
| Closed with no reopen | Ticket closed and never reopened | Customers may simply give up or phone instead |
| No repeat contact in a set period | No new contact from the same customer on the same topic within, say, 7 days | Needs good topic matching, and customers may call about a different topic |
| Customer confirms | The customer is asked "Was your issue fully resolved?" | Needs survey responses, and some customers do not reply |
None of these three figures is dishonest. They measure different things. The danger is mixing them: reporting the agent-marked 78% this month and the stricter 69% next month and seeing a "fall" that never happened. Pick one definition, write it down, and keep it. For contract and board reporting, prefer the stricter ones.
Why FCR matters. Every repeat contact costs money, frustrates the customer and clogs the queue. A support team that improves FCR usually improves satisfaction and lowers cost at the same time, which is rare.
What FCR hides. A team can raise FCR by refusing to escalate when it should, or by closing tickets that were not solved. A high FCR must therefore be read next to the reopen rate and customer satisfaction.
Reopen Rate and Repeat Contact Rate
Reopens are an honest signal, because customers vote with their behaviour. A rising reopen rate points to rushed closing, thin knowledge, poor handovers or incomplete fixes.
Resolution Rate and Time to Resolution
Resolution time is a clock from the customer's point of view, so it includes waiting for the customer to reply. Many teams measure two versions: elapsed time, which includes the waits, and handling time, which counts only active work. Know which one you are quoting.
Escalation Rate
The share of tickets passed up to a higher tier. As covered in our guide to L1, L2 and L3 support, a very high rate points to training or permission gaps, and a very low one may mean agents are hanging on to problems. Always split it by reason.
Backlog and Ticket Age
Look at age, not only count. Ten tickets that are a day old are fine. Ten that are six weeks old are a quiet disaster. Report the oldest ticket and the number older than your promised time.
Speed and Responsiveness Metrics
First Response Time (FRT)
Three traps:
- Automatic replies do not count. An auto-acknowledgement is useful but is not a response. Count the first reply from a person that moves the case forward.
- Averages mislead. A few very long waits drag an average up, and a flood of instant replies drags it down. Use the median (the middle value) and a percentile (for example, "90% of customers get a first reply within X").
- Business hours versus clock hours. A reply to a message sent at midnight is quick if your clock starts at nine the next morning, and slow if the customer was awake. State which you use.
Average Handle Time (AHT)
AHT is useful for staffing: more minutes per contact means more agents needed. It is risky as a performance target, because a lower AHT is not better by itself. Agents who feel timed will rush, avoid asking the question that would reveal the real problem, or end contacts before they are solved, and repeat contacts rise. Use AHT to plan capacity and to find process problems (a step that makes every contact slow), not to rank people.
Service Level, Wait Time and Abandon Rate (Voice and Live Chat)
These matter for live channels, where the customer is waiting right now. Service level is more informative than ASA because it tells you what share of customers got an acceptable wait.
Why staffing is not a straight line. Here is a result that surprises many managers. Suppose a phone line receives 120 calls an hour and each call takes 4 minutes of agent time. That is 8 hours of work arriving every hour, which queueing theory calls a workload of 8 "Erlangs". You might guess that 8 agents are enough. They are not, because calls arrive at random and queues form whenever the next call finds everyone busy. A standard queueing formula called Erlang C estimates how many agents you need for a given service level. Using it for this example:
Notice three things. With 10 agents, service falls to about 65% answered within 20 seconds. With 9, it collapses to about 40%. Going from 11 to 12 agents helps a lot, but going from 14 to 16 hardly changes the result. Being one agent short can be far worse than being one agent over, and each extra agent after the first few has less effect than the one before.
Contact Rate
Contact rate is one of the most valuable measures a support lead can give to the rest of the company. If contacts per 100 orders rise after a product change, support is telling you that something confusing or broken was shipped.
