Pointing a customer to a help article counts as deflection, not resolution. Deflection means the customer did not reach a person; resolution means their problem was actually solved. The two get blurred in reporting, and an agent tuned for deflection learns to send customers elsewhere rather than finish the job it was asked to do.
What is the difference between the two?
Deflection rate measures how many customer contacts were diverted away from staff, usually to self-service content such as an article, a frequently-asked-questions page or a form. Resolution rate measures how many contacts ended with the customer's issue settled.
They answer different questions. Deflection asks where the customer went. Resolution asks whether the customer got what they needed.
| Deflection | Resolution | |
|---|---|---|
| Question it answers | Did the customer avoid a person? | Was the problem solved? |
| Counted when | The customer is sent to self-service and does not escalate | The issue is settled, ideally confirmed |
| Customer who gave up | Counts as a success | Counts as a failure |
| Customer who called instead | Often counts as a success in the chat channel | Counts as a failure |
| What it rewards in an agent | Sending people elsewhere | Finishing the task |
A conversation can be deflected and resolved, if the article really did answer the question. It can also be deflected and unresolved, which is where the trouble lies.
Why do the two get confused?
Partly because both reduce the number of contacts reaching staff, so both show up as savings. Partly because deflection is easy to measure: the system knows it sent a link. Resolution is hard: the system has to find out whether the problem went away, which usually means watching what the customer does next.
When the hard measure is missing, the easy one fills the gap. Reports start to treat "customer sent to self-service and did not come back to this chat" as solved. That counts silence as satisfaction.
What happens to customers who were deflected but not helped?
They go somewhere. Some try the same chat again with different wording. Some switch to email or the phone, where the contact now costs more and the customer is more frustrated. Some complain. Some give up, which may look like a saving but often means a problem that will return later as a missed payment, a cancelled policy or a lost customer.
In regulated markets, giving up has another dimension. A customer who could not get an answer about a charge, a claim or a right may lose something they were entitled to. A firm that measured that interaction as a success has a measurement problem with real consequences.
When is pointing to an article the right answer?
More often than critics of deflection admit. If a customer asks how to reset a password and a clear article explains exactly that, sending the article is good service. The customer gets a complete answer they can follow at their own pace, and nobody needs to retype what the article already says.
The test is whether the article resolves the question as asked. Three conditions should all hold:
- The question is informational, with a single correct answer that does not depend on the customer's account.
- The article answers that exact question, not a related one.
- The agent checks, briefly, whether the article helped, and offers to continue if not.
Where any of those fail, the article is a deflection in the bad sense. A customer asking why their refund has not arrived does not need an article on refund timelines. They need someone to look at their refund.
How can you tell whether a deflection resolved anything?
By following the customer, not the conversation. Useful signals include:
- Repeat contact: did the customer contact the firm again about the same issue within a set period, on any channel?
- Channel switching: did they phone or email shortly after the chat ended?
- Explicit confirmation: did they answer yes when asked whether the article solved the problem?
- Downstream events: did a complaint, cancellation or missed payment follow?
None is perfect alone. Together they let you split deflections into those that resolved the issue and those that merely moved it. That split is far more informative than either rate alone. The approach fits within the wider set of measures in measuring a support agent beyond resolution rate.
What does this mean for how the agent is built?
An agent built to resolve behaves differently from one built to deflect. It looks at the customer's account before deciding whether an article will do. It answers in the conversation when the answer depends on the customer's situation. It checks whether its answer helped. And it hands over to a person when it cannot resolve the issue, rather than sending one more link.
That last behaviour is the clearest difference. An agent that knows when to stop and pass the case on will look worse on deflection and better on everything that matters. That is the stance of the AI support agent that knows when to say no: saying "I can't solve this, but a colleague can" is more useful than an article that does not answer the question.
What should you do differently?
Stop reporting deflection as a success measure on its own. Report it next to repeat contact and channel switching, so every deflection figure comes with evidence of whether it helped. Change the agent's brief so that sending an article is one option among several, chosen only when the article answers the question. And review a sample of deflected conversations regularly, asking a single question of each: was this customer's problem solved?
Questions people ask
Is a high deflection rate always bad?
No. For simple informational questions it can reflect good service. It becomes a problem when it is high for questions that depend on the customer's account or need action.
How long should the repeat-contact window be?
Long enough to catch customers who tried again after a delay, but short enough that a new issue is unlikely. The right window depends on the product; set it deliberately and keep it stable.
Should the agent ask whether an article helped?
Yes, briefly. A one-line check gives you direct evidence and gives the customer an easy route to continue if it did not.