Use this one
Decagon, Aisera, Atera
Share of all support requests
480deflected
1,000who asked for help
48%
The strict reading, and the only one that falls when customers cannot find your help center.
Ticket deflection
Ticket deflection that does not bill you per deflection
Zendesk, Intercom and Help Scout each charge between $0.75 and $1.50 every time their AI deflects a ticket. Optlo runs on your own AI account, so you pay the model's list rate.
1,000 conversations arrive
700answered from your own docs
300handed to a person
The short answer
Ticket deflection is resolving a customer's question before it becomes a support ticket. The customer still gets an answer; a person never has to write it. The deflection rate is self-service resolutions divided by total support requests, times 100.
Three ways to deflect, and only one of them scales
Deflection is as old as the help center, and each version of it reaches a different group of customers.
| Approach | Reaches | Misses |
|---|---|---|
| Help center articles | Customers who go looking for an answer before they write in. | Everyone who does not. 380 of 1,000 never opened self-service in the month below. |
| A bot with a decision tree | The questions someone thought to anticipate, on the paths they mapped. | Anything phrased slightly differently, and it irritates the customer on the way past. |
| An agent reading your docsUse this one | Whoever writes in, wherever they write in, in the channel they already chose. | Questions your documentation does not answer yet, which it hands to a person rather than inventing. |
How to measure it
One month of support, counted three ways, with three different answers
Published deflection benchmarks range from roughly 15 percent to 85 percent, and most of that spread is arithmetic. Here is a single month of support, scored by the three formulas that rank for this term.
Magentrix
Share of self-service sessions
480deflected
620who opened self-service
77%
Leaves out the 380 who emailed without opening self-service, which is where most of the gap comes from.
Zendesk
Self-service score
4,100article views
520tickets created
7.9×
A ratio of reading to filing, so it climbs when your articles get more traffic even when the same number of tickets arrive.
What it costs
On per-resolution pricing, deflection is the meter
The vendors below sell deflection and bill for it by the unit. That makes the deflection rate two things at once: the number measuring how well the agent works, and the number they calculate your invoice from.
- Zendesk$1,050
- Intercom$693
- Help Scout$525
- Your own provider key$16.80
Why per-resolution pricing gets more expensive as you get better
Every one of these vendors charges a fee each time the AI succeeds, and each counts a success slightly differently.
| Vendor | Rate | Billed |
|---|---|---|
| Zendesk | $1.50 | per automated resolution |
| Intercom | $0.99 | per Fin resolution |
| Help Scout | $0.75 | per AI resolution |
| Freshdesk | Meters | per 72-hour Freddy session, resolved or not |
At 1,000 conversations and a 70 percent deflection rate, resolution based billing is $525 to $1,050 a month, on top of whatever you pay per seat.
On Optlo the same month costs $16.80 in AI usage. You connect an OpenAI, Anthropic or Gemini key and the provider bills that account at its published rate. The figure does not move with the deflection rate, because the agent reads and writes about the same number of tokens whether it resolves the conversation or hands it over.
The better the agent gets, the more you pay for it. Improving the deflection rate by fine-tuning your own documentation results in a larger per-resolution bill. With Optlo you can bring your own key, and the only thing that rises is the deflection rate.
What actually deflects
Four things to check before you trust a deflection rate
Three of them decide how high the rate can go. The fourth decides whether it is telling you the truth.
Grounding
Does it answer from your documentation, or from the internet?
Nothing else matters as much. An agent reading your help docs, your site and your pasted FAQs answers the specific question about your refund window; one reasoning from general knowledge produces something plausible and wrong, and the customer writes in anyway. Ask what happens when the answer is not in the source material, and prefer a tool that says it does not know.
Placement
Does it work where the customer already is?
A help center only deflects the people who go looking for it. In the worked example above, 380 of 1,000 never opened one. Answering inside the chat widget, and on the reply to an inbound email, is what pulls that second group into range.
Actions
Can it finish the job, or only describe it?
Half of what looks like a question is really a request: where is my order, change my address, resend the invoice. An agent that can look the record up and act closes those; one that can only explain how to do it has handed the work back to the customer, and will see them again next week.
Honesty
Is the handoff easy, and is the re-contact rate visible?
A deflection rate rises if you bury the contact link, and every number on this page can be gamed that way. The check is re-contact rate: how many deflected customers came back within a week with the same problem. If a tool reports deflection but not re-contact, it is reporting the half of the story that flatters it.