Quality Assurance (QA) Scores
Surveys tell you what customers thought. Quality assurance tells you whether the conversation met your standards, whether or not the customer filled in a survey. It is done by reviewing a sample of conversations against a scorecard.
A typical scorecard rewards:
- Accuracy: correct information, correct steps, correct policy.
- Resolution: the problem was solved or properly escalated.
- Process and compliance: identity checks, required statements, correct notes.
- Communication: clarity, tone, empathy, plain language.
- Ownership: next steps clearly stated, customer left informed.
Some items are critical (an "auto-fail"): giving wrong financial information, breaking a privacy rule, being rude. A single critical failure should fail the whole review no matter how good the rest was.
Sampling. Review a random sample across agents, channels, times of day and contact types, plus extra samples of escalations, complaints and new starters. Reviewing only the "best" or "worst" conversations distorts the picture.
Calibration. Two reviewers can score the same conversation differently. Calibration sessions, where reviewers score the same contacts and compare, keep scoring consistent. Without them QA scores say more about the reviewer than about the agent.
Fairness. Share the scorecard in advance, explain it, let agents see their reviews and challenge them, and use QA to coach rather than punish. Our quality and compliance support service covers this work in more detail.
Efficiency, Capacity and Cost Metrics
Cost Per Contact Versus Cost Per Resolved Contact
This is why in-house or outsourced customer support comparisons should always use cost per resolved contact, not cost per hour.
Metrics That Keep Each Other Honest: Pairing Rules
Every metric has a cheap way to improve it that hurts customers. The defence is to pair each metric with its counter-metric.
| If you push this | People may do this | Pair it with |
|---|---|---|
| Faster first response | Send empty replies to stop the clock | Resolution time, CSAT, reply quality in QA |
| Lower handle time | Rush calls, skip checks | FCR, repeat contact rate, QA accuracy |
| Higher FCR | Close tickets early, avoid escalating | Reopen rate, CSAT, escalation quality |
| Higher CSAT | Ask for five stars, send surveys after good calls only | Survey coverage, response counts, QA |
| Fewer escalations | Hold on to problems too long | Time to resolution, customer effort |
| More tickets closed | Split one problem into many, close without fixing | Reopen rate, resolved-by-customer-confirmation |
| Lower cost per contact | Understaff, cut training | Cost per resolved contact, attrition, QA |
How to Set Targets When You Have No Benchmark
New teams ask: "What is a good CSAT?" or "What FCR should we aim for?" The honest answer is that published figures come from many different definitions, industries and customer mixes, so they are shaky. A better path is to build your own baseline.
- Define each metric exactly, in writing, with the formula, the data source and what counts as in or out.
- Measure for four to eight weeks without targets. You will learn the normal range and find data problems.
- Look at the distribution, not only the average: best and worst weeks, differences between channels, priorities and teams.
- Set three levels: a floor (below this, we act), a target (where we aim to be) and a stretch (excellent). Targets are usually set a little above your normal range, not far above it.
- Set targets by channel and priority. A chat target should not equal an email target, and a critical issue should not share a target with a how-to question.
- Review quarterly. When a metric is comfortably met for several periods, raise it or move your attention to the next constraint.
- Use external benchmarks as context only, after checking that their definitions match yours.
Building a Scorecard and Dashboard
The aim of a scorecard is to let someone understand the health of support in two minutes. Fewer, better chosen numbers beat a wall of charts.
One page per audience.
- Executives: five or six numbers with trend arrows: CSAT, FCR, time to resolution, cost per resolved contact, contact rate, and one line on risks.
- Support leaders: the above, plus backlog age, QA, escalation reasons, staffing and service level.
- Team leads and agents: their own QA and customer comments, response counts, coaching notes, and team trends.
Cadence. Daily for queue health (backlog, service level, oldest ticket). Weekly for quality and trends. Monthly for scorecards and contract reviews. Quarterly for targets and definitions.
612 responses, margin about 3 points. Up 1 point.
Reopens within 7 days: 6%.
Median. 90th percentile: 42 min.
