Customer Service Metrics: 15 KPIs to Track and Improve Support

Your first response time looks great. Your queue is basically empty. And somehow the same customers keep writing back.

I've watched this happen at a lot of DTC brands. The dashboard says support is crushing it, but the inbox tells a different story. A ticket marked "closed" doesn't mean the customer's problem went away. Sometimes it just means someone hit send.

So here's my list of the customer service metrics that actually matter, how to measure each one, and how to tell when a number is lying to you. Fifteen in total, grouped by what they tell you. Some you'll check every morning. A couple you'll only look at once a quarter. All of them are worth understanding before you pick a support partner or rebuild your reporting.

Let's get into it.

What Are Customer Service Metrics?

Customer service metrics are the numbers you use to understand how your support team is performing. How fast you reply, how often you solve the problem, how customers feel afterward, what it all costs.

KPIs are the handful of metrics you've decided to hold yourself accountable to. Every KPI is a metric. Most metrics aren't KPIs.

Example: you might track average handle time because it helps with staffing math. But your actual KPI for the quarter could be "get reopen rate under 5% before BFCM." One is context. The other is the goal.

It also helps to split metrics into two buckets:

• Operational metrics tell you how the machine is running. Response time, backlog, handle time, SLA compliance.
• Customer experience metrics tell you how it felt on the other side. CSAT, effort, whether they came back.

You need both. Teams that only watch operational numbers end up fast and frustrating. Teams that only watch CX numbers can't explain why things changed.

Why Are Customer Service Metrics Important?

Because "support feels busy" isn't something you can fix.

Metrics turn vague feelings into specific problems. CSAT dipping on refund tickets. Backlog spiking every Monday after a weekend promo. Reopens climbing on one agent's macros. Once you can see it, you can do something about it.

They also help you plan. If you know your contact rate is roughly 4 tickets per 100 orders, you can look at your BFCM forecast and figure out how much coverage you actually need. Way better than guessing and drowning on Cyber Monday.

And they keep everyone honest about quality. It's very easy to make a team look efficient on paper. It's much harder to make customers stop writing back. The right mix of metrics shows you both sides.

Last thing. Support is one of the few places where you talk to customers one on one, usually right when something has gone wrong. How that conversation goes has a real effect on whether they order again. Metrics are how you find out if you're helping or hurting.

15 Essential Customer Service Metrics to Measure

Customer Satisfaction and Experience Metrics

1. Customer Satisfaction Score (CSAT)

CSAT asks customers to rate their support experience, usually on a 1 to 5 scale, right after a ticket is resolved.

How to calculate it: (Number of 4 and 5 ratings ÷ Total responses) × 100

The overall score is fine for a board slide. The useful stuff happens when you break it down by contact reason. Low CSAT on "where is my order" tickets is usually a shipping or carrier problem, not an agent problem. Low CSAT on returns might mean your policy is the issue. Read the comments too. That's where the real answers are.

2. Net Promoter Score (NPS)

NPS asks how likely someone is to recommend your brand, on a 0 to 10 scale.

How to calculate it: % of Promoters (9 to 10) minus % of Detractors (0 to 6)

Heads up: NPS measures how people feel about your whole brand. Product, price, shipping, packaging, all of it. Support influences it, but it doesn't own it. Use it as a gut check on the relationship, and compare NPS for customers who contacted support against those who didn't.

3. Customer Effort Score (CES)

CES asks some version of "How easy was it to get your issue resolved?" Usually on a 1 to 7 scale.

How to calculate it: Sum of all scores ÷ Number of responses

This one's underrated. Customers forgive a lot. What they don't forgive is having to explain their problem three times, dig up their order number twice, or get bounced from chat to email. High effort scores usually point to process problems you can actually fix.

Resolution Quality Metrics

This is the section I'd spend the most time on. Closing a ticket and resolving a problem are two very different things, and most dashboards treat them like the same thing.

4. First Contact Resolution (FCR)

FCR tracks the share of tickets solved in a single interaction, with no follow-up needed.

