Customer Support Metrics That Actually Matter in 2026

IN THIS ARTICLE

1. 01The difference between CSAT, NPS, and CES
2. 02Mean time to resolution (MTTR): the metric that lies if you are not careful
3. 03First contact resolution (FCR): the most underrated metric
4. 04Ticket volume trends: measuring demand correctly
5. 05Agent utilisation: the metric most teams ignore
6. 06What AI unlocks in terms of measurement
7. 07Building a metrics framework that drives decisions

Support leaders report on a lot of numbers. Ticket volume, resolution time, CSAT, NPS, FCR, agent utilisation. But many teams report metrics that look good in a board deck without meaningfully improving either customer experience or team health. The problem is not the data: it is knowing which numbers actually change when your support operation improves.

This guide cuts through the noise. It covers the metrics that matter, the ones that mislead, and the new measurements that become possible when AI enters the picture.

The difference between CSAT, NPS, and CES

Three customer satisfaction surveys dominate support measurement, and they measure meaningfully different things. Choosing the wrong one for your context gives you data that does not inform the decisions you actually need to make.

CSAT (Customer Satisfaction Score)

CSAT measures satisfaction with a specific interaction. "How satisfied were you with your support experience today?" Rated 1 to 5, CSAT tells you about the individual ticket resolution. It is transactional and immediate. A high CSAT means the customer was satisfied with this particular experience.

Use CSAT when: you want to measure ticket-level quality, identify which agents or issue types are producing poor experiences, and catch problems immediately after they occur. CSAT is your day-to-day health indicator.

NPS (Net Promoter Score)

"How likely are you to recommend us to a colleague or friend?" Rated 0 to 10, NPS measures overall brand loyalty and is not a support-specific metric. A customer can have an excellent CSAT score on their latest ticket and still be a detractor on NPS because of pricing, product limitations, or other factors entirely unrelated to support.

Do not use NPS as a support quality metric. It will not tell you whether your agents are performing well. Use it at the company level to understand customer loyalty, but do not hold support teams accountable to it: too many variables are outside their control.

CES (Customer Effort Score)

"How easy was it to resolve your issue today?" Rated 1 to 7, CES measures friction. Research from Gartner found that reducing customer effort is more strongly correlated with loyalty than delighting customers. A support interaction that is low-effort but not exceptional is more valuable than an exceptional one that was frustrating to get to.

CES is particularly valuable for support teams because it directly measures what you can improve: reducing the number of contacts needed to resolve an issue, making information easier to find, and eliminating unnecessary steps in the resolution process.

Mean time to resolution (MTTR): the metric that lies if you are not careful

MTTR is one of the most-reported metrics in support. It is also one of the most misinterpreted. A low MTTR looks good. But a low MTTR combined with a high reopen rate means your team is closing tickets prematurely: not resolving them.

Always track MTTR alongside reopen rate. A good benchmark: reopen rate below 8% indicates resolutions are genuine. Above 15% suggests systemic quality issues. Some teams create artificial low MTTRs by closing tickets after one response and letting the customer re-open. This looks good in dashboards and destroys customer trust.

Also segment MTTR by ticket category. Overall MTTR can look healthy while hiding a category with 48-hour average resolution times. Category-level data is where you find the problems.

First contact resolution (FCR): the most underrated metric

First contact resolution: the percentage of tickets fully resolved in a single interaction: is the metric most directly correlated with both customer satisfaction and operational efficiency. Every ticket that requires a follow-up contact represents a failure: the customer had to come back, the agent has to find context again, queue capacity is consumed twice for one issue.

• Industry benchmark for FCR: 70 to 75% is considered good for most industries. Above 80% is excellent.
• Low FCR causes: agents lack authority to resolve (they need approval), agents lack information (knowledge base gaps), tickets are routed to the wrong specialist, customers are given partial answers to avoid complexity.
• How to improve FCR: expand agent authority and decision-making latitude, invest in knowledge base quality, improve routing accuracy, and use AI drafts that surface complete answers from the knowledge base.

Ticket volume trends: measuring demand correctly

Raw ticket volume is a vanity metric if you are not normalising it by customer base size or interaction count. A growing company will have more tickets. That is not necessarily a problem. Tickets per customer or tickets per order is a more meaningful signal.

More useful is the ticket volume by category trend. If billing tickets are growing while order tickets are flat, something specific is happening in billing. If your general FAQ category is growing, your knowledge base is not keeping pace with your product.

The ideal trend: overall ticket volume grows more slowly than your customer base (meaning your product is becoming more self-service) and the distribution shifts toward complex, high-value cases as AI handles the simple ones.

Agent utilisation: the metric most teams ignore

Agent utilisation measures what percentage of an agent's available working time is spent on active support work. Teams rarely track this, but it is critical for workforce planning and for understanding whether your routing and triage systems are working.

Target utilisation rate for support agents: 70 to 80%. Below 70% means agents have significant idle time: a staffing or routing problem. Above 80% means agents are consistently at capacity, which increases error rates and agent burnout and means you are one sick day from an SLA crisis.

What AI unlocks in terms of measurement

Manual support operations have measurement blind spots. The most useful data is often trapped inside ticket content: the sentiment, the specific issue type, the language the customer used. Extracting this at scale requires AI.

• Sentiment analysis at scale: AI can classify every ticket by sentiment (positive, neutral, negative, frustrated, urgent) automatically, giving you a real-time read on customer emotional state across your full ticket volume.
• Topic clustering: AI can surface emerging issue categories before they become visible in traditional category reports. A new bug, a confusing feature, or a policy problem will show up in AI topic clustering days before it registers in manual categorisation.
• AI handling rate: the percentage of tickets fully resolved by AI: a metric that only exists with AI in the loop. A good target is 50%+ for teams with a well-configured knowledge base.
• Draft acceptance rate: if your agents use AI-drafted replies, the percentage they accept as-is versus edit significantly versus reject tells you how well-calibrated the AI is to your brand and knowledge base.
• Escalation reason analysis: AI can categorise why tickets escalated from automated to human handling, identifying systematic gaps in knowledge base coverage or routing logic.

Delyt's analytics dashboard surfaces all of these metrics in real time, including the AI-specific ones. You can explore the analytics capability on the features page.

Building a metrics framework that drives decisions

The goal of metrics is not to report numbers: it is to inform decisions. For each metric you track, you should have a clear answer to: "What decision does this metric help me make?" If you cannot answer that question for a metric, it is probably a vanity metric.

A practical support metrics framework for 2026 should cover four areas: customer experience (CSAT, CES, FCR), operational efficiency (MTTR, reopen rate, agent utilisation), demand signals (ticket volume by category, trend over time), and AI performance (AI handling rate, draft acceptance rate, escalation reasons).