Transform complex support workflows
20 Customer Service Metrics to Track in 2026 (+ Formulas)
Most customer service metrics were built for a support operation made entirely of people. Add an AI agent to the queue and the same formulas start averaging two very different operations into one number. A rising first contact resolution rate can mean AI is resolving routine requests cleanly, or that human agents are marking harder cases resolved too soon, and a blended figure cannot tell you which.
Gartner predicts that by 2029, agentic AI will autonomously resolve 80% of common customer service issues without human intervention, leading to a 30% reduction in operational costs. Some support teams already run close to that mix. At Aptean, Enjo resolves 300K+ cases every year, volume equivalent to 120 agents handled by AI. Report one average across an operation like that and it describes neither half of it.
Customer service metrics for 2026: all 20, with formulas and benchmarks, plus how to measure AI and human resolution separately. Start free with Enjo
What Are Customer Service Metrics?
Customer service metrics are the measurements a support team uses to judge how well it resolves requests. They split three ways: operational data your systems record, experience data customers report, and resolution data that separates what AI resolved from what people resolved. Common examples include CSAT, first contact resolution, average resolution time, and autonomous resolution rate.
Anyone asking what are customer service metrics in 2026 is really asking about that third axis, because it is the one their current dashboard does not have. Until an AI agent joined the queue, every number on a support dashboard described the same population of resolvers. Now two resolvers share the work, and a single average describes neither.
Customer support metrics and customer service performance metrics describe the same set of numbers, and the labels get used interchangeably. What matters is not the label but whether the scorecard can tell you which resolver produced the result.
A useful scorecard should not depend on one number. CSAT can show customers are satisfied without telling you whether the issue was resolved on first contact. First Response Time can show a customer got a quick reply without telling you whether that reply solved anything.
The strongest measurement framework connects several numbers, so leaders see both the customer outcome and the operational cause.
What Counts As A Customer Service Metric
By definition, a customer service metric is any measurable value the support operation produces. Ticket volume, First Response Time, CSAT, and reopen rate all qualify.
Not every one of them belongs on a dashboard. Tracking customer service metrics well means picking the handful that explain each other, then reading them together.
A customer service scorecard built on satisfaction numbers alone cannot tell you why satisfaction moved. Pair every experience metric with a resolution-quality metric and an efficiency metric.
It is important to note that, metrics and targets are different things, and the customer service KPIs worth putting on a dashboard are a much shorter list.
Operational Data, Experience Data And Resolution Data
Every metric on this page sits on one of two sides. Operational data is what your systems record: ticket counts, response times, resolution times, costs. Experience data is what customers report: satisfaction, effort, willingness to recommend.
Operational data tells you what happened. Experience data tells you how it felt. Reading either one without the other is how a support team ends up confidently wrong about its own performance.
The five families map onto that axis like this:
Resolution quality straddles both sides deliberately. A ticket can be marked resolved in your system while the customer still considers the problem open. Reopen rate is where that disagreement surfaces, which is why it belongs on both sides of the axis rather than one.
Why One Metric Cannot Carry A Support Scorecard
CSAT captures how customers felt about a specific interaction. It does not explain what caused the result.
A customer can give a positive rating after waiting hours for a response. A ticket can close quickly and reopen because the resolution did not hold. AI can handle more conversations while leaving human agents a harder queue.
Customer service metrics therefore need to cover five dimensions:
- Customer experience: CSAT, NPS, CES
- Resolution quality: FCR, reopen rate, escalation rate
- Operational efficiency: FRT, ART, AHT, ticket volume, backlog
- Self-service and AI: self-service resolution, autonomous resolution
- Business impact: cost per resolution, retention, churn
The 20 Customer Service Metrics At A Glance
Every metric below, with its formula, the industry benchmark where a credible one exists, and whether it needs to be split by resolver once AI is in the queue. Eight of the twenty have no credible published benchmark, and saying so is more useful than inventing one.
The 20 Customer Service Metrics Worth Tracking
Customer Service Satisfaction Metrics
1. Customer Satisfaction Score (CSAT)
What it measures: how satisfied a customer was with a specific support interaction. A short survey asks customers to rate satisfaction on a numerical scale.
