The customer support metrics that actually matter (and the ones that lie)

Most support dashboards track a dozen metrics and act on none, because a number by itself is easy to game and easy to misread. The five that actually matter — first response time, CSAT, resolution rate, deflection, and re-contact rate — each answer a real question about your support, and each can lie if you read it alone. The discipline is not collecting more numbers; it is reading the right ones in pairs, so they keep each other honest.

First response time: the one customers feel

First response time — how long a customer waits for any human or AI acknowledgement — is the metric your customers experience most directly, and the expectation depends entirely on channel. On live chat, the bar is seconds to a couple of minutes; on email, hours is normal and acceptable. The reason it matters so much is emotional, not logistical: a fast first response tells a worried customer they have been heard, even before the problem is solved. A customer will forgive a slow fix far more easily than a slow acknowledgement — silence is what turns a question into a complaint. But speed alone is a trap, which is why it must be read next to satisfaction.

CSAT: did the answer actually land

CSAT — the share of customers who rate their experience positively after a resolution — is the closest thing support has to a truth serum, and healthy teams tend to land around 80 to 85 percent satisfied, roughly 4.2 to 4.3 out of 5. It is the counterweight to every speed metric, because it catches the failure mode those reward: answering fast and unhelpfully. A support team can hit every speed target and still be quietly failing, and CSAT is the number that says so out loud. The caveat is response bias — angry and delighted customers answer surveys more than the contented middle — so read the trend over time, not any single week's score.

Resolution rate and deflection: the volume story

These two are about how much work actually gets done, and where. Resolution rate is the share of tickets you close; deflection (or containment) is the share resolved with no human at all, by your help center and AI. In 2026 the enterprise median for AI deflection sits near 40 percent, with the top quartile past 60 percent and the best-seeded setups reaching 70 to 90 percent on well-documented topics. Deflection is the most exciting number on this list and the most dangerous, because it is trivial to fake. An AI can "deflect" a ticket by frustrating someone into giving up — which looks identical to success on the dashboard. That is why deflection is meaningless without the metric that catches its lie.

Re-contact rate: the honesty check on everything

Re-contact rate — the share of "resolved" tickets where the same customer comes back within a few days — is the metric that keeps all the others honest, and it typically runs around one in ten. It is the antidote to vanity deflection: a ticket the AI "closed" that reopens 48 hours later was never resolved; it was deferred. Re-contact on AI-handled tickets runs slightly higher than on human-handled ones, which is exactly the signal you want to watch as you lean on automation. A resolution that does not stay resolved is not a resolution — it is a delay with better branding. Track it, or your deflection number is fiction.

It is worth internalizing why re-contact runs a little higher on AI-handled tickets than human ones: an AI is superb at the clear question and weakest at the ambiguous one it should have escalated, so the tickets it gets wrong tend to be the ones it should never have tried. That is not an argument against AI — it is an argument for confidence-gating, and re-contact is the number that tells you whether your gate is set correctly. Watch re-contact climb and you have found the exact place your automation is overreaching — which is far more actionable than a deflection score that only ever flatters you.

Read them in pairs, or they lie

The through-line of every metric above is that none of them can be trusted alone. First response time paired with CSAT stops you rewarding speed over usefulness. Deflection paired with re-contact stops you celebrating tickets you merely postponed. CSAT paired with volume stops a tiny, happy sample from flattering you. A single support metric is a headline; a pair of them is the story. The teams that measure well are not the ones with the most dashboards — they are the ones who never look at a number without asking which second number would expose it if it were lying.

The metrics that look important but lie

Just as valuable as knowing the five that matter is knowing which popular numbers to distrust. Raw ticket volume is the classic offender: it rises when you launch, rises when you grow, and rises when your product confuses people — three completely different stories that one number cannot tell apart. Average handling time is another, because a low figure can mean ruthless efficiency or rushed, unhelpful replies, and you cannot tell which without CSAT beside it. "Tickets closed per agent" rewards speed and closing rather than solving, and quietly punishes the teammate who takes the hard cases. Even a single glowing CSAT week can be a tiny, self-selected sample rather than a real improvement. A metric that measures activity instead of outcomes will always reward looking busy over being useful — and a team measured on the wrong number becomes very good at gaming it. Before any figure goes on a dashboard, ask what behaviour it will encourage when someone is trying to make it go up, because they will.

One last discipline holds all of this together: keep the time range still. Support numbers swing with weekends, launches, and seasonality, so a metric is only meaningful against the same window in the previous period. A CSAT that "dropped" may just be a Monday; a deflection rate that "jumped" may be a quiet holiday week of easy questions. A metric without a comparison period is a number without a verdict — compare this thirty days against the last thirty, and read the trend, not the point.

A minimal dashboard that tells the truth

You do not need twelve charts. You need five numbers, read in pairs, on a consistent time range:

• First response time, next to CSAT — fast and useful, not just fast.
• Deflection, next to re-contact rate — work removed, not work postponed.
• CSAT trend over time, not any single week, to survive response bias.

That is enough to know whether your support is genuinely improving or just looking busier. Everything else is a footnote to these five.

We build the tooling for exactly this, so here is the honest version. Iris ships Reports and Insights that track these numbers the way this post argues you should: first response and resolution times, CSAT with its positive and negative reasons clustered, and — the pairing that matters most — deflection alongside the questions the AI could not answer, so you see work removed and work merely deferred side by side. Insights surfaces your knowledge gaps as a ranked list, turning "the AI failed here" into "write this article next." It is priced in rupees with a free plan, and the AI is metered at a low, flat, published per-answer rate. If you want a support dashboard that keeps you honest rather than one that flatters you, that is the instinct we built it on.