Ticket deflection rate: formulas and how to measure it
Ticket deflection rate, measured properly: five formulas run on one month of data, a setup checklist, and how to credit articles and videos for the tickets they prevent.
Berkay Akpolat14 min read
Updated November 3, 2026
Ticket deflection rate is the share of people who look for help in your help center and don't go on to open a ticket. A defensible formula: support-seeking sessions with no ticket within 72 hours, minus sessions that visibly failed, divided by all support-seeking sessions, × 100. The common formulas give very different numbers, so pick one and keep its rules fixed.
Below: five formulas run on one month of data, a setup checklist, article and video attribution, and what deflection saves.
What is ticket deflection rate?
Ticket deflection rate measures how often self-service (help articles, videos, an AI assistant, community answers) stops a question from becoming a ticket. It's a self-service metric, not a support-team metric: it rises when customers find answers, not when agents work faster.
Three terms get mixed up:
| Term | What it counts | Typical source |
|---|---|---|
| Deflection | People who sought help and didn't open a ticket | Help center plus help desk data |
| Resolution | People whose issue was actually solved in self-service | Surveys, "Was this helpful?" votes, no re-contact |
| Containment | Chatbot or phone-menu conversations that never reached a person | Bot or IVR reports |
Deflection is the easiest to measure and the easiest to inflate: someone who gives up and churns quietly also "didn't open a ticket" (Gainsight calls this false deflection). Resolution is what you actually want. Good measurement pushes deflection toward resolution.
How do you calculate ticket deflection rate? Five formulas compared
There is no standard. Help desks, chatbot vendors and analysts each use their own. These are the five you'll meet, plus chatbot containment:
1. Self-service score (used by Zendesk, "also known as the self-service ratio and the ticket deflection ratio")
Self-service score = help-center sessions ÷ users in tickets
Gets right: it needs only two numbers you already have. Gets wrong: it's a ratio (e.g. 4:1), not a rate, and it rises with traffic even if nobody is helped.
2. Session-based rate (many help-center and chatbot vendors)
Rate = (help-center sessions − sessions followed by a ticket) ÷ sessions × 100
Gets right: easy to explain as a percentage. Gets wrong: every abandoned visit counts as a success, so it runs high.
3. Adjusted session rate (analytics-led teams; a practical step toward stricter definitions such as Intercom's and Metabase's)
Rate = (support-seeking sessions − sessions followed by a ticket − other sessions that visibly failed) ÷ support-seeking sessions × 100
Gets right: removes bots, staff, browsing and visible failures, and uses a re-contact window. Gets wrong: needs visits joined to tickets, and still counts people who left silently.
4. Contact-form deflection (teams that suggest articles on the ticket form)
Rate = people who left the form after a suggested article, with no ticket within 72 hours ÷ people who started the form × 100
Gets right: counts people who were about to write in. Gets wrong: covers only one entry point.
5. Baseline reduction (support leaders reporting to finance)
Reduction = (tickets per 1,000 customers before − after) ÷ before × 100
Gets right: measures the outcome the business cares about. Gets wrong: mixes in product changes, seasonality and growth.
Not deflection: chatbot containment (chatbot vendors)
Containment = (bot conversations − conversations handed to a person) ÷ bot conversations × 100
It measures the bot, not your help content, and counts a frustrated user who closes the chat as a success.
Intercom's definition is the strictest published one: "the number of resolved sessions which were resolved by self-serve support divided by the total number of resolved sessions seeking support", where "users who stopped seeking support are not included" and a session is a 24-hour period. Formula 3 can't go that far without asking every visitor, so it still counts people who left silently: read it as a ceiling. For a floor, count only sessions with a positive signal (a Helpful vote, say) and no ticket within the window, as Metabase recommends. Report both if you can; the gap between them is your silent abandonment.
Our recommendation: report formula 3 (with its positive-signal floor) monthly as your deflection rate, and formula 5 quarterly as the business result. Use formula 4 if you suggest articles on your contact form. Keep Zendesk's score (formula 1) only if your team already tracks it, and never put it on the same chart as a percentage.
One month, five answers: a worked example
Here's one month of made-up data for a SaaS company with 25,000 customers, run through all five formulas.
| Input | Value |
|---|---|
| Help-center sessions | 20,000 |
| Support-seeking sessions (searched or opened an article; bots and staff removed) | 12,000 |
| Support-seeking sessions followed by a ticket from the same person within 72 hours | 1,400 |
| Other support-seeking sessions that visibly failed (ended on a search with no results, or a "Not helpful" vote; no ticket followed) | 2,600 |
| Tickets this month | 2,400, from 2,000 people |
| Contact-form starts | 1,000 |
| Form abandoned after opening a suggested article | 220, of whom 40 wrote in within 72 hours |
| Tickets per 1,000 customers: same month last year → this month | 120 → 96 |
| Formula | Calculation | Result |
|---|---|---|
| 1. Self-service score | 20,000 ÷ 2,000 | 10:1 |
| 2. Session-based rate | (20,000 − 1,400) ÷ 20,000 | 93% |
| 3. Adjusted session rate | (12,000 − 1,400 − 2,600) ÷ 12,000 | 67% |
| 4. Contact-form deflection | (220 − 40) ÷ 1,000 | 18% |
| 5. Baseline reduction | (120 − 96) ÷ 120 | 20% fewer tickets per 1,000 customers |
Same company, same month: 10:1, 93%, 67%, 18% and 20%. None of them is wrong. They answer different questions. This is why vendor benchmark bands (Chatling's calculator, for example, labels 40–60% "Good") mean little unless you know which formula produced them, and why you should only compare your number with itself over time.
