Deflection

Deflection is keeping a customer query out of a human queue. Often times this is through pointing the customer to a help article, a self-service flow, or an automated reply, and sometimes a customer is kept out of the queue because their problem is fully resolved and back-office action has been taken (more commonly referred to as “resolution”). One key thing to know about the term “deflection” is that if a customer never speaks to a human, the case is considered deflected, regardless of whether or not the customer’s problem is actually solved.

Deflection is a containment metric, not a resolution metric. A high deflection rate shows a system is successfully absorbing volume that would otherwise reach a human. It does not show whether the customer's problem was actually fixed: a customer pointed to an FAQ that does not answer their question counts as deflected even if they come back with the same issue an hour later.

How deflection differs from resolution

Deflection asks whether a query stayed out of the human queue. Resolution asks whether the customer's problem actually got fixed, back-office action included, so a vendor optimised purely for deflection can post an impressive rate while a meaningful share of those customers come back, escalate, or complain. Gradient Labs' guide to deflection vs resolution in AI customer service covers what to ask a vendor when a deflection rate and a resolution rate are quoted side by side.

Where deflection metrics mislead in regulated industries

In banking, lending, and insurance, a deflected-but-unresolved query often resurfaces as a complaint, a regulatory referral, or reputational damage, because the customer had a real financial problem that a help article did not fix. A deflection-first system tends to perform well on simple, low-stakes queries and poorly on the complex ones that matter most in a regulated business, precisely the cases where getting it wrong carries the highest cost. This is why resolution, not deflection, is the metric worth building an evaluation around in financial services.

Frequently asked questions

Is a high deflection rate always a good sign?

Not on its own. It shows volume is being absorbed, but says nothing about whether the customer's problem was solved. Read deflection rate alongside resolution rate and repeat-contact rate before treating it as a success metric.

How is deflection different from containment rate?

The two are close to synonyms in practice; containment rate is the more common term in insurance and some other sectors, and deflection rate is more common in support and helpdesk tooling. Both measure whether a query stayed out of the human queue, not whether it was solved.

Can an AI agent deflect and resolve at the same time?

Yes, when the agent actually completes the case rather than just answering a question. The goal is not to avoid deflection, it is to make sure whatever is deflected was also genuinely resolved.