Cut ticket volume without hiding from your customers

Deflection earns a bad name when it means burying the contact link. Done honestly it means answering the question so well that nobody needs to write in.

Start free How to measure it

No credit card. Baseline first, ideally.

can I change the delivery address after ordering?

Yes, until the order is packed. Open the order in your account and edit the address there.

Orders & delivery

Talk to a human

Never hidden, never argued with

Answers, not obstacles: The route to a human stays visible the whole time.
No interception: Anything outside its knowledge goes straight through.
Measured honestly: Answer rate read next to refusals, not on its own.
Improves as you write: Each gap you close is a question that stops arriving.

A busy in-house support floor at mid-morning, headsets on, one person leaning back rubbing an eye.

Volume grows with the customer base, headcount does not

Support load scales with signups while the team scales with budget. The usual fixes — a harder-to-find contact form, a mandatory help-centre search, a longer phone tree — reduce tickets by making customers give up, which shows up later as churn.

First-response time rises every quarter regardless of process changes
A large share of tickets are answered by pasting the same link
Deflection tactics push CSAT down as fast as they push volume down
Nobody can say which articles actually prevent tickets

Deflection that does not show up later as churn

Volume falls for two different reasons: because the answer got easier to find, or because contacting you got harder. Only one of those survives a year, and the difference is entirely in how the assistant behaves when it cannot help.

Instant on the repetitive

Hours, policies, plan differences, setup steps — resolved in seconds rather than in a queue.

It never argues

Ask for a person and you get one. Making customers fight a bot is what damages satisfaction.

Escalations carry the thread

When it does hand over, your agent starts from what has already been ruled out.

Refusals as a metric

Logged and grouped, so a rising answer rate can be checked against whether it started guessing.

The overnight backlog

Questions arriving outside your hours are resolved rather than queued for Monday.

A loop, not a launch

Write the missing article, the next crawl picks it up, that question stops arriving.

Deflection by answering, not by hiding

The only durable way to reduce tickets is to make the answer easier to get than the ticket. The bot resolves the repetitive questions in seconds with a citation, and makes no attempt to intercept a question it cannot handle.

Answered in seconds: The questions your team answers by pasting the same link, handled before anyone opens a ticket.
No interception: A question it cannot answer is handed on immediately, not talked around.
The contact route stays visible: Reaching a person is never hidden behind a bot that will not give up.
Escalations carry context: The transcript arrives with the handover, so nobody repeats themselves.

Measure deflection honestly

Answer rate on its own is a vanity metric — a bot that answers everything confidently scores perfectly while inventing policies. Read it next to refusal volume and the content-gap list to see whether resolution is real.

Answer rate: Share of conversations the bot answered rather than refused.
Refusals, grouped: What it could not answer, ranked — the actual roadmap for your help centre.

Every refusal makes the next month better

The gap list turns support into a loop rather than a treadmill. Write the missing article, the bot picks it up on the next crawl, and that question stops arriving. Volume falls because the content improved, not because the customer gave up.

The honest part

Falling tickets is not the same as fewer problems

Volume drops for two very different reasons, and they look identical in a dashboard. Either the answer got easier to get, or contacting you got harder. Only one of those is still working in a year, and no amount of reporting on ticket count alone will tell you which one you built.

The route to a person stays visible and never argues back
Anything outside its knowledge goes straight through, not into a loop
Refusals are logged and grouped rather than quietly absorbed
Answer rate is only meaningful read next to what it refused

We publish no deflection percentage, because the honest number depends entirely on how much of your repetitive volume is already written down somewhere. The gap report is what tells you where yours would land.

A man alone at the end of a long meeting table late in the day, reading a screen with the flat expression of someone weighing a number.

Tickets fell. Now read the other number.

↓ tickets＋→ satisfaction

Deflection. The answer got easier to find.

↓ tickets＋↓ satisfaction

Attrition. Contacting you got harder.

One number alone says nothing

Getting a number you can defend in a review

Most published deflection figures are overstated, usually by accident — counting every bot conversation as a ticket that would otherwise have been filed. Here is the version that survives scrutiny.

Analytics cards showing conversations, answered rate, unanswered questions, negative ratings, handoffs and leads captured. — Answered rate on its own is the easiest number in support to move for the wrong reason.

1
Record a baseline first
Ticket volume per week, normalised against active customers or sessions. Without a before, any after is unfalsifiable.
2
Launch and leave it alone
Give it several weeks. Weekly volume is noisy, and the first fortnight is distorted by the novelty of a new widget.
3
Measure the change in tickets
Not the count of chats. The honest metric is normalised ticket volume before against after — actual displacement.
4
Read satisfaction alongside it
Falling tickets with falling CSAT is not deflection, it is attrition. The two numbers only mean anything together.

Deflection, without the weasel words

Short answers. If yours is not here, the assistant on this page will try it — and tell you honestly if it cannot.

Will deflection hurt customer satisfaction?

It does when deflection means making support hard to reach. It does not when it means answering the question in two seconds with a link to the source. The distinction is whether the bot admits what it cannot do — one that argues with a customer instead of handing over is the thing that damages satisfaction.

How much volume can realistically be deflected?

It depends almost entirely on how much of your repetitive volume is already documented somewhere. Sites with a thorough help centre see the largest share, because the bot can only resolve what your content covers. The gap report tells you what to write to move the number.

Does it replace my helpdesk?

No. It sits in front of it, resolves what your content covers, and routes the rest into the tools your team already uses with the conversation attached.

All solutions