First Contact Resolution — The Metric That Falls When Self-Service Works (2026)

Quick answer: FCR asks what share of customer issues you settle the first time the customer comes to you. It is among the oldest and most-quoted numbers in customer service, it is a contact-center metric rather than a bot metric, and it behaves in a way that catches people out when they deploy automation. Because the denominator counts contacts, and because a chatbot changes which contacts survive to reach a human, a self-service program that is working should be expected to push FCR down while nothing about your service gets worse. How far down depends entirely on your request mix, and the worked example below moves it 1.3 points. If you have set an FCR target and also bought a bot, you have set those two things against each other without meaning to.

Three words, three decisions

The formula is not the hard part:

FCR = contacts resolved on first attempt / total contacts

Some organizations run the same metric with issues in the denominator instead. That is a fourth choice sitting on top of the three below, and it changes what a transfer or a bot handoff does to the number. This page uses the contact denominator throughout, which is the one SQM's benchmark uses.

The name is the hard part. Each of its three words hides a choice that your reporting has already made, usually by default, usually without telling you.

First. First within what window? A customer who calls back eleven days later about the same order has either broken your FCR or not, depending on whether your reopen window is seven days, fourteen, or thirty. Shorten the window and FCR rises without a single conversation improving.

Contact. A contact is one attempt by one customer to reach you. But is a transfer between two agents one contact or two? Is an email followed by a chat one issue with two contacts, or two issues? And the modern question this page is really about: does a chatbot session count as a contact at all? Different answers produce different numbers from identical events.

Resolution. Resolved according to whom. The customer, who alone knows whether the problem went away, or the system, which knows only that nothing came back. Our chatbot resolution rate entry works through the same fork on the bot side, and the conclusion is the same on both: an inferred resolution is a much weaker claim than a confirmed one.

None of these is a trick question with a right answer. The practical rule is that an FCR figure travels with its window, its contact convention and its resolution test, or it does not travel at all.

The benchmark everyone quotes, and where it came from

If you have seen a number attached to FCR, it most likely traces to SQM Group, which has published annual benchmarks for more than twenty-five years and in our reading is the source most vendor pages paraphrase without naming it. Its standards are widely repeated: a good FCR rate is 70 to 79 percent, world-class is 80 percent or higher, and only 5 percent of call centers reach world-class.

The current published figure is less flattering than the standards suggest. In SQM's 2024 benchmark, released in February 2025, the aggregated average across all industries was 69 percent, with individual centers ranging from 43 to 88 percent. That is not merely below the good band. SQM's own 2022 report sets the low end explicitly at "69 percent or below," which it labels as needing improvement. The industry average has landed inside the band its own benchmarker calls a problem.

It has also been drifting, and here the honest statement is that two of the four numbers are published and two are inferred. SQM prints 71 percent for its 2022 report and 69 percent for its 2024 one. Its "1 percent lower than the previous year" and "2 percent lower than the previous year" phrasings imply roughly 72 and 71 for the intervening years, and we read those phrasings as percentage points rather than relative change because that is the reading consistent with the printed averages on both pages. Treat 72, 71, 71, 69 as a shape rather than a dated series: SQM published its 2024 figures in February 2025, a thirteen-month lag, but published its "2022" figures in May 2022, so the two reports' data years may not sit where their labels suggest.

Now the part that matters more than any of those numbers. Here is the population SQM measures, in its own description:

That is a telephone survey of people who telephoned a large North American contact center. It is a careful instrument and it is not a description of a small business running a bot on WhatsApp. When a chatbot vendor puts "the industry average FCR is around 70 percent" on a product page, the number has been lifted out of that population and dropped next to a product that was not in it.

There is a sharper version of the same point, and it comes from SQM rather than from us. Its 2022 report states its own view that the post-call survey method is the only way to accurately benchmark FCR, a view worth reading next to the fact that SQM sells post-call surveying. The chatbot platforms in our review set report resolution from system logs, which is what their documented analytics measure, and none of them documents a post-contact customer survey of the kind SQM describes. So by the standard of the organization that publishes the benchmark, the number your dashboard shows is not comparable to the benchmark it is being compared against. That is not a criticism of the platforms. It is a reason to stop treating 70 percent as a target you inherited.

Does a bot session count as a contact?

This is the fork that decides whether automation appears to help or hurt, and in our reading it is almost never stated. Take one customer, one question, one bot session that fails, and one agent who fixes it. Three conventions, three different truths:

• The bot session is not a contact. The denominator is unchanged and the agent resolved it first time. FCR unaffected. Automation is invisible.
• The bot session is contact one, the agent is contact two. The issue took two contacts and is a non-FCR by construction. Every single handoff becomes a failure and FCR collapses in proportion to how often the bot escalates.
• The bot session replaces the contact. The bot session is the first contact and the agent turn is treated as a continuation of it, so the issue resolves on first contact. FCR unaffected or slightly improved.