How to calculate it: (Tickets resolved on first contact ÷ Total tickets) × 100

Define "first contact" clearly before you start tracking. Most teams count a ticket as FCR if the customer doesn't write back about the same issue within 7 days. A lot of teams treat 70% as a reasonable bar, though it depends heavily on your mix. Subscription brands with lots of "skip my next box" requests will naturally run higher than brands dealing with damaged shipments.

5. Overall Resolution Rate

This is the share of tickets that ended with the customer's issue actually solved.

How to calculate it: (Resolved tickets ÷ Total tickets received in the period) × 100

The trap here is auto-close. If your helpdesk closes tickets after 48 hours of silence, those all count as "closed." Some of those customers got what they needed. Some gave up. Some went straight to a chargeback. Track resolved and closed as separate statuses if you can.

6. Ticket Reopen Rate

The share of resolved tickets that come back because the problem wasn't fixed.

How to calculate it: (Reopened tickets ÷ Resolved tickets) × 100

If I could only look at one number to judge a support operation, it'd be this one. Reopen rate is the honesty check on everything else. Fast response times, great handle times, huge resolution numbers... if reopens are climbing, none of it is real.

Watch it by agent, by macro, by contact reason, and by channel. And if AI is replying to customers, watch AI-handled reopens separately from human ones.

Speed and Operational Efficiency Metrics

Speed matters. Customers hate waiting. But every metric in this section can be pushed in the wrong direction if nobody is watching quality at the same time.

7. First Response Time (FRT)

How long a customer waits for the first real reply.

How to calculate it: Sum of first response times ÷ Number of tickets (or better, use the median)

Use median, not average. A few tickets stuck over a long weekend will wreck your average and hide what most customers actually experience. And don't count auto-acknowledgements. "We got your message!" is not a response.

8. Average Resolution Time (ART / TTR)

How long it takes from the moment a ticket opens to the moment it's actually resolved.

How to calculate it: Total time to resolution across tickets ÷ Number of resolved tickets

Look at the long tail. Your median might be 6 hours while 10% of tickets take four days. Those four-day tickets are usually your angriest customers, and they're often stuck waiting on a warehouse, a 3PL, or a refund approval.

9. Average Handle Time (AHT)

The active time someone spends working a ticket, including research and writing the reply.

How to calculate it: Total active handling time ÷ Number of tickets handled

Super useful for staffing and for spotting tickets that are way more complicated than they should be. Please don't turn it into an agent target, though. When people get rewarded for short handle times, they rush. Rushing creates reopens.

10. Ticket Volume and Backlog

Volume is how many tickets come in. Backlog is how many are still open and unresolved at the end of a day or week.

How to calculate it: Count tickets created per period. Backlog = Open tickets at period end.

For ecommerce, normalize volume against orders. Tickets per 100 orders (your contact rate) tells you if support demand is growing faster than the business. If it is, something upstream is broken. Product pages, shipping, a confusing subscription flow, something.

11. SLA Compliance Rate

The share of tickets that hit your committed response or resolution time.

How to calculate it: (Tickets meeting SLA ÷ Total tickets) × 100

Set different SLAs by channel and priority. Live chat customers expect minutes. Email customers can usually wait a few hours. A VIP subscriber with a failed payment probably shouldn't sit in the same queue as a question about restock dates.

12. Escalation Rate

The share of tickets that get passed to a senior agent, a specialist, or another team.

How to calculate it: (Escalated tickets ÷ Total tickets) × 100

Too high usually means your frontline doesn't have the knowledge or the authority to fix things. Can they issue a refund? Send a replacement? If not, everything escalates. Too low can be a problem too, especially with AI in the mix. A system that never escalates might be holding on to tickets it has no business answering.

Business Impact Metrics

13. Customer Retention Rate and Churn Rate

Retention is the share of customers who stick around over a period. Churn is the flip side.

How to calculate retention: ((Customers at end of period − New customers acquired) ÷ Customers at start of period) × 100

How to calculate churn: (Customers lost ÷ Customers at start of period) × 100

Support doesn't own retention. Product and price and shipping all play a part. But you can still learn a lot by comparing repeat purchase rate or subscription churn for customers who had a good support experience against those who had a bad one. If the gap is big, that's your business case.

14. Cost per Resolved Ticket

What it actually costs you to get one customer problem solved.

