Deflection rate is a vanity metric
What to measure instead, and why “resolved without a human” is the only number that pays a bill.
Significo team4 min read

The slide everyone likes
Picture a monthly review. On the screen there is a big number: the share of conversations the bot handled without passing them to an agent. Everyone nods. Fewer conversations for the team, lower costs, a busy bot. It feels like progress.
Then someone from the support team mentions that the phones were unusually busy this month. Nobody quite connects the two.
That number is the deflection rate, and we have learned to be careful with it. It tells you where a conversation did not go. It does not tell you whether the customer got what they came for.
Why it fools us
A customer who gives up halfway through a chat and calls the next day counts as deflected. So does a customer who gets an answer to a slightly different question and leaves more confused than before. On the dashboard, both look like wins.
It gets worse. If the goal is to keep people away from agents, the easiest way to raise the number is to make reaching a person harder. Hide the contact options, add another menu, suggest one more article. Deflection goes up while the experience goes down.
And the question does not disappear. It comes back through another channel, usually with more frustration attached, and now it costs more because the customer has to explain everything again.
Count what was actually solved
The better question is simple: was the request resolved? Resolved without a human means the AI agent gave a correct, complete answer or finished the task, and the customer did not have to come back about the same thing. That saves time for your team and for your customer. It is the number that pays the bill.
It is harder to measure than deflection, because you need to know what happened afterwards. This is where ticketing earns its keep. When every conversation becomes a ticket with a status, you can see whether a request was closed, reopened or followed by a new contact on the same topic.
A few numbers that tell you more
We like to look at resolution together with a handful of other signals. How quickly does a customer get a first response, including at night and over the weekend? How long does a request stay open from the first contact to the final answer?
Repeat contacts about the same issue are often the most honest signal of all, because they point straight at answers that were unclear or incomplete. Handovers matter too. When the AI passes a conversation to a person, does the agent get the context they need, or do they start from zero? And the topics customers raise most often show you where better information, or a better process, would make the question go away.
Break these down by channel and topic and patterns appear. A topic that is solved well in chat but keeps coming back after phone calls usually points to a gap in information or process, not to one team doing a poor job.
None of this has to be perfect on day one. What matters is that the numbers describe the customer’s experience rather than the bot’s activity.
How we approach it in SIGNIFICOAI
In SIGNIFICOAI, the AI agent and your team work in the same ticketing system. Every call, email, web form and message becomes a ticket with an owner and a status, and anything handed to a person carries its history with it.
Ticket insights show counts, trends, status and priority. Telephony statistics show when calls come in and how they are handled. And with Siggy, the AI assistant inside the software, a manager can simply ask which topics are rising this week instead of building a report by hand.
Because the AI and the team share the same data, it becomes clear which requests the AI resolves well, which ones it hands over and where answers need work.
Would your customers care?
Here is a test we use for any support metric. Would your customers care about it? Nobody has ever been pleased to be deflected. People care about a clear answer, quickly, without repeating themselves.
Start your reporting there, and automation becomes a way to solve more requests well rather than a way to keep people at arm’s length. That is a goal your team and your customers can both get behind.
Like what you read? Add Significo as a preferred source on Google.