Same events. Three answers, spread across the whole plausible range. Before anyone argues about whether the bot raised or lowered FCR, someone has to say which of those three the reporting is doing, and in our experience of vendor documentation that convention is rarely written down anywhere a buyer would find it.

Our own view, offered as a view rather than a standard: the second convention is the honest one for a business that wants to know whether automation is helping, and it is also the one that makes FCR unusable as a management target during a bot rollout. Those two facts are not in tension. They mean FCR is a good diagnostic and a bad scoreboard, which is the same conclusion our customer effort score entry reaches about a different instrument for a different reason.

Why a working self-service program should lower your FCR

Here is the mechanism, and it is arithmetic rather than opinion.

FCR is a weighted average across request types, and request types resolve at very different rates. SQM's 2024 report publishes the split: general inquiries 73 percent, account maintenance 72, orders 71, billing 69, claims 61, technical support 60, and complaints last at 48.

Self-service does not take a random sample of those. It takes the top of the list. Lookups and simple status questions are exactly what a bot handles well, and they are exactly the categories with the highest FCR. Deflect them and the contacts that still reach a human are more heavily weighted toward claims, technical support and complaints, which resolve worst. The per-category rates need not move at all for the blended number to fall.

A worked example, using SQM's published category rates and a plausible small-business mix. Start with 1,000 contacts:

The blended FCR is 69.6 percent. We chose the mix to land near SQM's published 69 percent, so that agreement is construction rather than corroboration: it makes the example legible, not realistic. Now deploy a bot that absorbs 60 percent of general inquiries and 40 percent of orders, and change nothing else. Contacts fall from 1,000 to 700, a 30 percent reduction that any operator would count as a win. The blended FCR falls to 68.3 percent.

A drop of 1.3 points. It is a small number and it is exactly the size that gets mistaken for a performance regression, because it is the same order of magnitude as the two-point year-on-year move SQM reports for the whole industry: large enough to look like news, small enough to be pure arithmetic. Nothing in the example got worse. The easy work left the queue.

Three consequences follow, and they are ours:

• Do not put FCR in a bonus plan during a self-service rollout. You will be paying people to keep easy contacts in the queue.
• Report FCR by request type, never blended, once a bot is live. Category-level FCR is stable under mix shift; the blended figure is not.
• Read FCR next to total contact volume. FCR down 1 point with volume down 30 percent is a good quarter. FCR down 1 point with volume flat is a real problem.

One honest limit on all of this. SQM does not attribute its declining averages to any single cause, and it is worth being precise about what it does say, because the obvious misreading runs backwards. Its comment is that the 2 percent drop was smaller than it expected given 85 percent work-from-home and 34 percent annual agent turnover, both roughly double pre-covid levels. That is a note about pressures FCR withstood, not a diagnosis of what pushed it down. Separately, average handle time reached 697 seconds in its 2024 report against 589 in its 2022 one, which SQM does not connect to the FCR trend either. We are not claiming self-service caused the industry decline, and no more is SQM claiming anything else did. We are claiming that the mix-shift mechanism exists, that it runs in a known direction, and that it is enough on its own to move a single company's number by most of what the whole industry moved in a year.

FCR against the bot metrics we already publish

The four numbers get used interchangeably. They differ in what they count and in who is being asked.

The useful way to hold them apart is that FCR is a metric about your organization and the other three are metrics about your bot. That is why FCR cannot be replaced by them and why it also cannot be run alongside them naively. Our deflection versus containment entry covers the bottom two rows in detail, and the short version is that both of them credit a customer who gave up.

We should also record what we have not asked. Across our fifteen published platform reviews, neither "first contact resolution" nor "first call resolution" appears in a single one, and "resolution rate" appears in three. We have been asking vendors about the bot-session metrics and never about the organizational one, which is a gap in our review protocol rather than a judgment about the platforms. The three platforms carried in this entry's metadata are ones we have evaluated hands-on, namely Intercom, Tidio and SendPulse, and none of those reviews answers it either. The question we are adding is narrow and answerable: when a session escalates, does the platform's reporting treat the bot turn and the agent turn as one contact or two.

Where it breaks

Quoting a number without the window. A seven-day reopen window and a thirty-day one produce different FCRs from identical service. The window is part of the metric.

Comparing a log-derived figure to a survey-derived benchmark. The benchmark's own publisher says the survey is the only method it considers accurate. Comparing across methods is comparing across instruments.

Blending FCR while running a bot. Mix shift will move the blended number more than performance does. Report by request type.

