What's the difference between deflection rate and resolution rate?

Deflection rate and resolution rate sound close enough that they often get treated as the same thing.

They are not.

Deflection rate measures whether AI kept a ticket away from a human. Resolution rate measures whether the customer’s problem was actually solved. The gap between those two numbers is where a lot of AI support deployments quietly fail.

Every vendor pitches the deflection number. Every customer experiences the resolution number. If you are only tracking one, it is almost certainly the wrong one.

What is deflection rate?

Deflection rate is the percentage of customer service tickets handled without escalating to a human. It is the single metric most AI customer service vendors lead with, because it produces a clean dashboard number that improves dramatically the moment the bot goes live.

The mechanics are straightforward. A ticket comes in. The AI takes the first attempt. If the customer doesn’t escalate, doesn’t reopen the conversation, and doesn’t ask for a human, the vendor counts the ticket as deflected. The denominator is total tickets; the numerator is whatever didn’t bounce back to your team.

Deflection rate measures activity. It tells you the AI touched the ticket. It does not tell you the ticket was resolved.

What is resolution rate?

Resolution rate is the percentage of tickets where the customer’s problem was actually solved end-to-end, in a single interaction, without follow-up contact within a reasonable window (most teams use 7 days).

It’s a harder metric to compute. You need to track repeat contacts, escalation paths, customer-reported satisfaction signals, and sometimes the eventual outcome (refund issued, order replaced, account reactivated). It is not a number the bot vendor can show you in real time on a dashboard, because the data it needs lives in lagging customer behaviour.

That lag is why vendors don’t lead with it. It is also the reason resolution rate is the only metric that actually tracks customer experience.

Where the two diverge

The simplest example: a customer asks the bot for a refund. The bot replies with the return policy and asks if they would like to start a return. The customer reads the policy, says “I’ll wait for a human,” and closes the chat.

By most vendors’ definitions, that ticket is deflected. The AI handled the first response. The customer didn’t escalate inside the chat. The dashboard ticks up.

What actually happened: the customer is now waiting for a human anyway, the ticket reopens the next day under a fresh thread, and your team handles it cold without the context of the bot conversation.

Deflection: 1. Resolution: 0.

That 8% is the AI-alone resolution number for refunds across 16,000 brands. The deflection number for the same category can sit far higher in vendor reports, because the bot’s first reply counts even when the customer rejects it.

Across the inbox, the same gap holds. Order status approaches a real deflection ceiling near 60-70% and a resolution number that tracks it closely. Refunds, returns, and complaints sit under 10% on resolution regardless of what the deflection figure says.

What you should actually be tracking

Two numbers, not one. Run both, report both, and break each down by ticket category.

Resolution rate per category. Order status, FAQ, and password reset will track high; refunds, complaints, and subscription disputes will track low. The gap between categories is the real shape of your AI deployment.

Repeat-contact rate within 7 days. Tickets the bot “resolved” that come back to you under a new thread are the clearest signal that the first answer didn’t land. Customer behaviour catches what your dashboard doesn’t.

Customer-escape rate. Track the percentage of chats where the customer types “human,” “agent,” “person,” or “speak to someone.” That ratio measures how many of your customers are actively rejecting the bot, regardless of whether the bot eventually answered them.

Trustpilot and review-site velocity. Customer-side signals lag three to six months. If your deflection rate is climbing and your rate of negative reviews is climbing alongside it, the bot is touching tickets it isn’t resolving.

Closing

Deflection is the number the vendor sells. Resolution is the number the customer experiences. The vendor who can break both numbers down by ticket category, side by side, is the one telling you the truth about what the AI is doing.

One number hides the trade-off. Two numbers describe what the customer actually went through.

Sources

• Gorgias 2026 State of Conversational Commerce. 8% of refund and return requests handled by AI alone end-to-end across 16,000 brands.
• What’s the right deflection KPI?. Per-category deflection ceilings: 60-70% on order status, under 10% on refunds.
• AI in customer service is two products. Where deflection AI fits and where it doesn’t, and what agent assist does instead.