First Contact Resolution: Formula and Benchmarks

First contact resolution (FCR) is the share of issues finished in the customer’s first contact, with no repeat contact about the same problem. SQM Group’s 2024 benchmark of 500+ North American call centers put the all-industry average at 69%, with 70% to 79% called good and 80% or more world class. Those are phone figures.

Key takeaways

• FCR is a ratio with a window. Issues resolved on first contact divided by eligible issues, with a recontact window you write down (Microsoft’s FCR metric reference uses 7 days, as Customer Care Staff notes).
• Benchmarks are phone data. SQM’s 2024 numbers range from 73% for general inquiries to 60% for technical support and 48% for complaints. We found no public FCR benchmark built for email and chat at small software teams, and ticket-based figures use different definitions.
• An AI answer counts as first contact if it holds. If the customer does not come back about the same issue inside your window, count it. A silent customer is not proof.
• A billing “resolution” is not FCR. Vendors charge for a resolved conversation by their own rule, which is a different test from whether the problem stayed fixed.

What does first contact resolution count?

FCR counts issues that were finished the first time the customer reached out. A Shopify merchant asks why an app block won’t appear on their storefront, an agent sends installation steps and closes the ticket, and the merchant still can’t display the block. The ticket is closed and the issue is open. FCR exists to catch that gap.

For a software team, the unit is the customer conversation, and the test is whether the customer can finish the job they contacted you about. A fast reply and a tidy queue mean nothing if the task is unfinished.

The metric started in inbound voice support and spread to email, chat, interactive voice response, websites and social channels, as the first-call resolution overview explains. That overview says FCR is best judged from the customer’s perspective, citing two books, one from SQM Group, rather than an internal case status. Your help desk’s resolved tag is an internal status. The storefront block appearing is the resolution.

First contact resolution rate = issues resolved on the first contact ÷ eligible issues, expressed as a percentage.

The definition in BenchmarkPortal’s sample industry report is the same shape: contacts completely resolved during the first inbound contact, divided by contacts handled, with no additional contact needed. The same report notes that FCR figures are gathered by customer surveys, self-reported by agents, or both, and that the method affects the result.

Counting rules decide what the percentage says. The numerator must come from the same group of issues as the denominator, so don’t compare incoming conversations from one period with resolved conversations from another. Two teams can pull different rates from the same queue if one counts reopened threads and the other doesn’t.

Which labels belong beside the number?

A useful FCR definition carries these labels:

• Contact unit: What counts as the initial interaction.
• Eligible cases: Which incoming issues belong in the calculation.
• Resolution test: What evidence shows the task is finished.
• Recontact window: How long you watch for the same problem returning.
• Channels included: Where related customer contacts are checked.

Recording details is not the same as deciding the outcome. A law-firm guide to legal intake makes the same split, calling its intake record a summary and not a decision. In software support, collecting the merchant’s error message isn’t the same as fixing the error.

How do you measure first contact resolution?

Count a resolution only when the issue is finished in the initial interaction and no related return invalidates it. A reply alone doesn’t pass that test. Neither does an automated resolved flag.

Set up the measurement during week one

1. Set the denominator. Include all eligible incoming conversations in a defined cohort. Keep genuine unresolved cases in that cohort, even when they’re awkward or require engineering.
2. Write stable exclusions. Remove spam and duplicate threads. Treat a customer’s unrelated new question as a separate issue, not a failure of the earlier resolution. Exclude proactive outreach from the inbound denominator.
3. Record an initial resolution state. Let the agent or AI mark whether the request appears complete, with a short reason such as a completed account correction or customer confirmation.
4. Apply a recontact window. Mark the original case unsuccessful if the customer reopens it or starts another conversation about the same unresolved issue inside the window.
5. Publish the measurement method. State whether your score comes from customer responses or inbox records, and don’t swap one for the other without saying so.

Don’t finalize a cohort before its recontact window has elapsed. Otherwise newer conversations get an easier test because customers haven’t had time to return.