Treating every escalation as a failure. Under the strict contact convention it is one, arithmetically. Operationally a fast handoff is the correct outcome for a judgment case, and our escalation playbook is about making it fast rather than about avoiding it.

Optimizing FCR against customer effort. An agent who resolves the issue in one contact by keeping the customer on hold for eleven minutes has scored an FCR and produced a bad experience. The two metrics need to be read together or the first one will be gamed.

Assuming your platform has a convention. Many report a single blended self-service figure and leave the contact question unanswered. Ask before you build a target on it.

Related terms

• Chatbot resolution rate — the bot-session equivalent, and the entry this one is deliberately split against.
• Chatbot containment rate — non-escalation counted as success, which is what a loose FCR quietly becomes.
• Deflection vs containment — the two volume metrics and their shared blind spot.
• Customer effort score — the instrument that catches the one-contact resolution that cost the customer an hour.
• Chatbot CSAT — the satisfaction floor to read against any resolution number.
• Human handoff — the event whose contact treatment decides which way FCR moves.

FAQ

What is first contact resolution?

It is the share of customer issues settled on the first contact, with no callback or follow-up needed. It is measured across a contact center rather than inside one tool, the denominator is contacts rather than conversations with a bot, and the standard method is to ask the customer shortly afterwards rather than to infer it from whether anything came back.

What is the formula for first contact resolution?

Contacts resolved on the first attempt divided by total contacts, over a defined period. Some organizations put issues in the denominator instead, which is a further choice that changes what a transfer or a handoff does to the number. The arithmetic is trivial and the definitions are not: you also have to fix the reopen window that decides what counts as a callback, the convention that decides what counts as one contact, and the test that decides what counts as resolved.

What is a good first contact resolution rate?

SQM Group, which has benchmarked North American call centers for over twenty-five years, publishes 70 to 79 percent as good and 80 percent or higher as world-class, and reports that only 5 percent of centers reach world-class. Its 2024 benchmark put the all-industry average at 69 percent, which falls inside the band SQM itself defines as needing improvement, namely 69 percent or below. We publish no benchmark of our own, and we would treat those figures as describing inbound North American voice contact centers surveyed by telephone rather than as a target for a small business running a chatbot.

Is first contact resolution the same as first call resolution?

They are the same instrument under two names. "First call resolution" is the older term from when the contact center was a telephone operation, and "first contact resolution" is the channel-neutral version that arrived once email, chat and messaging did. Some organizations still use the two to mean voice-only versus all-channels, so it is worth confirming which sense is meant before comparing numbers.

Does a chatbot improve first contact resolution?

It depends entirely on a convention almost nobody publishes: whether a bot session counts as a contact. If bot sessions sit outside the denominator, automation is invisible to FCR. If a bot session followed by an agent counts as two contacts, every escalation is a non-FCR and the number drops sharply. If the bot turn and the agent turn are treated as one continuing contact, it barely moves. Ask your platform which one it does before you draw any conclusion.

Why did our FCR drop after we launched a chatbot?

Most likely mix shift rather than a service failure. Bots absorb the request types that resolve best, so the contacts still reaching a human are weighted toward the types that resolve worst, and the blended average falls without any category getting worse. On this page's worked example, deflecting 30 percent of contacts moves blended FCR down 1.3 points on unchanged per-category performance. Check FCR by request type and read it next to total contact volume before concluding anything went wrong.

Should a small business track FCR at all?

Probably, but as a diagnostic rather than a target, and by request type rather than blended. The version that pays for itself is the repeat-contact question: which issues bring people back, and why. That is a list you can act on. A single blended percentage compared against a contact-center benchmark is not. Our chatbot metrics guide covers assembling the wider set, and our chatbot KPIs guide covers which of them survive contact with a board.

How do I improve first contact resolution?

SQM's published root-cause work attributes non-FCR contacts to the organization 49 percent of the time, the agent 38 percent and the customer 13 percent, and lists the top repeat-contact reasons as customers chasing the status of an unresolved issue, customers disconnected while on hold, agents lacking the knowledge to resolve, requests left incomplete, and customers redirected to another company. In our reading three of those five are process failures rather than skill failures, though SQM does not classify them that way. On the automation side the equivalent lever is knowledge coverage, which is the subject of our resolution rate playbook.