Why it matters: CSAT gives support leaders direct feedback on the customer experience.
Benchmark: SQM Group puts the contact centre average at 78%, with 85% and above as its top performance band, measured across more than 500 North American call centres.
What to watch: segment by channel, issue type, team, and resolution method. If AI handles a growing share of requests, compare AI-resolved CSAT against human-resolved CSAT rather than one blended score.
2. Net Promoter Score (NPS)
What it measures: customer loyalty, via how likely a customer is to recommend the company or product. Net Promoter Score was created in 2003 by Fred Reichheld, a partner at Bain & Company, and introduced in his Harvard Business Review article "The One Number You Need to Grow". Bain continues to develop the methodology through the Net Promoter System.
Customers group into:
- Promoters: 9 to 10
- Passives: 7 to 8
- Detractors: 0 to 6
Why it matters: NPS is a relationship-level metric. Read it alongside interaction-level measures such as CSAT and CES.
Benchmark: Retently's 2026 analysis puts the median at 41 for software companies, with anything above 30 read as strong and above 70 as exceptional.
What to watch: NPS is not a substitute for support-specific metrics. A customer can be loyal to a product and still have a poor support interaction.
3. Customer Effort Score (CES)
What it measures: how much effort a customer spent getting an issue resolved. A common question is how easy it was to get the issue resolved.
How to measure it: use a consistent numerical or descriptive scale and track the score over time.
Why it matters: CES exposes friction satisfaction scores miss. A customer can eventually get the right answer after repeating information, switching channels, or contacting support three times.
What to watch: read CES alongside FCR, escalation rate, and resolution time.
The metric comes from "Stop Trying to Delight Your Customers" by Matthew Dixon, Karen Freeman and Nick Toman, published in Harvard Business Review in 2010. Across more than 75,000 customer interactions, the authors found that exceeding expectations does little to build loyalty, and that Customer Effort Score predicts loyalty better than satisfaction measures or Net Promoter Score.
Resolution And Service Quality Metrics
4. First Contact Resolution (FCR)
What it measures: the share of issues resolved during the first interaction, with no follow-up.
Why it matters: FCR shows whether customers get complete resolutions without repeated contacts.
Benchmark: SQM Group's 2024 study of more than 500 North American call centres puts the average at 69%, with a range of 43% to 88%. SQM's top band starts at 80%, which roughly 5% of centres reach.
What to watch: never read FCR alone. A ticket can close on first contact and still be wrong. Read it alongside reopen rate and CSAT to see whether faster closure also means better resolution.
5. Ticket Reopen Rate
What it measures: the share of resolved tickets customers reopen because the original resolution did not hold.
Why it matters: reopen rate is the quality check on every resolution metric. A falling Average Resolution Time looks positive while masking tickets returning to the queue.
What to watch: segment by issue type, channel, queue, and resolution method. Split AI and human results only once the sample size makes the comparison meaningful.
6. SLA Compliance
What it measures: how consistently support meets defined response or resolution commitments.
Why it matters: a healthy average response time does not guarantee important queues are meeting their commitments.
Benchmark: the Freshworks 2025 Freshservice Benchmark Report, drawn from more than 187 million tickets across 10,551 organisations, puts resolution SLA adherence at 96.16% on IT service desks.
What to watch: break SLA performance down by priority, channel, queue, and issue type.
Human Support reporting inside Insights covers volume, SLAs, CSAT, responsiveness, and team metrics.
Response And Resolution Time Metrics
7. First Response Time (FRT)
What it measures: the time between a customer submitting a request and receiving the first response.
Why it matters: FRT measures customer waiting time, and it matters more once AI and humans share the workload. AI responds immediately to eligible requests while human agents work a queue. One blended FRT hides that difference.
Benchmark: the Freshworks 2025 benchmark puts average first response time at 9.36 hours on IT service desks. No credible cross-industry average exists per channel, so treat any published per-channel target as a vendor's top performers rather than a norm.
What to watch: segment by resolution method and channel. Read it alongside Average Resolution Time, because a fast first reply does not mean a fast resolution.