Median, business hours. Oldest open: 6 days.
Contacts per 100 orders. Down from 4.6.
120 reviews, 2 critical fails.
Dashboard hygiene.
- Show the definition next to each number, or one click away.
- Show sample sizes beside every survey score.
- Use trend lines over a few months, not single-month bars.
- Use medians and percentiles for time measures.
- Always include comments. Numbers say what happened. Customers' own words say why.
Metrics by Channel
| Channel | Most useful speed measure | Resolution measure | Experience measure |
|---|---|---|---|
| Phone | Service level, wait time, abandon rate | FCR, repeat contacts | CSAT, CES |
| Live chat | Wait time, first response, concurrent chats per agent | FCR, chat reopen | CSAT |
| First response (median, percentile) | Time to resolution, touches per ticket | CSAT, comments | |
| Social media | First response, time to move to private channel | Resolution rate | Sentiment, CSAT |
| Self-service | Search success, time to find an answer | Contacts avoided, article helpfulness | Effort, feedback |
Reading the Pattern: Five Worked Diagnoses
The real skill is not calculating metrics but reading them together. Here are five patterns that appear again and again, with the questions to ask before you act. In each case, treat the explanation as a hypothesis to test by reading real tickets, not as a verdict.
Pattern 1: Satisfaction Is Falling but First Contact Resolution Is Flat
What it might mean. Customers are getting problems solved at about the same rate, but they are less happy. Look for changes in how they are treated or what they are asked to do: a new policy customers dislike, a tone change after a script update, longer waits that do not show up in FCR, a new self-service step that adds effort, or a product change that makes customers more anxious.
What to check. Read the low-score comments from the last two weeks and tag them by theme. Compare CSAT by contact reason, by channel and by wait time. Check whether CES moved.
Pattern 2: Handle Time Fell and Satisfaction Fell Too
What it might mean. Agents are rushing. A well-run change (better tools, a better knowledge base) lowers handle time without hurting satisfaction. If both move, the saving probably came from shortcuts.
What to check. QA scores for "discovered the real issue" and "explained next steps". Repeat contact rate within seven days. Whether a new target or league table on handle time was introduced.
Pattern 3: First Response Time Is Excellent but Resolution Takes Longer Every Month
What it might mean. The team is working to the clock that is measured. Replies go out fast but do not move the case forward ("We are looking into this"), and the real work waits. It can also mean complexity is rising or the escalation route is slowing down.
What to check. Number of touches per ticket. Time spent waiting in escalation queues. The share of first replies that ask for information already in the ticket. Whether a first-response target exists without a matching resolution target.
Pattern 4: NPS Rose, CSAT Fell
What it might mean. The two answer different questions. NPS may have risen because of a product release, a price change or a marketing campaign, while the support experience got worse. Or the samples differ: NPS surveyed a broad base, CSAT only people who contacted support.
What to check. The survey populations and dates. Whether anything outside support changed. Whether support-driven NPS (asked after contact) tells the same story as CSAT.
Pattern 5: the Escalation Rate Jumped
What it might mean. A new product release with undocumented behaviour, a gap in the knowledge base, a new batch of untrained agents, a change in what L1 is allowed to do, or a genuine incident.
What to check. Escalation reasons split by week. The ticket subjects that make up the jump. Whether a product or policy change shipped the week before. Whether the notes quality changed.
Metrics When You Outsource Support
Outsourcing makes measurement more important, not less, because the people doing the work are no longer in the next room. Three things matter most: agreeing definitions, separating targets from penalties, and seeing the data yourself.
Write the Formula Into the Contract
Most disputes over service targets come from two parties meaning different things by the same word. A good agreement defines each metric with its exact formula, the data source, the time zone, what is excluded (for example, contacts caused by an outage you declared) and who calculates it.
