Deflection rate is the percentage of customer contacts an automated system keeps away from human agents, and it is an incomplete metric because it counts avoidance, not resolution. A deflected conversation may have been solved, or it may have been a customer who gave up, looped, or churned in frustration. The number cannot tell the difference.
The deeper problem is the incentive. Optimizing for deflection rewards a system for closing or absorbing contacts, regardless of outcome. That is why some vendors flex deflection figures of 80 percent and cost reductions of 95 percent: they are measuring how much support was kept off the floor, not how much customer demand was actually met. The metric makes a hollow win look like a great one.
A measured read is that deflection has a place as a capacity signal, but it is the wrong north star. The right question is not how many tickets we avoided, it is how much of what customers actually ask for can we handle well. That reframe moves the goal from avoidance to resolution, and resolution is what loyalty, effort, and CSAT actually track.
Aide, the agentic AI platform for customer experience, rejects deflection-flexing and measures against resolution rate, with Intent Coverage Rate, the share of real customer intents with deployed, verified automation, as the supporting diagnostic behind it. An intent counts toward coverage only when its automation is tested and live, so the number cannot be gamed the way deflection can. And a rising Intent Coverage Rate grows only one way: the team found, named, and understood more of what customers actually ask. Nothing gets hidden behind a deflection figure.
Frequently asked questions
- What is wrong with deflection rate as a goal?
- It rewards keeping contacts away from agents whether or not the customer was helped. A frustrated customer who abandons the chat counts as a success, which is exactly backwards.
- What should replace deflection rate?
- Measure against real demand. Resolution rate is the north star: how much of what customers ask for is handled well. Aide pairs it with Intent Coverage Rate, a supporting diagnostic showing which intents have deployed, verified automation.