8. Average Resolution Time (ART)
What it measures: elapsed time between a request opening and the issue being resolved.
Benchmark: Freshworks reports 21.96 hours on IT service desks and 29.08 hours on wider employee service teams in its 2025 report.
What to watch: compare ART against FCR, reopen rate, and CSAT. Reducing ART only helps while resolution quality holds steady.
9. Average Handle Time (AHT)
What it measures: the active time an agent spends handling an interaction.
Why it matters: AHT and ART are the two most commonly cited customer service performance metrics for operational efficiency. AHT specifically helps leaders understand human-agent workload and capacity.
What to watch: do not turn AHT into a standalone speed target. If agents close conversations faster but CSAT falls or reopens, the number is moving the wrong way.
Agent Assist supports this part of the workflow, surfacing case summaries, reply suggestions, knowledge, and sentiment in the agent's existing helpdesk.
Support Workload And Operational Metrics
10. Ticket Volume
What it measures: the number of support requests received in a defined period.
Track it by channel, issue type, product, customer segment, time period, and resolution method. Pull the counts from your AI ticketing or helpdesk system.
Why it matters: rising volume signals a product issue, service incident, documentation gap, or a change in customer behavior.
What to watch: lower volume is not automatically good. Check whether customers are resolving issues through self-service or whether requests are going unanswered.
11. Ticket Backlog
What it measures: the number of unresolved support requests waiting for action.
Why it matters: backlog shows accumulated demand rather than incoming demand alone. It is one of the few customer support metrics that measures pressure rather than throughput.
What to watch: read backlog alongside ticket volume, FRT, ART, and SLA compliance. Growing backlog on stable incoming volume points to a capacity, routing, or process problem.
12. Ticket Escalation Rate
What it measures: how often a request moves from its initial resolution path to another team or a human agent.
Why it matters: escalation is not automatically a failure. Some requests need human judgment. For AI resolution, the useful question is why the escalation happened: missing knowledge, a missing action, low confidence, or a request type that should always reach a human.
Benchmark: none published. The closest adjacent figure is SQM Group's call transfer rate, 19% on average with 15% or below rated good, but a transfer between agents is not an escalation out of AI resolution and the two should not be compared.
What to watch: a rising escalation rate on stable ticket types usually signals a knowledge gap. Escalation (under AI Agents) defines when, how, and where requests hand off to humans, and what context travels with the handoff.
Self-Service And AI Customer Service Metrics
13. Self-Service Resolution Rate
What it measures: the share of eligible requests resolved through self-service, with no human agent involved.
Why it matters: more useful than page views or help center searches. A customer reading an article is not necessarily a customer whose issue was resolved. Coverage depends on how well your AI knowledge base maps to real request patterns.
Benchmark: Gartner surveyed 5,728 customers and found only 14% of customer service issues are fully resolved in self-service, rising to 36% for issues customers described as very simple.
What to watch: track it alongside ticket volume and escalation. If self-service climbs while customer outcomes hold or improve, the team is removing demand rather than shifting it.
Help Center is a requester-facing self-serve portal. Article Generation (under Help Center) builds articles from a URL, connected documents, or helpdesk ticket patterns, and the Self-Improving Loop drafts new articles from resolved conversations for review.
14. Autonomous Resolution Rate
What it measures: the share of eligible support requests resolved by AI with no human intervention.
Why it matters: this shows how much eligible work AI resolves without transferring the conversation.
The denominator matters. Including requests that were never appropriate for AI resolution understates automation performance.
What to watch: pair it with CSAT, reopen rate, escalation rate, and resolution time. A higher automation rate only counts while the resulting resolutions stay useful and durable.
Current practice increasingly separates real automated resolution from conversations that merely ended or were contained. A conversation that ended because the customer gave up is not a resolution, and counting it as one inflates every downstream number.
15. AI-Assisted Resolution
What it measures: the contribution AI makes to a request a human ultimately resolves. That contribution includes case summaries, reply suggestions, knowledge retrieval, ticket autofill, sentiment detection, translation, and draft refinement.