SLA, KPI and Service Credit: Three Different Things
| Term | What it is | Typical use |
|---|---|---|
| SLA | A promised minimum level of service | Response and resolution times, availability |
| KPI | An important measure tracked over time | CSAT, FCR, QA score |
| Service credit | An agreed reduction in fees when an SLA is missed | Compensation, and an incentive to stay on target |
Where possible, make SLAs few and clear and keep KPIs for improvement conversations. A contract that punishes every missed number encourages a vendor to hide problems. A contract that mixes a few firm commitments with joint reviews of the wider KPIs encourages honesty. Note that incentives and penalties differ between businesses and should be agreed with your own advisers.
See the Data Yourself
Ask for read access to the helpdesk and survey data, not only a monthly spreadsheet. Agree a shared dashboard, a weekly operational review and a monthly business review. Agree who will run calibration sessions on QA, so that the partner's reviewers and yours score the same way.
Questions to Ask a Partner About Metrics
- Which definitions do you use for first response, resolution and FCR? Can we agree ours in the contract?
- How do you sample for quality, and how often do you calibrate with us?
- How do you stop agents gaming scores? What pairs of metrics do you watch?
- What will we see, how often, and can we see raw data?
- How do you report problems early, before the monthly review?
- What happens to targets in the first ninety days? (See our 90-day transition plan.)
Our guide on how to choose a support outsourcing partner goes into evaluating reporting and governance, and our quality and compliance support and customer care pages describe how we run scoring, calibration and reporting for clients.
Ten Common Mistakes (And Their Fixes)
- Changing definitions mid-stream. Fix: version your definitions, and restart trend lines when you change them.
- Reporting averages for time. Fix: median and a percentile, plus the worst case.
- Scoring agents on tiny samples. Fix: show response counts and set a minimum before showing a score.
- Measuring only what is easy. System data is easy; customer truth is harder. Fix: add surveys, QA and comment review.
- One target for every channel and priority. Fix: set targets per channel and per priority.
- Treating CSAT as a performance rating. Fix: use it for coaching conversations and trend, not league tables.
- No counter-metric. Fix: pair each speed or volume number with a quality one.
- Ignoring the comments. Fix: tag them, count themes and act on the top three.
- Blaming support for company-wide problems. Fix: split metrics by contact reason and feed product issues to product teams.
- Reporting without acting. Fix: every review ends with named owners, actions and a date.
A 60-Day Plan to Fix Your Measurement
Days 1 to 10: define. List every metric you report. Write the formula, data source and owner for each. Cut anything nobody uses. Choose one definition of FCR and write it down.
Days 11 to 25: collect. Turn on automatic, complete CSAT surveys with a free-text box. Start a weekly QA sample with a simple scorecard. Record the baseline for each metric by channel.
Days 26 to 40: pair and calibrate. Add counter-metrics. Run the first calibration session. Review twenty tickets behind every metric that moved.
Days 41 to 55: set targets. Set floor, target and stretch from your baseline, per channel and priority. Share them with the team, explaining what each is for and what it is not for.
Days 56 to 60: publish and review. Publish the one-page scorecard. Hold the first monthly review with named owners. Schedule the quarterly definition review.
Putting It All Together
Customer customer support metrics are a language for talking about service. Used well, they show whether customers are being helped, where effort is wasted and where to invest. Used badly, they become a game that people win and customers lose.
The core ideas:
- Define every metric precisely and keep the definition stable.
- Use all four families: experience, resolution, speed and quality with efficiency.
- Respect sample size. A survey score from a handful of responses is noise.
- Pair speed with quality. Every shortcut has a counter-metric that exposes it.
- Measure resolution honestly. Prefer customer confirmation and repeat contact over agent opinion.
- Read tickets, not only charts.
- Put definitions and data access into any outsourcing contract.
We run customer support and sales support for businesses, so we write from the day-to-day work. This guide is general information, not legal, financial, medical or compliance advice. Rules differ by country and change, so check what applies to you with your own advisers. We list our sources at the end. If you spot something out of date, write to info@connecgen.com.