Three ways to measure first contact resolution

The Customer Care Staff benchmark comparison reports SQM Group’s 2025 aggregated average of 70% from a post-call survey, alongside 62% of consumers reporting first-time resolution in XM Institute’s 2025 study. Those figures aren’t directly comparable. Different respondents and collection methods can produce different answers without either calculation being dishonest.

For a Shopify app, a merchant might start in chat, continue by email, and return after a release. Match contacts by merchant, app and underlying problem. Convot’s Outcomes report takes the narrower route, calculating FCR as resolved conversations that were not reopened, so a new conversation about the same problem is not matched automatically. Counting only the first chat makes the dashboard cleaner without improving the outcome. Use your help desk reporting workflow to inspect the conversations behind the result, not just the headline.

What is a good first contact resolution rate?

For phone support, SQM Group’s 2024 data put the all-industry average at 69%, with a range from 43% to 88% across 500+ call centers (SQM’s 2024 FCR benchmark by industry, published February 2025). SQM calls 70% to 79% good and 80% or higher world class, and found that only 5% of call centers reach the top band. Its 2025 research reports an average of 70%, as reported by Customer Care Staff.

Every figure below is phone-call data from SQM’s 2024 benchmark, as summarized in Digital Minds BPO’s call center benchmark post and on SQM’s own page.

Choose a comparison band by issue type, not by chasing 80% across every conversation. A bug thread isn’t an inquiry just because both arrive in the same inbox. These are call center numbers, so for chat and email at a five-person software team they are orientation, not a target.

Separate your queue by issue type, channel and product. A blended rate can fall because the mix changed: a release that generates more technical investigations lowers the overall number even when responders handle each category just as well. An influx of easy questions can lift it without making support any better.

Read movement as a diagnostic signal. A rising rate can mean clearer documentation, better routing or more complete first replies. A falling rate can point to missing knowledge or unnecessary handoffs, so look for the point where the first responder stopped making progress. Without a stated scope, method and return rule, an external benchmark is orientation, not a performance contract. Your own stable, segmented baseline is more useful for deciding what to fix next.

How do you count FCR when an AI agent answers first?

An AI answer counts as first contact resolution when the customer does not come back about the same issue inside your recontact window. The contact is the customer’s, not the staff member’s. What changes with AI is the evidence: a resolved flag set by the AI is the weakest kind, so rely on the return window.

Definitions differ on handoffs. Some teams treat any transfer as a miss and others count only customer-initiated returns. Pick one in writing and report the AI-answered and human-finished conversations on separate lines, so a better AI mix doesn’t hide behind one blended percentage.

A billing “resolution” is a different test. Cove AI, for example, charges $0.20 on a paid plan once per conversation in which it sent at least one automatic reply and the conversation was then resolved, whether Cove or the team finished it. That is a pricing rule. It tells you nothing about whether the issue stayed fixed.

A worked example: 1,000 conversations with and without an AI first touch

The figures here are an illustration with stated assumptions, not a measured result. Assume 1,000 customer conversations in a month, a 7-day recontact window, and customer-initiated recontacts about the same issue as the failure test.

Without AI. A person answers everything. Assume the team holds 70% of conversations with no recontact, close to SQM’s 2024 and 2025 averages. FCR is 700 of 1,000, or 70%.

With an AI first touch. Split the 1,000 conversations using the mix we measured on our own Appointo help center in the 30 days to 14 October 2026 (Convot data): Cove AI evaluated 808 customer messages in 230 conversations, answered 16% directly, drafted a reply for the team on 55%, and escalated 29% to a person after no answer met its confidence threshold. Those are shares of messages, not conversations, so applying them to conversations is an approximation. The assumed hold rates are ours.

The 70% for drafts reuses the human-only rate, and the 60% for escalations borrows SQM’s 2024 technical support figure because these are the conversations the AI could not answer. Neither is a Convot measurement.

The baseline treats every conversation alike, while the AI scenario gives the hard 29% a lower rate, so the comparison understates the AI mix. Apply 60% to the same hard 29% in the human-only baseline and the gap narrows. The useful point is that FCR is not guaranteed to rise just because an AI answers first, since 84% of these conversations still end with a person. How sensitive is the total to the AI’s own hold rate?