How to calculate it: Total support costs (people, tools, partners) ÷ Number of resolved tickets

Note the word "resolved." A lot of teams calculate cost per ticket handled, which makes cheap, sloppy answers look efficient. A $2 reply that reopens twice is way more expensive than a $5 reply that fixes it the first time. This is the metric to use when you're comparing support vendors.

AI Customer Service Performance Metrics

15. AI Resolution Quality and Accuracy

If AI is answering customers on your behalf, you need to know if it's right. Not just if it replied.

There isn't one formula here. Track a few things together:

• Accuracy rate. From QA sampling, the share of AI replies that were factually correct and followed your policies.
• AI resolution rate. AI-handled tickets that were actually resolved ÷ total AI-handled tickets.
• AI reopen rate. Compare it directly to your human reopen rate.
• Escalation quality. When AI hands off, does the agent get full context, or does the customer have to start over?
• CSAT on AI-handled tickets. Same survey, separate cut.

What you want to avoid is judging AI on containment. "AI handled 70% of tickets" sounds amazing until you find out a chunk of those customers wrote back angry the next day.

How to Choose the Right Customer Service Metrics for Your Business

You don't need all fifteen on one dashboard. Pick based on where you are.

Start with your goal. Trying to keep CSAT steady while volume doubles? Focus on FRT, backlog, and SLA compliance, with reopen rate as your guardrail. Trying to cut costs? Cost per resolved ticket and FCR. Rolling out AI? Accuracy, AI reopen rate, and escalation quality.

Look at your channels. Brands heavy on chat should care a lot more about first response time than brands that live in email. If you're running phone support, AHT matters more.

Get a baseline first. Track a metric for at least four to six weeks before you set a target. Your numbers in a sale month and a slow month will look completely different.

Pair every speed metric with a quality metric. This is the big one. FRT with reopen rate. AHT with CSAT. Resolution rate with QA accuracy. If you optimize one number alone, you'll almost always break another one without noticing.

How to Measure and Track Customer Service Metrics

Use Help Desk Reports and Ticket Data

Your helpdesk already tracks most of the operational stuff. Response times, resolution times, volume, SLA compliance, tags.

The catch is tagging. If tickets aren't tagged consistently by contact reason, you can't break anything down in a useful way. Spend an afternoon cleaning up your tag list. Fewer tags, clearer definitions. Future you will be grateful.

Collect Customer Feedback

Send CSAT surveys right after resolution while the experience is fresh. One question plus an optional comment box. That's it. Long surveys get ignored.

Run CES on specific flows you're trying to improve, like returns or subscription changes. Save NPS for a quarterly pulse across your whole customer base.

Monitor Quality Through Ticket Reviews and QA

Surveys only tell you about customers who bothered to respond. QA tells you about everyone else.

Review a sample of tickets every week against a simple scorecard. Was it accurate? Did it follow policy? Did it actually solve the problem? Was the tone right? If AI is in the mix, review its tickets with the same scorecard. Tools built for customer support quality monitoring can help you cover far more tickets than manual spot-checks.

Build a Customer Service Metrics Dashboard

Once you know what you're tracking, put it in one place. A good customer support dashboard shows quality and speed side by side, so nobody celebrates a faster FRT while reopens quietly climb.

Keep the daily view short. Five or six numbers max. Put the deeper cuts one click away. If you want to go further on building this out, here's our breakdown of ecommerce customer support analytics.

How to Improve Customer Service Performance Using Metrics

Numbers don't fix anything by themselves. Here's the loop that actually works.

Spot the trend. Look week over week, not day over day. One bad Tuesday is noise. Three bad weeks is a pattern.

Dig into the root cause. Pull the actual tickets behind the number. Read twenty of them. You'll usually figure out what's going on faster than any report will tell you.

Change one thing. A macro, a policy, a routing rule, a help center article. One at a time, so you know what worked.

Measure again. Give it two to four weeks and check the same metric, plus whatever metric it's paired with.

Here's a real-world style example. Say a brand gets its first response time down from 6 hours to 40 minutes. Everybody's thrilled. Then a month later reopen rate has gone from 8% to 14%.