How do you measure ticket deflection, step by step?
Define a support-seeking session
Count sessions where someone searched your help center, opened an article, watched a help video or asked the help center's AI assistant. Drop bots, your own staff (filter by IP or signed-in role) and visits that only touched the home page. This is your denominator.
Connect help-center visits to tickets
You need to know whether the same person opened a ticket after their visit. The usual ways: identify signed-in users in your analytics tool and pass the same user ID to your help desk, or match by email when someone submits the contact form. Log at least: session ID, search query and result count, article views, helpful votes, contact-form starts, suggested-article clicks, form abandons, and tickets with requester ID and topic tag. Visitors you can't join (often most of a public help center) belong in a separate, labeled estimate, not mixed into the joined rate. Without any join, you can only report formulas 1 and 5.
Pick a re-contact window and freeze it
A session counts as deflected only if the person didn't open a ticket within a fixed window. Published definitions use 24 hours (Intercom's session), 48 hours (Helply's re-contact rule) or 72 hours (Metabase); a short window misses people who try again the next day. Whatever you choose, keep it fixed, or your trend line measures your rule changes.
Subtract visible failures
Remove sessions that ended on a search with no results, or with a "Not helpful" vote. These people didn't find an answer, even if they didn't write in. Count each session once: a failed session that also led to a ticket is already in the ticket count.
Report per topic, not only in total
Group sessions and tickets by the same topics: billing, SSO, exports, and so on. A total of 67% can hide a topic at 30% that generates half your tickets. Use your ticket tags as the topic list, and map each help article to one tag.
Put it next to ticket volume
Report deflection rate together with tickets per 1,000 active customers. If deflection goes up but tickets per customer don't go down, you're counting the wrong sessions.
How do you attribute deflection to articles and videos?
This is where the useful decisions are. Three methods, from simplest to strictest:
1. Topic before and after. When you publish or rewrite the article for a topic, compare that topic's tickets per 1,000 customers for the four weeks before and the four weeks after. Check that other topics stayed flat over the same weeks; if everything dropped, it was probably seasonal.
2. Article-level adjusted rate. Run formula 3 for each article: sessions that viewed this article, minus those followed by a ticket on the same topic within 72 hours, minus "Not helpful" votes. Rank articles by tickets that still follow them. The top of that list is your rewrite queue.
3. Format test for video. To learn whether adding a how-to video lowers tickets, add videos to half of a set of similar articles (say 10 of your 20 most-viewed how-to articles, picked alternately or at random) and leave the rest. After four to six weeks, compare tickets per topic between the two halves, and only trust the result if each half has enough tickets to show a difference (a few dozen a month, not a handful). Look at video completion too: a video that most people stop at 25% isn't doing the work. Nielsen Norman Group's usability research found people use video as a secondary source, especially for long, multistep processes, so test video on multistep tasks first.
| Question | Method | Data you need |
|---|---|---|
| Did the new article reduce tickets on its topic? | Topic before and after | Ticket tags, publish date |
| Which articles still let tickets through? | Article-level adjusted rate | Visits joined to tickets |
| Does video lower tickets for our tasks? | Format test | Similar articles, video completion, ticket tags |
| Is self-service paying off overall? | Baseline reduction | Tickets per 1,000 customers, by quarter |
What is a good ticket deflection rate?
There's no reliable public benchmark, because nobody uses the same formula. The ranges you'll see in vendor posts are rarely sourced. Measure against your own trend.
For context on how hard self-service is, Gartner's survey of 5,728 customers found that only 14% of customer service issues were fully resolved in self-service, and even for issues customers called "very simple", only 36% (Gartner, 2024). In the same survey 73% of customers used self-service at some point in their journey. Most people try; most don't finish there.
The same survey names the most common reason for failure: in 43% of cases, customers couldn't find content relevant to their issue. Earlier Gartner surveys, with different samples, put full self-service resolution at 9% in 2019 and 13% in 2020: low, every time it's been measured (more in our customer self-service statistics). That's a content problem you can measure directly, through searches that return nothing.
How much does ticket deflection save?
Savings per month = deflected tickets × cost per ticket. Deflected tickets = monthly tickets × the share that are how-to questions × the share of those self-service resolves.
