Deflection rate shows up in every vendor pitch now. It's the number that gets circled on the slide deck, the metric that justifies the software spend, the thing that makes sense to executives who are trying to understand whether their investment in digital support is actually working.
The number usually looks good. "Our AI self-service deflected 40% of tickets." Sometimes it's 60%. Sometimes vendors claim even higher. It's a satisfying number because it's big and directional and easy to communicate upward.
But deflection rate was never designed to answer the question everyone actually wants answered: did we help the person?
It was designed to answer a different question. A narrower one. A logistics question. Can we avoid creating a support ticket in the system? That's what deflection measures. It measures whether an interaction stayed out of the queue.
Those are not the same thing. Not even close.
Someone might get deflected into a chatbot, spend ten minutes getting nowhere, then manually go create a ticket anyway. That's still a deflection in the metric. Or they might get deflected into an FAQ that doesn't actually address their specific situation. They leave frustrated, they go to social media, and their problem becomes someone else's to solve. Still a deflection. Or they might find an answer that solves their immediate problem but creates three downstream issues nobody knew about until later. Deflected, but not actually helped.
The metric doesn't care about any of that. It only cares whether a formal support ticket got created in the system. Everything else is invisible to the number.
This matters because when you optimize for deflection, you're not actually optimizing for helping people. You're optimizing for moving them somewhere else. There's a real incentive to make self-service options just sophisticated enough that people give up trying to get a real answer, not because they found one, but because the friction was too high. That's deflection. That's a win in the metric. That's not help.
The weird part is that nobody thinks they're doing this. Every vendor sincerely believes their self-service tools are helping people. Every organization rolling out new deflection strategies thinks they're improving the customer experience. And genuinely, sometimes they are. Good self-service can actually be faster and more satisfying than waiting for someone to respond. But the metric doesn't distinguish between that scenario and the scenario where someone just gave up.
We'd need completely different measurements to know the difference. Outcome-based measurements. Did the person actually solve their problem or at least move closer to solving it? Did they feel like they got help? Did they come back to support again for the same issue? Did the resolution stick, or did it just push the problem somewhere else?
Those measurements are harder to get. They require following the person's journey, not just watching whether a ticket materialized. They require actually asking whether someone was satisfied, not just whether they disappeared from the support queue.
So instead we optimize for what's easy to measure. And what's easy to measure is also what's easiest to game. The incentives align perfectly toward deflection strategies that look great in dashboards but might be making the actual experience worse.
It's not a conspiracy. It's just what happens when the metric becomes the goal and nobody stops to ask whether the metric was measuring the right thing in the first place.