The Only 7 Customer Service KPIs That Actually Matter

Most support dashboards track 15 to 20 metrics. Most of them are noise. Total ticket volume tells you how busy you are, not how good. Response time tells you how fast you reply, not whether you fixed anything. Here are the seven numbers that actually drive decisions.

1. Customer Satisfaction Score (CSAT)

After each interaction, ask: "How satisfied were you?" on a 1 to 5 scale. The percent who pick 4 or 5 is your CSAT. E-commerce benchmark: 75% to 85%. Below 70% means something is broken. Above 90% is rare at scale.

AI agents keep CSAT at 78% to 85% for the queries they handle, mostly because speed matters. A correct answer in 8 seconds scores higher than the same answer after a 6-minute wait. Keep surveys to one question and send them right after the chat. Target a 15% to 25% response rate.

2. Net Promoter Score (NPS)

"Would you recommend us to a friend?" on a 0 to 10 scale. Promoters (9 to 10) minus detractors (0 to 6) is your NPS. E-commerce benchmark: 30 to 50 is good, above 50 is strong, below 20 is trouble.

AI helps NPS indirectly through faster answers and 24/7 availability. Measure NPS quarterly, not monthly. Monthly surveys annoy customers and inflate negative scores.

3. First Contact Resolution (FCR)

What percent of issues get solved in one interaction? This might be the most important number because every unresolved first contact creates a second (or third) interaction, doubling your cost.

Human agents hit 70% to 75% FCR. AI agents with good knowledge bases reach 80% to 88% on their supported topics because they have instant access to all docs and never forget a policy. Do not count transfers as resolutions. Only count it if the customer did not come back about the same issue within 48 hours.

4. Average Handle Time (AHT)

Total time per interaction, including hold time, talk time, and follow-up work. Human benchmark: 6 to 10 minutes in e-commerce. AI agents: 15 to 45 seconds.

Here is the twist. After AI takes over the easy stuff, human AHT may go up, and that is fine. Humans now get the hard cases. A rising human AHT alongside a falling blended AHT means the system works. Do not push agents to rush calls just to hit a time target. A 5-minute call that creates a callback costs more than an 8-minute call that solves it.

5. Cost Per Resolution (CPR)

Total support spend divided by total resolved tickets. The number your CFO cares about most. Human-only: $8 to $14 in the GCC. Blended AI and human: $2 to $5. AI-only: below $2.

A company doing 15,000 tickets a month at $10 CPR spends $150,000. With AI handling 60% at a blended $4 CPR, that drops to $60,000. When calculating your baseline, include everything: training, management, QA, software. Otherwise the savings look smaller than they are.

6. Auto-Resolution Rate

What percent of conversations does the AI solve with zero human help? This is different from deflection, which counts conversations the AI touched, whether or not it fixed anything.

Target: 40% to 55% in the first 90 days, rising to 60% to 75% as you fill knowledge gaps. If you are under 30%, the problem is usually the docs, not the AI. Platforms like Oris AI track the difference between deflection and real resolution by measuring confirmed fixes through post-chat CSAT.

7. Escalation Rate

What percent of conversations go from AI to a human? Not all escalations are bad. Good escalations happen when the AI spots low confidence or negative sentiment. Bad escalations happen when the AI had the answer but failed to find it.

Target: 25% to 40% escalation, with under 10% being wrong escalations. A company starting at 45% in month one should reach 30% by month three. Do not make it hard for the AI to escalate. It is better to over-escalate slightly than to trap customers in a loop with an AI that cannot help.

Skip the Vanity Metrics

Total ticket volume is not useful on its own. It goes up when you launch a product and down when traffic drops. Response time matters only paired with resolution. A 30-second first reply that takes three chats to solve the issue is worse than a 2-minute reply that fixes it the first time. Measure what connects to outcomes. These seven do.