12 CX Metrics Worth Tracking in 2027 (and 6 to Stop Reporting)

Most CX dashboards are honest about the wrong things. They report, to two decimal places, how busy the team was — tickets closed, calls handled, average seconds per contact — and stay conveniently vague about whether the customer got what they came for. In the agentic era, when an AI agent can qualify, resolve, and route first contact, that gap stops being a reporting quirk and becomes a strategic blind spot. This is a metrics audit: twelve numbers worth keeping, six worth retiring, and one rule underneath all of them.

The rule: measure the outcome, not the activity.Every metric on the keep list answers “did the customer get what they needed, and what did that cost?” Every metric on the cut list answers “how much did we do?” — a question that looks like performance and quietly rewards the wrong behavior. When automation enters the picture, activity metrics don’t just mislead; they actively point you the wrong way.

Why the old dashboard breaks under automation

A human team and an activity metric are roughly aligned: more tickets closed usually did mean more customers helped. Point an AI agent at the same scoreboard and the alignment snaps. An agent optimized for deflection will happily “resolve” a conversation by ending it. An agent optimized for handle time will rush the exact caller who needed a human. The metric you choose becomes the behavior you get, instantly and at scale — which is why the audit matters more now than it did when the front line was entirely people.

Under automation, a metric isn’t a description of behavior. It’s an instruction for it.

The 12 to keep

Grouped by the question they answer. None of these are new inventions — the shift is treating them as the headline instead of the footnote.

Four of these deserve a sentence more, because they’re the ones teams most often skip.

• Time-to-first-response earns the top slot on evidence. Response speed on inbound demand is repeatedly the difference between a won and a lost customer — and a large share of inbound calls in home-services still go unanswered entirely (Invoca’s industry data, 2024, vendor-published). You cannot resolve what you never answered.
• Escalation qualityis the metric almost no one reports, and it’s the one customers feel most. Consumer research consistently finds people are wary of AI support — a majority tell researchers they want companies to be more careful with it (Zendesk CX Trends, 2025). What they hate isn’t the AI; it’s a cold transfer that makes them repeat the whole story. Score the handoff, not just the headcount.
• Cost per resolved outcomeis the number that survives contact with a CFO. It’s the through-line to our cornerstone on outcome-based CX economics, and it’s why the cut list below exists at all.
• After-hours capture rateis the metric that makes the business case obvious. Most demand doesn’t politely arrive between nine and five, and a missed call after hours is rarely a call that comes back — the customer simply books with whoever answered. Tracking the share of nights-and-weekends contacts you actually resolve turns “we’re always on” from a slogan into a number you can put next to recovered revenue.

The 6 to stop reporting

Retiring a metric doesn’t mean you never look at it. It means you stop putting it at the top of the deck, stop setting targets against it, and stop letting it drive decisions. Each of these fails the same test: it can improve while the customer experience gets worse.

1. Raw ticket / call volume as a goal. It measures demand, not performance. Volume going up can mean marketing worked or that your product broke. As a headline it rewards being busy.
2. Deflection rate. The most dangerous number on the old dashboard. It counts people you kept away from a human — and scores a frustrated customer who gave up exactly the same as one who was helped. Replace it with containment-with-satisfaction.
3. Average handle time as a target. Fine as an operations input, poison as a goal. Optimizing for short conversations punishes the complex, emotional contacts that most needed the time. Independent analysis of AI in support already warns the real savings are modest — closer to 20–35% net, not the 60–80% in vendor headlines (NBER, “Generative AI at Work,” 2023) — so shaving seconds is chasing the wrong lever.
4. Tickets-per-agent-per-hour. A productivity metric from the staffing era. Once an agent handles the routine majority, the humans are left with the hard cases — so this number should fall, and punishing it for falling is backwards.
5. Monthly-averaged CSAT. Not wrong, just blunt. A single rolling number hides which interactions failed and buries the signal you could act on. Keep satisfaction — measure it at the moment, per contact.
6. Vanity self-service “resolution.”Marking a chat resolved because it ended isn’t resolution. True self-service resolution today is far lower than dashboards imply — often in the mid-teens (Lorikeet self-service research, 2025, vendor-published). Verify with the reopen rate or don’t count it.

Making the switch without a six-month project

You don’t need a data warehouse rebuild. You need to change what leads the conversation.

1. Pick the three keep-list metrics closest to revenue — time-to-first-response, cost per resolved outcome, escalation quality — and put them at the top of the next report.
2. Move deflection, handle time, and raw volume into an appendix. Keep them as inputs; strip them of targets.
3. Add one honesty check — the reopen / repeat-contact rate — so nobody can inflate resolution by closing early.

The tell that it worked: your reviews stop being about how hard the team worked and start being about what customers actually got. That’s a different meeting, and a better one.