Side by side
Where Optlo differs, and where it does not
The top of this table is parity. Every tool here reads your documentation, hands off to a person and reports a deflection rate. The difference is underneath, in who owns the AI account and therefore who captures the price of every answer.
| What you are paying for | Zendesk | Intercom | Help Scout | Freshdesk | Optlo |
|---|---|---|---|---|---|
| Answers from your own documentation | Yes | Yes | Yes | Yes | Yes |
| Hands the thread to a person mid-conversation | Yes | Yes | Yes | Yes | Yes |
| Reports a deflection rate out of the box | Yes | Yes | Yes | Yes | Yes |
| The same bill whether the agent resolves or hands off | No | No | No | NoBilled per 72-hour Freddy session | YesConversations are a capacity unit, not a success fee |
| AI charged at your provider's list rate | No | No | No | No | YesYou hold the key and the account |
| Choose the model doing the deflecting | No | No | No | No | YesOpenAI, Anthropic or Gemini |
| Your AI cost falls when providers cut prices | No | No | No | No | Yes |
Capabilities and prices checked August 2026. Vendors change plans often; the sources below are where each was read from.
Want one of these in depth?Optlo vs ZendeskOptlo vs IntercomOptlo vs Help ScoutOptlo vs Freshdesk
Getting started
Deflecting by this afternoon
Three steps, one of which happens on your AI provider's site rather than ours.
Connect your AI provider
Create a key with OpenAI, Anthropic or Google and paste it in. Usage bills to that account at their list rates, so the deflection you win shows up in your provider dashboard rather than on an invoice from us.
Point it at what you already wrote
Give it your help docs, your website or a pasted FAQ. It answers from those, and says so when it does not know rather than inventing something. This step is the deflection rate; everything else is trimming.
Set the escalation rule, then watch re-contact
Decide when a person takes over: low confidence, a named topic, an unhappy customer. Then read the first week's deflected conversations and count how many customers came back. That number, not the deflection rate, tells you whether to widen the scope.
Questions about deflection, answered plainly
What is ticket deflection?
Ticket deflection is resolving a customer's question before it becomes a support ticket. The customer still gets an answer; a person never has to write it. In practice that means an agent answering from your documentation inside the chat widget, a help center article that lands, or a workflow that completes a request like an order lookup without a human touching it.
How do you calculate ticket deflection rate?
Divide the number of people who got their answer without a ticket by the number who tried to get help at all, then multiply by 100. In the worked example on this page, 480 of 1,000 support requests ended without a ticket, which is a 48% deflection rate. The denominator is where published figures diverge: counting only the people who opened your help center turns the same month into 77%.
What is a good ticket deflection rate?
Published benchmarks for this metric run from about 15 percent to 85 percent, which is too wide to plan against, and most of the spread is definitional rather than real. Vendors reporting the high end are usually dividing by self-service sessions; the ones reporting 15 to 30 percent are dividing by every contact attempt on complex technical products. Measure yours the strict way, write down which formula you used, and compare it only against itself month over month.
What is the difference between deflection rate and containment rate?
Containment is measured inside one channel: of the conversations your bot started, how many finished there without escalating. Deflection is measured across the whole front door: of everyone who wanted help, how many never needed a person. A bot can contain 90 percent of its own conversations while deflecting very little, if most of your customers bypass it and email you instead.
What is the difference between deflection and resolution?
Deflection is about whether a ticket was created. Resolution is about whether the problem went away. They come apart in the direction that costs you money: a customer who gives up on your help center and never writes in counts as deflected and is not resolved. Pair the deflection rate with re-contact rate and CSAT, or you are measuring silence.
Does a higher deflection rate cost more?
On per-resolution pricing it does, directly. Each deflected ticket is a billable event: Zendesk states $1.50 per automated resolution, Intercom about $0.99 per Fin resolution, Help Scout $0.75 per AI resolution. At 1,000 conversations a month, moving from a 40 percent deflection rate to 70 percent adds $450 to the Zendesk line alone. Optlo bills for active conversations instead, so a conversation costs the same whether the agent resolves it or hands it to a person, and the model tokens go to your own provider account at list rates.
Can ticket deflection hurt customer satisfaction?
It can, when you raise the rate by making a person harder to reach rather than by answering better. That version lifts the metric and lowers everything else, because the customers it deflects are the ones who gave up. Keep the handoff one click away, and watch re-contact rate as the counterweight.
Build AI customer support on your terms
Bring your own AI, stay in control, and optimize as you scale.
Try free for 7 days
* No credit card required