Reference
Support benchmarks, without the spin
“Is that good?” is the hardest question in support. Here is what real teams report across the metrics that matter — segmented by channel and industry, and every number traceable to its source.
Customer Satisfaction (CSAT)
Read CSAT against your own channel and industry, not a single global number: national cross-industry satisfaction hovers around 77 on the 100-point ACSI scale, while survey-based CSAT for individual channels runs higher.
All 8 figuresFirst Response Time (FRT)
FRT expectations are set by channel, not a single SLA: live chat and phone are measured in seconds, email in hours, so always benchmark each channel on its own curve.
All 6 figuresResolution Time (MTTR)
Full resolution runs an order of magnitude longer than first response; the blended average sits near 3-4 days, while top-quartile teams close in under a day.
All 3 figuresFirst Contact Resolution (FCR)
The cross-industry FCR average is about 70%; anything at or above 80% is world-class, and the spread across contact centers runs from roughly 50% to 90%.
All 7 figuresAverage Handle Time (AHT)
AHT is a cost input, not a quality target; the blended benchmark is ~6 minutes but is meaningless unless segmented by industry and interaction complexity.
All 7 figuresCost per Contact
Cost per contact scales with how much human time a channel consumes: self-service and AI cost cents-to-a-few-dollars, while phone is the most expensive assisted channel.
All 6 figuresTicket Volume Growth
Rigorous cross-company YoY volume benchmarks are scarce; the reliable signal is directional intent, with a majority of leaders planning for meaningfully higher volume.
All 1 figuresSelf-Service Success / Deflection
Distinguish deflection from resolution: a bot may deflect 40%+ of queries, but only a fraction end in genuine self-service resolution, so always adjust for re-contacts.
All 5 figuresAgent Attrition / Turnover
Contact-center attrition scales with team size and seniority; large centers churn far faster than small ones, and entry-level agents leave at multiples of the manager rate.
All 5 figuresQA Score / Internal Quality Score (IQS)
The commonly cited IQS benchmark is ~88%, with high performers targeting 90%+; note that manual QA still reviews only a tiny sample of conversations.
All 5 figuresCustomer Effort Score (CES)
CES lacks consolidated public industry benchmarks, so use these scale-based thresholds and track your own trend rather than chasing an external number.
All 4 figuresNet Promoter Score (NPS)
NPS varies widely by industry; above 0 is positive, 30+ is good, 50+ is excellent, so always compare against your sector rather than a universal bar.
All 7 figuresAbandonment Rate
Read abandonment against channel and service level, not one universal target: voice and live chat cluster around 7-8% while asynchronous channels sit near 3%, and an identical rate can look healthy or alarming depending on how fast the answered contacts were actually served.
All 8 figuresAverage Speed of Answer (ASA)
ASA is driven as much by how it's measured as by how good the team is: efficient cloud-contact-center panels report single-digit seconds, while live mystery-shopped consumer wait times run past three minutes — so compare only within the same channel, sector, and measurement window, and treat percent-within-seconds targets (the 80/20 rule, 911 standards) as a different animal from a raw average.
All 9 figuresLive-Chat Concurrency
Concurrency is a capacity lever, not a target to maximize: the sustainable number of simultaneous chats falls as issues get more complex and rises with agent tenure and tooling. Most teams run 2-4 per agent -- new hires at 1, seasoned agents pushing 4-6 -- and stretching higher trades response quality and FCR for raw throughput.
All 6 figuresBot/AI Containment Rate
Read containment by channel and technology tier, not as one headline number: legacy touch-tone IVR contains under ~10% of contacts while modern AI agents contain the majority. Remember that "contained" also counts abandons and dead-ends, so always pair it with CSAT and re-contact rate before you trust it.
All 9 figuresEscalation Rate
There is no single escalation-rate benchmark: it scales with product complexity and support model. The cross-industry average sits near 10-15%, but complex B2B, technical, and telecom desks run 18-28%, while low-complexity consumer sectors stay in single digits. Read your number against your own segment, and treat AI-to-human handoff as a separate rate.
All 9 figuresAgent Occupancy
Occupancy has an optimum, not a maximum: the healthy consensus for inbound voice sits around 75-85%, but read it against your call type — transactional retail and predictive-dialer outbound run higher, while healthcare and technical support run lower because of longer after-call work — and treat anything sustained above 90% as a burnout-and-attrition signal rather than a win.
All 9 figuresReopen Rate
Reopen rate only means something once you segment by complexity and pin down how "solved" is defined: cross-company averages sit around 3-5%, under 5% is excellent and over 10% is a red flag, but the tracked number understates reality when customers re-raise the same issue as a brand-new ticket.
All 7 figuresService Level (e.g. 80/20)
The famous 80/20 (80% of calls answered within 20 seconds) is an inherited default with no research behind it, not a law. Read it as one point on a spectrum: tighten toward 90/10 for VIP, critical, or emergency lines, relax toward 70/30 for low-urgency queues, and set a band (roughly 78-85%) tuned to the abandonment rate you can actually live with rather than chasing a single number.
