From Deflection to Resolution: Rewriting Your Support Metrics

Open almost any support dashboard and the biggest number on it is deflection rate — the share of contacts that never reached a human. It’s a comfortable metric because it always goes up and to the right. It’s also the wrong one. Deflection measures how many people you kept away, not how many you actually helped, and those two numbers can move in opposite directions.

This is a migration guide. If your team still optimizes for volume avoidance, the goal here is to swap in definitions and targets that measure resolution — and to give you the language to explain the change to a CFO who has spent three years watching the deflection line climb. One argument, start to finish: you get what you measure, so measure whether the problem got solved.

What deflection actually hides

A deflected contact is one that didn’t escalate to an agent. That definition treats two very different outcomes as identical: the customer who got a real answer and left satisfied, and the customer who gave up. Both count as wins. That’s the flaw — a rage-quit and a resolution look the same on the chart.

And the abandonment problem is not small. Genuine self-service resolution today is far lower than most dashboards imply — independent research from Gartnerhas put the share of customer service issues fully resolved in self-service in the mid-teens, not the majority. Meanwhile the same body of research finds that most consumers still prefer, or want the option of, a human for anything non-trivial. So a 70% “deflection rate” can quietly contain a large block of people who bailed and either called back through another door, churned, or posted about it.

The definitions you actually want

Before you can change a target you have to change what the words mean. Here is the vocabulary swap, stated plainly enough to paste into a metrics doc.

The two that do the most work are repeat-contact rate and cost per resolved outcome. Repeat-contact rate is the honesty check on everything else — if resolution is real, people don’t come back. And cost per resolved outcome is the number that translates support into the language finance already speaks, because it divides by the thing you actually wanted (a solved problem) instead of the thing you were trying to avoid (a conversation).

How to measure resolution without kidding yourself

Resolution is harder to measure than deflection precisely because it can’t be inferred from the absence of an escalation. You have to triangulate it. Three signals, weighted:

• Repeat-contact within a window. If the same customer contacts you again about the same issue inside 7 days, the first contact was not resolved — regardless of how it was tagged. This is the strongest signal and the hardest to game.
• Explicit confirmation.A single “did this solve it?” at the end of the conversation. Cheap, and it catches the silent abandoners that repeat-contact alone misses.
• Downstream outcome. The appointment was actually booked, the payment actually went through, the order actually shipped. Where a real system event exists, trust it over any self-reported tag.

Notice what’s missing: agent disposition codes. “Resolved” as clicked by the person who wants the ticket off their queue is the least reliable input you have. Use it last, and only as a tiebreaker.

Deflection can be inferred from silence. Resolution has to be earned and then verified.

The migration, step by step

Don’t rip the old dashboard out. Run the new metrics alongside it for a cycle so the two stories can be compared — that comparison is your best evidence when you present the change.

1. Instrument repeat-contact first.Before you touch any target, start measuring how many contacts recur within 7 days. You almost certainly can’t see this today, and it’s the number that reframes everything.
2. Redefine a “win.” A win is a resolved outcome with no repeat contact — not a closed ticket. Write it down. Get the support lead and the finance partner to sign the same sentence.
3. Run old and new side by side for one reporting cycle. Keep deflection on the page. Add resolution rate, repeat-contact rate, and cost per resolved outcome next to it. Let the gap between the two be visible.
4. Switch the target. Once the new numbers are trusted, move the goal — and the incentives — to resolution and repeat-contact. Whatever the compensation or QA rewards is what your team will actually optimize.
5. Retire deflection as a headline. Demote it to a diagnostic, not a scoreboard. It still tells you where load is going; it just stops being the thing you celebrate.

Selling the change to leadership

The objection you’ll hear is that resolution rate will look worse than deflection rate did — and it will, at first, because it’s honest. Get ahead of it. The pitch to a CFO isn’t “our numbers went down,” it’s “our numbers got real, and here is the money hiding in the gap.”

Frame it as risk, not vanity. Every unresolved-but-deflected contact is a retention liability, and retention is where the economics live — acquiring a replacement customer is far more expensive than keeping one, a gap consistently documented in Bain & Company’s customer-loyalty research. When you can show cost per resolved outcome and a falling repeat-contact rate, you’re no longer defending a support budget — you’re protecting revenue that was leaking out the back.

Be equally honest about the upside. The realistic net cost reduction from automation is closer to 20–35% than the 60–80% vendors advertise, once you count escalations and the work to keep quality high — and independent studies of AI in real workflows, including NBER’sfield research on generative AI at work, land in that more sober range. A business case built on the honest number survives its second quarter. One built on the headline number doesn’t.

What good looks like

Six months in, the deflection number isn’t on the wall anymore. The wall shows resolution rate, repeat-contact rate, and cost per resolved outcome, and all three are trusted enough that nobody argues about them in the QBR. Your agents spend their time on the contacts that genuinely need a human, because the routine ones are being resolved — actually resolved — not just kept out of the queue. And when finance asks what support costs, you answer in dollars per solved problem, not tickets avoided.