Guide · AI and knowledge base
Self-service deflection: what it is and how to raise it honestly
In short
Deflection is the share of questions customers settle on their own, from a help centre or a bot, without waiting for a person. It’s worth raising: those customers get their answer sooner, and your team keeps its time for the harder cases. But count it carefully, because a customer who gave up isn’t a customer who got an answer.
Reviewed 9 November 2026. Terms used: RAG (retrieval-augmented generation), Ticketing system, CSAT vs NPS.
Seven steps to raise deflection without hiding your team
Measure first, then write, then put the answers where people ask. In that order.
Define it before you count it
Write down what counts: a question the bot or an article answered, with no hand-over, and the customer didn’t come back about the same thing soon after. Count a chat that went quiet apart from one that ended with an answer, and never count a customer who couldn’t find a way to a person.
One way to write it down Deflected: answered by the bot or an article, no hand-over, no new chat on the same topic within 48 hours.
Not deflected: handed over, left in the queue, or back the next day with the same question.
Start from your top questions
Export a month of chats and tickets and group them by question. Write the twenty most asked first, and check the list again next month.
Write one answer per question
Use the customer’s words in the title, put the answer in the first two lines, and the details after. One page per question works better than a long FAQ that answers everything halfway.
Put the answer where the question is asked
Don’t wait for customers to find the help centre. Answer in the chat they already opened, on your website or WhatsApp, and link the page so they can read more.
Keep the way to a person in plain sight
A bot that hides the human only makes customers repeat themselves later. Offer a person after one or two missed answers, and whenever someone asks.
Read what wasn’t answered
Every week, look at the questions the bot couldn’t answer and the searches that found nothing. They are your next articles.
Watch deflection next to satisfaction
Deflection that rises while ratings fall, or while the same customers come back, isn’t working. Read the three together each month.
Deflected, or just gone?
In the chat
- Counts as deflected
- Answered from an article, no hand-over
- Doesn’t count
- Left halfway through the menu
In the help centre
- Counts as deflected
- Read the article and didn’t write in
- Doesn’t count
- Searched, found nothing, left
The next day
- Counts as deflected
- No new contact on the same topic
- Doesn’t count
- Back with the same question
A person
- Counts as deflected
- Not needed
- Doesn’t count
- Wanted, but couldn’t be reached
How it works in Axonave
Deflection counted, and the questions it missed, on one page.
Reports → Chatbot counts each bot conversation as resolved, handed over, taken over or gone quiet, and calls resolved plus gone-quiet “deflected”. So read it beside the hand-over reasons, where customers stopped in each flow and the top unanswered questions, which sit on the same page. A message left with the bot counts as resolved. The AI answers only from your published articles and names the one it used. Questions it couldn’t answer, text a menu couldn’t match and agents’ searches that found nothing are grouped for you, and “Make a draft” writes an article for a person to edit and publish. Customers can search the same articles in your public help centre.
- Citations by article: which pages do the answering
- A hand-over to a person when the AI is unsure or the customer asks
- Chat reports show bot containment by day and channel
Reports → Chatbot in our demo workspace: what the bot settled alone, why it handed over, what the AI answered and where customers stopped.
Common questions
What’s a good deflection rate?
There’s no single number that fits every business, and we haven’t measured one. Take your first month as the baseline and judge changes against it, together with ratings and repeat contacts.
Does a chat that goes quiet count as deflected in Axonave?
Yes. Reports → Chatbot counts resolved and gone-quiet bot conversations as deflected, shows each one separately, and puts the hand-overs beside them.
Can customers still reach a person?
Yes. The AI hands over when it can’t answer, when the customer is upset or asks for a person, or after the unanswered questions you allow. In a WhatsApp menu, a “Talk to a person” option joins after repeated misses.
See it working with your own customers.
We reply within one business day by email or WhatsApp.