All 7 figuresCSAT Survey Response Rate
Read a CSAT response rate against the channel it came from, not a single target. An email CSAT that lands near 15% and an in-product survey near 27% can both be healthy, because the delivery method — not how customers actually feel — drives most of the gap. Median response rates across all formats sit around 10%, so a small denominator is the norm, not a failure; what matters is whether the respondents represent your customer base.
All 9 figuresCustomer Dissatisfaction (DSAT)
DSAT is the mirror image of CSAT — the share of survey responses that land in the negative bottom box, typically a 1 or 2 on a 5-point scale — and it has no consolidated primary benchmark, so vendor thresholds openly disagree: one guide calls anything under 5% excellent while another treats scores up to 20% as fine. Read it as bottom-box share rather than a universal grade — under ~5% is strong, 5-12% is normal operational friction, and north of ~20% signals a systemic problem — and always segment by industry, since complex sectors like SaaS and telecom carry structurally higher dissatisfaction. Watch the harder primary signal underneath your own surveys, too: ACSI's measured rate of customers who complained directly to companies hit a record 17.5% in early 2026, up from 15.1% a year earlier, even as headline satisfaction stayed flat at 76.7.
All 9 figuresNet Revenue Retention (NRR)
NRR only means something against a comparable cohort. Private B2B SaaS clusters around 101% at the median, public software runs higher near 108%, and the number climbs with deal size — from roughly 97% for SMB accounts under $25K ACV to about 118% for enterprise deals over $100K. Benchmark by contract value and company stage, not against public-company headlines.
All 8 figuresTransfer / Handoff Rate
There is no universal transfer-rate benchmark: the number tracks how clean your routing is and how broadly your agents are trained, both of which swing hard by sector. Most desks aim to keep it under 10%, and observed voice benchmarks land near 8-9%, but the figure only means something read by queue and call type — an order-status line should transfer far less than insurance verification. Treat anything above 20% as a routing or IVR-intent problem rather than an agent one, and remember the rate counts hops, not the harm each transfer does when the customer has to start over.
All 8 figuresTime to Proficiency
Time to proficiency starts on the hire date and ends only when an agent sustains the production bar independently; it is not the length of the training calendar. Published references span months because product and policy complexity matter, while a single vendor case study shows how far a redesigned performance-support system can move one operation. Use these as context, define your own sustained threshold, and compare cohorts inside the same role.
All 4 figuresKnowledge Review Interval
Knowledge review cadence is a risk control, not an outcome benchmark. Product documentation demonstrates intervals from weeks to a year and a three-month example, but it does not establish that any one interval produces accurate content. Assign faster review and event-driven triggers to high-impact answers, then monitor overdue reviews and customer or agent failure signals.
All 4 figuresAI Agent Task Success & Reliability
These figures are evidence about evaluation, not a universal production target. The tau-bench result shows that realistic multi-turn, tool-using service tasks remain materially harder than answer-generation demos, especially when a task must succeed repeatedly. Nubank's deployment result shows the upside of evaluation-driven iteration in one use case, but its percentage-point lift is a before-and-after result, not a cross-company benchmark.
All 4 figuresHelpdesk Public Seat Price
A dated list-price snapshot is useful for budgeting, not for declaring a cheapest platform. Zendesk's figures below require annual payment; Intercom's help documentation lists monthly full-seat costs. Included channels, AI, security, collaborator seats, usage, and add-ons differ across tiers, so normalize the actual requirements and billing term before calculating total cost of ownership.
All 6 figuresSupport AI Public List Price
Support AI is sold in incompatible units. Outcome and automated-interaction prices apply to autonomous work; per-agent prices apply to copilots. Even identically named units can count handoffs, workflows, or customer silence differently. Treat this as a July 2026 market snapshot, map each definition to the same workload, and compare cost per independently verified correct resolution.
All 5 figuresReading the evidence
A benchmark is context, not a quota.
Two figures can share a metric name and still measure different work. Before you compare your operation with any number here, reconcile the population, clock, channel, segment, and decision the figure was built to support.
Read the editorial standards- 01
Match the denominator
Check what counted, what did not, when the clock started, and whether the measure uses business or elapsed time.
- 02
Compare the segment
Channel, customer tier, issue complexity, geography, and service promise usually explain more than the headline mean.
- 03
Trace the source
Every figure keeps its named source, year, segment, and reviewed link so you can inspect the evidence page behind the number and see when it cites an upstream study.
- 04
Pair a guardrail
Read speed beside resolution and quality, cost beside repeat demand, and automation beside confirmed customer outcomes.
The Support Brief
New benchmarks and the research behind them in one sourced monthly issue.
Monthly. No ads. Unsubscribe anytime. Privacy