Why it matters: autonomous resolution and assisted resolution are different outcomes. If AI resolves 30% of eligible requests autonomously and assists agents on another 40%, reporting only autonomous resolution misses most of the operational effect.
What to watch: compare assisted work against AHT, ART, FCR, CSAT, and reopen rate.
Assisted volume is also set to grow for reasons outside your control. Gartner forecasts that by 2028, regulatory changes related to AI will increase assisted service volume by 30%, as rules mandating easy access to human agents encourage customers to request a human by default. If that holds, assisted resolution stops being a secondary metric.
Business Impact Metrics
16. Cost Per Resolution
What it measures: the support cost required to resolve a customer request.
Why it matters: cost per resolution connects support performance to the business case for automation. It is also the number most likely to move against you. Gartner forecasts that by 2030, cost per resolution for generative AI will exceed $3, higher than many B2C offshore human agents. The drivers are structural: data centre costs, AI vendors pivoting to profitability, and heavier token use on complex requests.
What to watch: where the data supports a fair comparison, segment by human-only, AI-assisted, and AI-resolved work. The goal is not pushing every request toward the cheapest path. Complex cases need human judgment. The question is whether routine work resolves efficiently without reducing service quality.
17. Cost Per Contact
What it measures: the average cost of a single customer interaction, across phone, chat, email, and every other channel.
Why it matters: cost per contact prices the interaction. Cost per resolution prices the outcome. The two move independently, and the gap between them is where inefficiency hides.
What to watch: read them together. Falling cost per contact alongside flat cost per resolution means interactions got cheaper without getting more effective, usually because issues now take more contacts to close.
18. Customer Retention Rate And Churn Rate
What it measures: the share of customers who stay over a defined period, and the share who leave.
Why it matters: retention connects support quality to revenue. High churn often traces back to product quality, support experience, or both.
Benchmark: SaaS Capital's 2025 research puts median net revenue retention at 102% for private SaaS companies in the $25,000 to $50,000 annual contract band, with the top quartile at 111%.
What to watch: segment by customer type, product line, and region. Read churn against escalation rate, because accounts generating repeated escalations churn before the numbers say so.
19. Customer Sentiment
What it measures: the emotional tone in customer messages, tickets, and chats, classified as positive, neutral, or negative.
How to measure it: apply natural language processing to support text, then track the classification over time by agent, product, or issue type.
Why it matters: sentiment catches frustration that satisfaction scores miss, because most customers never fill in a survey.
What to watch: monitor sentiment shifts after product releases and policy changes. Use negative sentiment to prioritise cases rather than to score agents.
20. Agent Performance Metrics
What it measures: individual agent outcomes, including agent-level CSAT, resolution rate, first contact resolution, handle time, and workload balance.
How to measure it: pull agent-level data from your helpdesk, then read it alongside qualitative review rather than on its own.
Why it matters: agent metrics identify coaching needs and workload imbalance. They also show whether AI is removing routine work or leaving agents a harder queue.
What to watch: never rank agents on speed alone. Once AI handles routine volume, human agents inherit harder cases, so raw handle time rises even as performance improves.
Why Self-Service Numbers Overstate Resolution
Vendors routinely publish deflection figures in the sixties and seventies. Gartner's survey of 5,728 customers found that only 14% of customer service issues are fully resolved in self-service. Both numbers can be true at once, because they measure different things.
A vendor deflection figure usually counts sessions that did not reach an agent. Gartner measured whether the customer considered the issue resolved. The gap between those two definitions is the single largest source of overstated automation performance in support reporting.
This is why the denominator discipline in the two metrics above matters more than the headline number. Before comparing your self-service or autonomous resolution rate against anything published, check whether the published figure counts contained sessions or confirmed resolutions. If it counts contained sessions, it is not comparable to a resolution rate.
How AI Changes The Way These Metrics Should Be Measured
Traditional support dashboards were built around human teams. AI introduces a second resolver, so several formulas need an added layer of reporting.
The gap this creates is already measurable. In McKinsey's State of Customer Care survey of 440 customer care leaders and executives, published in February 2026, 40% of top-performing organisations reported significantly improved customer experience scores over the previous 12 months, against 12% of laggards. Same metrics, very different readings.