When they dig in, the cause is a new set of fast-reply macros for damaged-item tickets. Agents were sending "So sorry! Can you send us a photo?" within minutes, and the ticket got marked solved. Customers sent the photo, the ticket reopened, and then it sat in the back of the queue. Faster first reply, slower actual fix, more frustrated customers.

The fix wasn't slowing down. It was keeping those tickets in a pending state until the replacement shipped, and letting agents approve replacements without a photo for repeat customers under a certain order value. FRT stayed fast. Reopens dropped back down.

That's the whole point of tracking more than one number.

How AI Changes Customer Service Metrics

AI makes some numbers look incredible overnight. Response time drops to seconds. Volume handled goes way up. Containment looks huge.

Let's be real, though. A lot of AI support tools are built to deflect. They answer something, anything, and count the ticket as handled. The customer either gives up or comes back. Neither of those show up in a containment rate.

That's why AI needs a different scorecard. Accuracy, resolution, reopen rate, and how cleanly it hands off when it's out of its depth. The question isn't "how many tickets did the AI touch?" It's "how many problems did it actually solve, and how many did it make worse?"

This is the gap we built Kim around. Kim is an AI-powered customer support service for Shopify and DTC brands where AI handles tickets and trained human Sentinels own the outcome.

Every ticket gets a confidence score. When the AI is above 90% confident, it resolves the ticket on its own. Below that, a Sentinel steps in to review the plan, re-prompt, or take it over. Before anything goes out, there's a check on the plan and another check on the reply itself. Two gates, not zero.

And because Sentinels correct the AI in real time, it keeps getting better. Human corrections typically drop from 100% to around 15% by day 90. Knowledge comes from your actual ticket history, not a stack of SOPs someone wrote two years ago.

What that means for your metrics: you get the speed that AI is good at without the reopens that usually come with it. Reopen rate stays the number we hold ourselves to.

If you want to see what that looks like on your own ticket data, grab 30 minutes with our team. We'll walk through your current metrics and show you where the gaps are.

Frequently Asked Questions

What Are Some Good KPIs for Customer Service?

CSAT, first contact resolution, ticket reopen rate, first response time, and cost per resolved ticket make a solid starting set. Together they cover satisfaction, quality, speed, and cost.

What Are the 5 Key Customer Experience (CX) Metrics?

Most teams land on CSAT, NPS, Customer Effort Score, retention rate, and churn rate. The first three tell you how customers feel. The last two tell you what they actually did about it.

What Are 5 Examples of Metrics to Measure Customer Service Performance?

First response time, average resolution time, first contact resolution, ticket reopen rate, and SLA compliance rate.

What Are 7 Qualities of Good Customer Service?

Accuracy, speed, empathy, clear communication, ownership of the problem, consistency across channels, and making things easy for the customer. If you had to pick one, pick ownership. Customers can tell when someone actually cares if the problem gets solved.

What Are the Most Important Customer Service Metrics?

It depends on your goals, but CSAT, FCR, and reopen rate are the three I'd never drop. They show whether customers are happy and whether their problems actually stayed solved.

What Is the Difference Between Customer Service Metrics and KPIs?

Metrics are anything you measure. KPIs are the few metrics you've tied to a specific goal and hold the team accountable for. You might track twenty metrics and only have four KPIs.

How Do You Calculate Customer Service Performance?

There's no single number. Most teams combine a satisfaction metric (CSAT), a quality metric (FCR or reopen rate), a speed metric (FRT or resolution time), and a cost metric (cost per resolved ticket). Read them together, never in isolation.

What Is a Good Customer Service Resolution Rate?

It varies by industry and ticket mix, but for DTC brands, getting the large majority of tickets resolved without a reopen is the real goal. Watch your reopen rate alongside resolution rate. A high resolution rate with a high reopen rate means tickets are getting closed, not fixed.

How Do You Measure AI Customer Service Performance?

Track AI accuracy through QA sampling, AI resolution rate, AI reopen rate compared to human reopen rate, CSAT on AI-handled tickets, and how cleanly the AI escalates when it should. Skip containment as a headline number. It tells you how much the AI touched, not how much it fixed.