For a cost per contact, use your own fully loaded figure if you have one. As a reference, Gartner's February 2024 cost benchmark puts the median at $1.84 per self-service contact and $13.50 per assisted contact such as phone, chat and email (Gartner, 2024; full report for clients). Self-service isn't free: for a net figure, subtract what a self-service contact costs. With Gartner's medians, that's $13.50 − $1.84 ≈ $11.66 saved per deflected contact.
Our free ticket deflection calculator runs this with your numbers: tickets per month, the share of how-to questions, the deflection rate, handle time per ticket, and the cost per ticket or agent hourly cost. It returns tickets deflected, agent hours saved and money saved per month and per year. For example, 2,000 tickets a month, 40% how-to questions and 30% resolved by self-service gives 240 tickets deflected a month; at $15 a ticket, that's $3,600 a month. Use a ticket-based rate there, not a session rate: formula 5's drop in tickets per 1,000 customers is a good real-world input, while 67% of sessions is not 67% of tickets.
How do you raise your deflection rate?
Work from the measurements above, in this order:
- Write the missing articles. Your list of searches with no results is a ranked to-do list. Customers told you the words they use. Here's how to turn help center searches into a writing list.
- Fix the articles that let tickets through. Rewrite the ones with the most tickets after a view and the lowest helpful rate: clearer title, answer first, exact labels, prerequisites. Our guide to writing help center articles shows how.
- Add a video to multistep tasks. Then check with the format test above that it helped.
- Put answers where people get stuck. Link the article from the screen in your product and suggest articles on your contact form.
- Keep content current. An article that no longer matches the UI creates tickets instead of preventing them. Here's how to keep documentation up to date when your UI changes.
Building the knowledge base from scratch? Start with how to build a knowledge base with AI.
Measuring and raising deflection with Clevera
Clevera is an AI platform that turns a screen recording into a narrated how-to video, a step-by-step help article and a page on your hosted help center, in 74 languages. The help center gives you the self-service side of the measurements above; your help desk still holds the tickets.
Insights on every plan. Each Clevera help center shows Page views, Visitors, Search success, Zero-result rate, Click-through, the Helpful rate per page and Assistant answered, the share of Ask AI questions answered from your pages. Readers are counted without cookies, and bots are left out. You can look back 30 days on Free, 90 days on Starter and 12 months on Pro and Custom.
Content gaps with a Write article button. Searches that found nothing are listed with how many people searched and when. Write article opens Compose with the search as the topic, and the article joins the help center for your next publish (it costs credits like any article).
Article suggestions on the contact form. On Starter and up, the help center's contact form suggests up to three of your pages while the reader types, and each message tells you which pages were suggested. That's the setup for formula 4; count form starts and abandons in your own analytics.
Send page views to your own analytics. On Starter and up, add Google Analytics 4, Plausible, PostHog or Segment so help-center page views land next to your product data. Joining those visits to tickets is still your job: it needs a user ID or email your help desk also has.
Fill the gaps faster. Record the task once and Clevera writes the help article with screenshots, and can make a narrated how-to video from the same recording for your format test; video analytics show completion and how far viewers got. On Pro and Custom, Autopilot can turn questions you paste in (up to 30) and searches that found nothing into draft articles that wait for your approval.
The help center is included on every plan, Free included. Starter is $29 a month billed yearly ($36 monthly) and adds your own domain and analytics integrations; Pro starts at $86.80 a month billed yearly for 2 creators ($124 monthly). Prices as of November 3, 2026; see pricing.
Start free or read how doc site insights are counted.
FAQ
How do you calculate ticket deflection rate? Take support-seeking help-center sessions that didn't lead to a ticket within a fixed window (often 72 hours), subtract sessions that visibly failed, divide by all support-seeking sessions, and multiply by 100. For example, (12,000 − 1,400 − 2,600) ÷ 12,000 = 67%.
What's the difference between Zendesk's self-service score and deflection rate? Zendesk's self-service score divides help-center sessions by users in tickets and gives a ratio such as 4:1. A deflection rate is a percentage of sessions that ended without a ticket. They can move in opposite directions: more traffic raises the ratio even if no extra tickets are avoided.
Is a high deflection rate always good? No. A rate built on abandoned sessions can rise while customers get more frustrated. Check it against tickets per 1,000 customers, helpful votes and searches with no results. In B2B, a deflected ticket from a key account can also be a signal your account team never sees.
How long should the re-contact window be? Published definitions use 24, 48 or 72 hours. Longer windows are stricter. The window matters less than keeping it fixed, so your trend is comparable month to month.
Can you measure deflection without linking visits to tickets? Partly. Without the join you can report Zendesk's ratio and the change in tickets per 1,000 customers, plus content signals such as zero-result searches and helpful votes. Zendesk's search report also counts "Tickets created": searches that led to a ticket. You can't tell which visits avoided a ticket.
Do videos deflect more tickets than articles? There's no reliable general answer; it depends on the task. Test it on your own content: add videos to half of a set of similar articles and compare tickets per topic after four to six weeks.