Why Blended Human And AI Numbers Hide Performance
Consider First Response Time. AI responds immediately to an eligible request. A human agent may need minutes or longer depending on queue volume. If AI starts handling more simple requests, blended FRT improves even while the human queue slows down.
The same problem appears in FCR, CSAT, ART, reopen rate, and cost per resolution. A single number makes two very different support operations look identical. This is where customer support metrics for SaaS teams break down first, because the volume mix shifts fastest there.
Measure AI And Human Resolution Separately
For metrics both AI and humans influence, report the outcome by resolver wherever the data supports it.
A blended metric hides both where performance is improving and where the human queue is getting harder.
Measure Autonomous Resolution Separately From Assisted Resolution
Autonomous resolution and assisted resolution answer different questions, so they need different measures.
Autonomous resolution gets measured on escalation rate, resolution quality, and the share of eligible work closed with no human involvement. Assisted resolution gets measured on its effect on the human path: AHT, ART, FCR, CSAT, and reopen rate.
Reported that way, a CS leader can answer two separate questions. How much work is AI resolving on its own? And how much faster is the team resolving what escalates?
How Enjo Platform Helps Support Teams Improve These Metrics
Enjo is an AI Support Agent that resolves support requests end to end, including actions in connected systems. Enjo Platform is the system built around it: Enjo handles the first line, your team handles exceptions in Inbox, Help Center covers self-service, and Insights reports the split.
Resolving Requests Autonomously
Enjo resolves eligible customer requests using approved knowledge and connected actions. Several named modules support that workflow:
- Knowledge (under AI Agents): the knowledge layer grounding every AI resolution.
- AI Actions (under AI Agents): retrieves data and executes operations in external systems.
- AI Flows (under AI Agents): multi-step orchestration with logic, fallbacks, and auditability.
- Escalation (under AI Agents): defines when and where a request moves to a human.
- Guardrails (under AI Agents and Security): compliance and policy controls.
This module structure matters for measurement because AI resolution is not limited to producing an answer. Where a request requires an action in another system, the workflow includes that action before the request counts as resolved.
Agent Assist For Faster Human Resolution
Agent Assist is a capability inside the helpdesk supporting agents who work escalations. It provides case summaries, reply suggestions, reply refinement, knowledge assistance, sentiment, ticket autofill, and AI translation.
Agent Assist capabilities connect directly to the human side of FRT, ART, AHT, FCR, CSAT, and reopen rate.
Agent Assist runs inside Salesforce Service Cloud, Zendesk, Jira Service Management, ServiceNow, and Freshservice, per Enjo's current integration coverage.
Help Center For Self-Service Resolution
Help Center gives customers a self-service layer before a request reaches a queue. Article Generation builds articles from a website URL, connected documents, or helpdesk ticket patterns. The Self-Improving Loop escalates unanswered portal questions and drafts new articles from resolved conversations for review.
Help Center creates a measurable chain from self-service resolution to ticket volume to human workload. If self-service resolution rises while customer outcomes stay healthy, routine demand is being resolved earlier in the journey.
Insights For Measuring Support Performance
Enjo AI Insights is the measurement layer, reporting across AI performance, human support, service trends, resolution outcomes, and ROI.
The operating loop is short:
- Identify the underperforming metric.
- Segment it by issue type, channel, or resolution method.
- Find the workflow causing it.
- Automate or assist the repeatable portion, using the customer service automation patterns that fit the workflow.
- Measure again.
How To Choose The Right Customer Service Metrics
You do not need every metric on this page on one screen. Metrics for customer service department reporting earn their place by changing what the team does next.
Work through this checklist when deciding what a customer service metrics dashboard should show:
- Does it explain something another metric cannot? Two numbers measuring the same thing add work rather than insight.
- Can you segment it by resolver? Any metric AI and humans both influence needs the split before it goes on a dashboard.
- Is it paired with a quality check? Speed metrics need FCR and reopen rate beside them.
- Does the denominator hold up? Autonomous resolution rate needs a clearly defined eligible population.
- Can you see it in one place? If AI resolves in Slack, tickets escalate to Zendesk or Jira, and self-service sits in a third tool, one number means three partial numbers.