A naive count would report 16% “AI resolution rate” from the flag alone, which is the headline that overstates success. The honest read is that FCR tells you whether the answers hold, and a separate measure, such as first reply time or minutes of staff time per conversation, tells you what the AI saved. We do not publish a recontact rate for Cove answers, because we have not measured one. The Appointo sample has only 3 helpful and 2 unhelpful votes on AI answers, too few to report.

What do repeat contacts cost a software team?

Repeat contacts add rework for a small team and a customer who expected to be finished, though we found no traceable cost figure. Contact-center cost studies exist, but the figures that circulate for them are hard to trace. A Substack write-up of call center economics repeats repeat-call and satisfaction figures from an unnamed benchmarking firm, so we do not quote them. Measure your own repeat work instead.

For a small team, the extra work is tangible. Someone rereads the logs, reconstructs the account state and checks whether the previous advice was followed. The merchant waits again, and the founder may get pulled out of product work to explain the same configuration.

Measure the whole resolution

Pair FCR with cost per resolution, not a vanity close rate. Include the effort needed to reach a working outcome across the original conversation and its follow-ups:

• Responder time: Reading, diagnosis, replies and verification.
• Handoff time: Preparing context and reviewing it again.
• Engineering time: Reproduction, investigation and support-facing updates.
• Correction time: Undoing an incorrect action or explanation.

A cheap first reply isn’t a cheap resolution if another person must finish the job. FCR doesn’t tell you which customers will leave either. The practical retention argument is narrower: finishing setup, billing explanations and how-to questions removes avoidable frustration. Review the reopen pile by cause, not by the responder who handled it.

Of last month’s reopens, which issue types produced the second contact, and what did that second contact cost in agent time?

When does a first-contact close hurt trust?

Suspected bugs, billing disputes, security or privacy concerns, multi-account setups and issues needing engineering validation shouldn’t be forced into a one-touch close. A confident answer isn’t enough when the responder can’t verify the underlying result. These cases need deliberate ownership rather than a premature resolved status.

• Suspected bug: Send reproduction steps, expected behavior, actual behavior and logs to engineering. Tell the merchant when you’ll update them.
• Billing dispute: Preserve the charge history and the merchant’s objection. Assign a decision-maker and give a next-update time.
• Security or privacy concern: Use the designated review path and name the next communication checkpoint.
• Multi-account setup: Record which accounts and permissions are involved, and pass that map to the owner without making the customer rebuild it.
• Engineering validation: Include what support has tested and what still needs confirmation, with a named owner for updates.

Containment is not resolution

Containment means the customer stops writing. True resolution means the underlying issue is fixed.

The Macha compilation of support benchmarks separates measured benchmarks from vendor claims and makes the same point about definitions: one vendor’s report defines its “resolution rate” as the percentage of tickets closed by anyone, so it is not an AI metric. A dashboard number is only as good as its definition.

Silence might mean the customer gave up, not that the software works. Read FCR beside reopen rate, repeat-contact rate, escalation accuracy, customer effort and defect recurrence. A correct escalation can lower first-contact completion while protecting the customer from an unsupported answer, so don’t punish responders for choosing the necessary path.

How do you raise first contact resolution?

Raise it by fixing the largest repeat-contact cluster, giving responders the facts and authority to finish the task, and writing documentation that ends in a verifiable result. A 2011 peer-reviewed paper by Abdullateef, Mokhtar and Yusoff studies how first call resolution mediates call center performance. We could not read its coefficients, so we do not quote them. The practical point stands without them: knowledge and customer context help only if the responder can finish the task.

Fix the process before adding pressure

Tag every reopen by issue type, product, plan, language and escalation reason. Then select the largest actionable cluster and be precise about the cause. An installation cluster might contain outdated screenshots, missing permissions and a real product defect, and those need different owners. Assign one concrete fix and an owner. Changing the workflow is more useful than asking everyone to improve their FCR.