Sources

• SQM Group. Call Center FCR Benchmark 2024 Results by Industry, published 6 February 2025, read 19 August 2026 — the source for the current figures on this page. Specifically: the aggregated all-industry FCR average of 69 percent and the 43 to 88 percent range; the description of that average as 2 percent lower than the previous year; the good standard of 70 to 79 percent, the world-class standard of 80 percent or higher, and the statement that only 5 percent of call centers reach it; the per-call-type rates used in this page's worked example (general inquiries 73, account maintenance 72, orders 71, billing 69, claims 61, technical 60, complaints 48); the average handle time of 697 seconds and its description as an 18 percent increase on the prior year; the work-from-home figure of 85 percent and annual agent turnover of 34 percent, both described as approximately double pre-covid levels; the source-of-error split for non-FCR contacts (organization 49 percent, agent 38 percent, customer 13 percent) and the top five repeat-call reasons; and the benchmarking methodology summarized in this page's population table, namely over 500 leading North American call centers, inbound customer service only, FCR calculated from two customer survey questions on an external post-call survey administered by SQM's in-house telephone survey agents within one business day, and a minimum sample of 400 surveys per participant. That table carries five of the ten practices SQM lists. The five we omitted are its industry-inclusion threshold of more than ten benchmarked centers, its statement that FCR survey questions are standardized across all participants, its peer-and-world-class comparison framing, its reporting of high-level repeat-call metrics, and its use of QA auditors plus speech-recognition technology to verify survey accuracy. The last of those bears on the accuracy argument this page makes and is named here rather than left out. sqmgroup.com
• SQM Group. What is a Good First Call Resolution Rate?, Call Center Industry 2022 FCR Benchmark Results, published 11 May 2022, read 19 August 2026 — the source for the earlier figures and for one statement that does not appear in the 2024 report. Specifically: the 2022 all-industry average of 71 percent, its 40 to 91 percent range, and its description as 1 percent lower than the previous year, which together with the 2024 report supports the approximate 72, 71, 71, 69 shape this page reports, which it deliberately does not date to specific years; the 2022 distribution of 5 percent world-class, 46 percent in the good band and 49 percent below 70 percent; the average handle time of 589 seconds; and the statement that the post-call survey method is the only way to accurately benchmark the FCR rate, which this page paraphrases rather than quotes, and which is carried because it bears directly on whether log-derived chatbot figures can be compared to this benchmark at all. Note on vintage: this page is four years old and its numbers are superseded by the 2024 report wherever the two overlap. It is cited here only for the items listed.
• Note on the publisher. SQM Group sells contact-center benchmarking, quality-assurance software and awards, and several of the figures above appear on pages that also market those services. We cite it because its measurement population and survey method are published in enough detail to be examined and disputed, which is not true of most FCR figures in circulation. Readers should weigh the commercial interest accordingly. We have no relationship with SQM Group of any kind.
• Worked mix-shift example. The category rates are SQM's, published in the 2024 report cited above. The 1,000-contact mix is our construction, chosen to blend to 69.6 percent near SQM's published 69 percent average so that the example reads plausibly, which is construction and not corroboration, and the deflection assumptions (60 percent of general inquiries, 40 percent of orders) are illustrative. The arithmetic is reproduced in full on the page so that a reader can substitute their own mix. This is a demonstration of a mechanism, not a measurement of any real deployment, and we make no claim that self-service caused the industry-level decline SQM reports.
• Site-wide string counts, run 19 August 2026. Across the fifteen published platform reviews (sample-reviews/{aisensy,blip,botpenguin,botpress,chatbase,chatfuel,intercom,landbot,manychat,sendpulse,tars,tidio,typebot,voiceflow,wati}-review.md), case-insensitive matches were "first contact resolution" in 0, "first call resolution" in 0 and "resolution rate" in 3. Before this publishing run the phrase "first contact resolution" matched one published entry, glossary/customer-effort-score.md, where it appears only in that entry's note explaining why the term was deferred to a later run — plus that file's build mirror under web/.content/ and our internal AUTODRAFT_LOG.md, neither of which is published content. It now also appears throughout this entry's same-day companion, academy/customer-self-service-guide.md, so the count is superseded by this run itself. Published as a transparency statement about our own coverage, not as a claim about the platforms.
• Ahrefs Keywords Explorer, US overview and volume-by-country, queried 19 August 2026 — the search-demand, difficulty, parent-topic and country-split figures in this entry's keyword note, including the check behind the deliberate decision to consolidate 'first call resolution' onto this URL rather than write a second page for it.
• Chatbotscape. Chatbot resolution rate, containment rate and deflection vs containment — re-read on 19 August 2026 to confirm the comparison table above reflects what those entries actually claim.
• Chatbotscape platform reviews — Intercom, Tidio and SendPulse are platforms we have evaluated hands-on and are the three carried in this entry's related-platform metadata. We have run no comparative test of FCR reporting or of escalation-contact conventions across them, and this entry therefore carries no ranking.
• Chatbotscape evaluation methodology. /methodology (continuously updated).