Deflecting is not stopping people from writing
The usual trap is counting as deflected any conversation that never reached a person, including the one where the customer gave up.
A hidden contact button raises deflection rate and damages the business. The metric only means something if "deflected" means "resolved without intervention", and it is verified by the customer not coming back.
Why it matters
It is the one that translates support into money.
Multiplied by what each ticket costs, it says how much self-service saves; without that cross, it is a percentage that decides nothing. It is also the one that best detects whether an AI investment is resolving or merely placing an obstacle in front of the contact form.
Calculating it without fooling yourself
Conversations closed without human intervention and without a reopen in the following days, divided by the total conversations started.
The second condition is what stops you counting the people who gave up.
The ceiling depends on the type of question
Questions whose answer is the same for everybody deflect well with documentation.
Questions that depend on account state do not deflect with documentation however good it is, because the article does not know which plan the asker is on. In a SaaS, that second group is large, and it is what sets the ceiling.
Signs you are measuring it wrong
Deflection rate rises and CSAT falls.
Or it rises and so does the number of customers writing through another channel, such as an email to their account executive or LinkedIn. Both mean the same thing. You have not deflected, you have hidden.
The formula, with an example
Conversations resolved with no person and no reopen, divided by conversations started, times one hundred.
The figures in the example are illustrative.
- Example. 1,000 conversations in a month, of which 350 close with no intervention and without the customer writing again, give 35%.
- What does not count. The conversation the customer abandons with no useful answer and the one they reopen two days later.