Give people facts and bounded authority

A responder handling a billing explanation needs the account’s plan and relevant billing history. A responder handling access needs to know who owns the account and which changes they’re authorized to make. Allow routine corrections, and preserve approval for disputed, sensitive or irreversible actions.

Write decision-ready documentation

An effective support article tells the reader when it applies, what to do in order, how to verify success, and when to stop and ask a person. Use the merchant’s vocabulary in titles, including error text. A practical knowledge-base building process should prioritize repeat-contact causes over publishing volume.

Bound automation to what it can support

Let an assistant answer from the team’s own documentation and show the source. When it isn’t sure, it should hand the conversation to a person with the collected context. Retrieval alone isn’t enough when completion requires an account action the assistant can’t perform. Your customer support automation rules should specify which questions can be answered from documentation and which require a person.

At a 30-day review, keep the week-one rules unchanged. Read FCR and reopen rate side by side and confirm you’ve removed the repeat-contact cause rather than changed its label.

Which fix should you choose first?

Pick one repeat-contact cluster from last month and fix the blocking condition. Don’t buy a new tool until you can name what prevented completion.

• Missing documentation: Write the article around the failed task and measure that issue type for 30 days using the same window.
• Missing account facts: Put the plan and billing history beside the conversation before assuming you need another hire.
• An unnecessary tool change: Keep help content inside the conversation workflow so the merchant doesn’t restart elsewhere.

Match the help desk to the work

Disclosure: Convot makes the help desk and AI agent described here. Convot is a help desk built by a solo founder for small software teams, with Shopify app developers as the sharpest fit. Cove AI answers from the team’s own docs, cites its sources, and hands off to a human when it isn’t sure. It starts in Suggest mode, where your team approves each draft, which makes the drafted share of your queue visible before anything is sent automatically.

For Shopify apps, the conversation shows the merchant’s plan, MRR and past revenue pulled from the Shopify Partner API, which helps responders tell a question about an entitlement from a question about a charge.

Crisp imports in one click. Moves from Zendesk, Intercom and Freshdesk are free migrations done with the founder, not automatic importers. A funded team seeking a mature AI agent and outbound campaigns may reasonably prefer Intercom. The right choice depends on the workflow you need.

Publish your measurement window and exclusions this week. Then remeasure the issue type you’ve chosen using completed outcomes, not a larger pile of closed tickets. If you want to test an AI first touch, read how Cove AI answers from your own docs and run it in Suggest mode on your next month of conversations.

FAQ

What is first contact resolution?

First contact resolution is the share of customer issues finished in the first contact, with no repeat contact about the same problem. Divide issues resolved on first contact by eligible issues, multiply by 100, and state your recontact window beside the result.

What is a good first contact resolution rate?

For phone support, SQM Group’s 2024 benchmark puts the average at 69% and calls 70% to 79% good and 80% or higher world class. By call type it ranges from 73% for general inquiries to 48% for complaints. These are call center figures, so use them as orientation for chat and email.

Does an AI resolution count as first contact resolution?

An AI resolution counts when the customer does not return about the same issue inside your recontact window. An AI-set resolved flag alone does not count, because a later return is unknown. Report AI-answered and human-finished conversations on separate lines.

How do you calculate FCR for a small team?

Count eligible conversations in a month, wait for the recontact window to pass, then count the ones with no return about the same issue. In the illustrative example above, with assumed hold rates, 679 of 1,000 conversations resolved on first contact gives 67.9%. Match returns by customer, product and issue.

Is a higher FCR always better?

A higher FCR is not always better, because forcing a one-touch close on a suspected bug, billing dispute or security concern can produce a false resolution. A correct escalation lowers first-contact completion and protects the customer, so read FCR beside reopen rate and escalation accuracy.

Convot brings live chat, a shared inbox, help content and a grounded AI agent together for small software teams. Bring one recurring setup or billing question and check whether the docs, account context and human handoff can help your team finish it without another contact. The shared inbox page shows how every app’s conversations land in one queue.

Tarang Agarwal is the founder of Convot. SQM Group’s 2024 benchmark figures were read from SQM’s published page on 9 Nov 2026. The worked example is an illustration with stated assumptions.