Ticket Deflection Rate: Formula and Meaning

Ticket deflection rate is the share of customer questions resolved without a support agent touching them. The standard formula is deflected contacts divided by total contacts, times 100. If 4,000 people started a conversation last month and 1,400 got what they needed from self-service or an AI agent without a human replying, your deflection rate is 35%.

Last updated August 2026.

The metric matters more than it used to, because it is now attached to an invoice. Several vendors bill you per AI resolution, and their definition of a resolution is not the same as your definition of a deflection. Getting the two confused is how support teams end up reporting a 40% deflection rate to the board while paying for something closer to 55% of conversations.

What does ticket deflection mean?

Ticket deflection means a customer got their answer without a support agent working the ticket. That can happen through a help center article, a chatbot or AI agent, an in-product answer, or a status page that explains the outage before anyone writes in. The word is a little misleading: nothing is being pushed away. The question is still answered, just not by a person.

Two things are commonly called deflection and only one of them is. A customer who reads an article and leaves satisfied has been deflected. A customer who reads an article, does not find the answer, and then opens a ticket has not been deflected, even though your knowledge base logged a view. Any measurement that counts article views as deflections will flatter you badly.

How do you calculate ticket deflection rate?

The formula is simple. The hard part is deciding what goes in the numerator.

Ticket deflection rate = (deflected contacts / total contacts) x 100

Total contacts should be every attempt to reach you, not just the ones that became tickets. If you only count created tickets, self-service success is invisible by construction, and your deflection rate will read 0% no matter how good the help center is. Count sessions that reached your support surfaces: chat conversations started, help center searches, AI agent conversations, and tickets opened.

For the numerator, the defensible definition is a contact that reached a support surface and did not result in a human reply within a set window. Most teams use 24 or 72 hours. Pick one, write it down, and stop moving it, because a deflection rate is only useful as a trend line.

What is a good ticket deflection rate?

Published figures cluster between 20% and 60%, and the spread is mostly about product type rather than tooling quality. Ecommerce stores with heavy order-status volume deflect at the high end, because "where is my order" is a lookup, not a judgment call. Complex B2B software sits lower, because the questions that arrive are the ones the documentation did not answer.

Be careful with vendor case studies here. A number like "we deflected 70%" usually excludes channels the AI does not cover, or counts a conversation as deflected the moment the customer stops replying. Treat any single figure with no stated definition as marketing rather than a benchmark, and compare yourself to your own last quarter instead.

Deflection rate is not the same as a billable resolution

This is the part that costs money. Vendors that meter AI usage each define the billable unit differently, and none of those definitions is your deflection rate. Here is what the major vendors published as of August 2026.

Read those rows against each other and the problem is obvious. Gorgias bills a ticket fee even when the AI handled it, so an automated interaction can carry two charges. Freshdesk counts a session, and one 72-hour email session may contain several exchanges you would report as a single deflection. Agentforce can bill twenty times for one conversation if the agent takes twenty actions, which is the point at which its per-action model costs the same as its $2 per-conversation model.

Practical consequence: build your deflection report from your own helpdesk data, then reconcile it against the vendor invoice separately. If those two numbers are within a few percent of each other, fine. If the invoice is 30% higher than your deflection count, you are paying for a unit you are not measuring. It is also worth checking that the export feeding the report is complete and current, because a silently stale or partial pipeline will quietly understate volume, and catching a broken data feed before it reaches the dashboard is cheaper than explaining a bad quarter after the fact.

How do you improve ticket deflection rate?

The highest-yield work is boring and specific. Pull your top 20 ticket drivers by volume for the last 90 days, and for each one ask whether a customer could have answered it themselves in under a minute. Usually a third of them could, and the reason they did not is that the answer is buried, out of date, or written for someone who already knows the product.

• Fix the top five drivers first. Support volume is heavily concentrated. Five articles rewritten properly usually beat fifty written quickly.
• Put the answer where the question happens. An order-status answer belongs in the chat widget and the shipping confirmation, not three clicks into a help center.
• Let the agent take an action, not just cite an article. Deflection improves sharply when the AI can look up the order, reschedule the appointment, or start the return, because that is what the customer actually wanted.
• Measure escalations by reason. Every escalation is a labelled gap in your content. Tag them and the roadmap writes itself.
• Stop deflecting the wrong things. Billing disputes, cancellations and anything with a legal or safety edge should reach a person quickly. Deflection on those is a churn metric in disguise.

Does higher deflection hurt customer satisfaction?

Not by itself. Satisfaction falls when deflection is implemented as an obstacle course, where the customer has to defeat three self-service prompts before a human appears. It rises when the self-service answer is genuinely faster than waiting for a reply, which for straightforward questions it usually is. Track CSAT split by resolution path, AI versus human, and watch the gap rather than the average.

If AI-resolved CSAT is more than a few points below human-resolved CSAT, the fix is scope, not tone. The agent is answering questions it should be handing over. Tighten what it is allowed to attempt and the gap generally closes.

What is the difference between deflection rate and resolution rate?

Resolution rate is the share of contacts that were resolved at all, by anyone. Deflection rate is the share resolved without a human. A team can run a 95% resolution rate and a 10% deflection rate, which just means people are doing almost all the work. Deflection is a subset, and reporting one as the other is a common way to make an AI rollout look better than it was.

Which channels can you actually deflect?

Deflection is easiest where the customer expects an instant answer and hardest where they expect a considered one. Web chat and in-app messaging deflect well. SMS and social DMs deflect well for status and scheduling questions. Email deflects worst, partly because people write longer, more compound questions by email, and partly because a 72-hour session window makes the measurement mushy.

If you are choosing tooling on the strength of a deflection promise, ask the vendor which channels the figure covers. A 55% deflection rate on web chat alone is a very different purchase from 55% across every channel you run, and the second one is what changes your headcount plan.

Getting the measurement in place

Start with one channel, one window and one written definition. Export the last 90 days, count contacts that reached a support surface, count how many got a human reply inside the window, and subtract. That is your baseline. Then instrument it monthly and change one thing at a time, because deflection responds to content changes, product changes and seasonality all at once, and you will not learn anything from a chart with five simultaneous interventions on it.

When you are ready to move deflection with an AI agent rather than with articles alone, the thing to look for is an agent that answers on the channels your customers already use and can complete the task, not just describe it. Our customer service chatbot answers, qualifies and books across web chat, SMS, WhatsApp and social DMs on a flat monthly plan, so a good deflection month does not arrive as a larger invoice. If you are still comparing vendors, the best customer service software comparison prices an identical workload across six different billing units, and what actually counts as a billable AI resolution goes through each vendor's definition line by line.