- Is the sample big enough? Splitting reopen rate by resolver on low volume produces noise rather than signal.
- Does it ladder to a business outcome? Cost per resolution, retention, and churn connect operations to the P&L.
- Is the definition written down? FCR, resolution, and escalation mean different things in different tools, and undocumented definitions drift.
One more decision most guides skip: which channels you segment by. Most metrics lists put social media response time on the dashboard because they are written for consumer contact centres. Your demand probably does not concentrate there.
Segment by the channels your requests actually arrive through. For most support teams that means Slack, Microsoft Teams, website chat, a Help Center portal, and email. Social response time belongs on a consumer scorecard, and putting it on yours usually means measuring a channel that carries a rounding error of your volume.
The exception is shared-channel support. If customers reach you in a shared Slack channel, that channel needs its own FRT and resolution-time line, because expectations there run closer to chat than to email.
The strongest scorecard connects customer outcome to resolution outcome to operational outcome to automation outcome to business outcome. For a deeper view of where automation actually moves these numbers, see how AI support agents resolve requests end to end.
What Support Metrics Should A SaaS Company Track?
Support for a product sold to organisations changes the metric set. Accounts matter more than individual requests, one frustrated admin can put a whole contract at risk, and volume concentrates around releases, migrations, and renewals rather than spreading evenly.
A SaaS team should track these seven first:
- Autonomous resolution rate, because routine product questions scale faster than headcount.
- Escalation rate by account, because a single account generating repeated escalations is a churn signal before it is a support signal.
- First contact resolution, read against reopen rate, because admins escalate internally when an answer does not hold.
- Average resolution time by severity, because a blocked production environment and a how-to question are not the same queue.
- Ticket volume per account, normalised for seats, because raw volume tracks growth rather than health.
- Self-service resolution rate, because documentation gaps show up here before they show up in ticket volume.
- Cost per resolution, segmented by AI-resolved and human-resolved, because that ratio is the whole automation business case.
Consumer support metrics like social media response time matter far less for a SaaS team than escalation quality and account-level patterns.
Frequently Asked Questions
What are the 5 key customer service metrics?
The five that carry most support scorecards are CSAT, first contact resolution, first response time, average resolution time, and, for any team running AI in the queue, autonomous resolution rate. The first four cover experience, quality and speed. The fifth is the one traditional lists still leave out, and without it a scorecard cannot tell you which resolver produced the other four numbers.
What is a good CSAT score?
SQM Group puts the contact centre average at 78%, with 85% and above as its top performance band, across more than 500 North American call centres. Read your own score by segment rather than against the average: AI-resolved and human-resolved CSAT usually differ, and a blended score in the low eighties can hide a human queue performing well below it.
What is autonomous resolution rate?
Autonomous resolution rate is the share of eligible support requests an AI agent resolves with no human involvement, calculated as requests resolved by AI divided by total eligible requests, times 100. The denominator decides whether the number means anything. Requests that were never appropriate for AI resolution belong outside it, and sessions that simply ended without the customer getting an answer are not resolutions.
What is the difference between customer service metrics and KPIs?
Metrics are everything your support operation measures. KPIs are the short list you are accountable for hitting. Ticket volume is a metric on every team; it becomes a KPI only where reducing it is an explicit goal. The practical test is whether a number changes what the team does next, and the customer service KPIs worth reporting upward are usually six to eight of the twenty metrics here.
How do you measure customer service performance when AI resolves part of the queue?
Split every shared metric in two, once for AI resolution and once for human resolution. First response time, resolution time, first contact resolution, CSAT, reopen rate, escalation rate and cost per resolution are all customer service performance metrics with separate AI and human readings, and a blended average describes neither. Then measure autonomous resolution and AI-assisted resolution separately, because they answer different questions.
How many customer service metrics should a team track?
Six to eight on the reported dashboard, with the rest available for diagnosis. Every metric on the dashboard should explain something the others cannot, carry a quality check beside it if it measures speed, and be splittable by resolver if both AI and humans influence it. Tracking twenty numbers and acting on none of them is a more common failure than tracking too few.
