Deflection Rate vs Resolution Rate: Which Number Actually Matters

The two numbers, precisely

Deflection rate is the share of conversations that ended without reaching an agent. It is easy to measure, because the system knows whether a ticket was ever assigned to a person.

Resolution rate is the share of conversations where the customer got the outcome they came for. It is harder to measure, because the system has to know something about the outcome rather than about routing.

Every conversation that a customer abandoned in frustration counts towards deflection. None of them counts towards resolution. That is the whole problem in one sentence.

Where the two numbers come apart

Take two stores that both report 60% deflection to their board.

• Store A gets mostly “where is my order” questions. The AI reads the tracking status and answers. Of the 60% deflected, roughly all were genuinely resolved.
• Store B sells made-to-measure furniture. The AI answers what it can, and for the rest it repeats a generic policy line until the customer gives up. Of the 60% deflected, perhaps half were resolved. The other half emailed again two days later from a different address, or did not come back at all.

Same headline number. Completely different businesses underneath it. And Store B’s second contact often shows up as a new conversation, which quietly improves the deflection rate again.

A calculation worth doing once

Take one month. Take the conversations the AI handled end to end, without a human. Then subtract:

1. Conversations where the same customer contacted you again within 48 hours, on any channel.
2. Conversations where the last message was from the AI and the customer never replied and never took the action the reply asked for.
3. Conversations that ended with a request for a human that was not honoured.

What remains is close to your real resolution count. Divide it by all conversations in the month.

The number is almost always lower than the deflection rate you were quoted. The size of the gap is the useful part: a gap of a few points is normal, a gap of twenty points means the assistant is suppressing contact rather than handling it.

Why benchmarks are close to useless here

The share of tickets that can be automated is decided by your ticket mix, not by the vendor. Order status, delivery windows, returns eligibility and address changes are answerable from data the system can look up. Sizing advice, warranty judgement calls, complaints and anything involving an exception are not.

Two stores on the same platform can sit thirty points apart on resolution rate and both be doing everything right. Before you accept any number in a sales deck, sort last month’s tickets into answerable from data and needs a decision. That ratio is your ceiling.

What to put on the dashboard

• Resolution rate as the headline, calculated as above.
• Deflection rate next to it, so the gap stays visible.
• Repeat contact within 48 hours, which is the earliest warning that the gap is growing.
• Escalation rate, and treat a rise as information rather than failure. See when AI should hand over to a human.

If you only have room for one number, use resolution rate. It is the only one of the four that cannot be improved by making your customers give up.