Key Takeaways
- 92% of contact centers have a formal QA program, but manual review covers only 2-5% of conversations, leaving the vast majority of interactions unmonitored
- The industry benchmark for Internal Quality Score (IQS) is 88%, yet only about one-third of support teams actually track IQS formally
- AI-assisted QA raises conversation coverage from 2-5% to 100% and is associated with 12-18% CSAT improvements and 40-50% reductions in compliance incidents
- Support teams that link QA scoring to structured coaching see 28% faster agent ramp-up times compared to teams without formal coaching loops
- QA programs with consistent execution correlate with first-contact resolution rates above 70%, versus the 50-60% FCR rates common in teams without structured review
Customer support quality assurance statistics 2026: what the data shows
Quality assurance in customer support has a structural problem that most adoption metrics obscure. Nearly every contact center has a QA program on paper. The gap is in execution: what fraction of conversations actually get reviewed, how consistently scoring gets applied, and whether review findings connect to coaching and process change.
AI-assisted QA tools have made 100% conversation coverage technically feasible for the first time. Meanwhile, most teams still rely on manual sampling that touches fewer than 1 in 20 interactions. The data below draws from Klaus (now Zendesk QA), MaestroQA, Zendesk, SQM Group, Freshworks, and AmplifAI to show where QA programs actually stand and what the performance gap between high- and low-coverage operations looks like.
QA program adoption rates
Adoption is near-universal. Execution is not.
| Metric | Figure | Source |
|---|---|---|
| Contact centers with a formal QA program | 92% | AmplifAI Customer Service Statistics 2026 |
| Support teams that formally track Internal Quality Score (IQS) | ~33% | Klaus/Zendesk QA Benchmark Report 2023 |
| Contact centers that struggle to find time for QA | 85% | AmplifAI Customer Service Statistics 2026 |
| Support professionals who find measuring quality challenging | 30% | Klaus Customer Service Quality Benchmark Report 2023 |
| Teams measuring across all three critical error types | 61% | AmplifAI Customer Service Statistics 2026 |
| Organizations that say QA improves service quality | 86% | Industry QA surveys, 2025 aggregate |
| Organizations that say QA boosts customer satisfaction | 76% | Industry QA surveys, 2025 aggregate |
The gap between the 92% that have a program and the 33% that track IQS formally tells most of the story. Having a QA program often means having a scorecard template. It does not necessarily mean consistent review cycles, calibrated scoring, or any formal feedback loop connecting review outputs to agent development.
Conversation coverage: the core gap in manual QA
How little of the actual conversation volume gets reviewed under manual processes is where most QA programs fall apart.
| Coverage Metric | Figure | Source |
|---|---|---|
| Share of interactions reviewed under typical manual QA | 2-5% | AmplifAI / Solidroad Call Center QA Data 2025 |
| Share of interactions reviewed under legacy QA systems | 1-2% | Intryc Customer Support QA Guide 2026 |
| Estimated share of conversations never reviewed | 81% | Industry analysis, multiple sources 2025 |
| Conversations reviewed at a mid-size insurance contact center (example) | 0.3% (40 of 12,000/week) | Lorikeet CX AI QA Tools Analysis 2025 |
| Manual scoring hours per week for a 50-agent team | 20-25 hours | Automated QA research, 2025 |
| Lag time for results under manual review | 3-5 days after interaction | Automated QA research, 2025 |
The 2-5% figure comes up consistently across sources and reflects a real constraint: manual QA requires a dedicated analyst, and a full review of a customer interaction takes time. At typical staffing ratios, even a well-resourced QA team cannot get above single-digit coverage of inbound volume.
The consequence is not just incomplete data. The 95%+ of conversations that go unreviewed are where compliance risks go undetected, coaching opportunities are missed, and the patterns driving customer dissatisfaction stay invisible. QA programs built on 2-5% sampling give a signal, not a picture.
Internal Quality Score (IQS) benchmarks
IQS is the most commonly used metric for tracking QA performance over time. It aggregates scorecard ratings across reviewed interactions and expresses team performance as a percentage against a defined standard.
| IQS Metric | Figure | Source |
|---|---|---|
| Industry IQS benchmark | 88% | Klaus Customer Service Quality Benchmark Report 2023 |
| Recommended IQS target for high-performing teams | 90%+ | Zendesk QA guidance, 2025 |
| Typical IQS at program start (MaestroQA case data) | ~70% | MaestroQA customer case studies |
| IQS after 6-8 months of structured QA (MaestroQA case data) | ~90% | MaestroQA customer case studies |
| IQS typical score range for compliance-heavy industries | 90%+ expected | Enthu.AI QA Scorecard Guide 2026 |
| IQS typical score range for teams weighting soft skills more | 80%+ expected | Enthu.AI QA Scorecard Guide 2026 |
The 88% IQS benchmark comes from Klaus's 2023 survey of over 4,000 customer service professionals across 98 countries, conducted with Aircall and Support Driven. It reflects a weighted average across support operations primarily in software, e-commerce, and B2B.
The MaestroQA trajectory from 70% to 90% over 6-8 months is a consistent pattern in structured QA rollouts: initial scores are lower partly because teams are calibrating against a new standard, and partly because the process surfaces issues that were previously invisible. Score improvement reflects both genuine performance gains and measurement refinement.
Grading methods and scorecard structure
There is no universal QA scorecard, but consistent structural patterns emerge across tools and industries.
| Scorecard Element | Common Practice | Source |
|---|---|---|
| Rating scale most commonly used | 1-5 or 1-10 per criterion | Zendesk QA Scorecard Guide 2026 |
| Section types used in scorecards | Standard, Bonus, Auto-fail | MaestroQA Help Center documentation |
| Common scorecard categories | Brand/tone, protocols, efficiency | MaestroQA QA scorecard research |
| Teams using auto-fail criteria for critical violations | Majority of formal QA programs | Industry QA documentation |
| Frequency of scorecard review and update | At least quarterly | Zendesk QA best practices guidance |
| QA coaching sessions tied to four or more reviewed calls | 71% of call centers | SQM Group Call Center QA research |
Auto-fail sections are a significant design choice. They allow specific criteria (misrepresenting policy, missing required disclosures, handling a safety issue incorrectly) to fail an entire interaction regardless of the score on other criteria. Their presence in most formal programs reflects that not all quality dimensions are equally weighted.
The 71% of call centers that coach agents on four or more reviewed calls per cycle indicates that most serious QA programs have moved beyond review-for-compliance toward review-for-development. Whether those coaching conversations are structured and calibrated varies considerably.
AI-assisted QA vs. manual review
Automated QA using large language models to evaluate conversations has shifted from pilot to production for a meaningful portion of support teams by 2026.
| Metric | Manual QA | AI-Assisted QA | Source |
|---|---|---|---|
| Conversation coverage | 2-5% | 100% | Multiple sources, 2025 |
| Time to results after interaction | 3-5 days | Near-real-time | Automated QA research 2025 |
| QA analyst hours per week (50-agent team) | 20-25 hours | Significantly reduced | Automated QA research 2025 |
| Bias and scoring consistency | Subject to evaluator fatigue | Consistent across all interactions | NICE AI QA research |
| CSAT improvement reported after AI QA adoption | Baseline | 12-18% improvement | Automated QA scoring research 2025 |
| QA cost reduction after AI QA adoption | Baseline | 25-30% reduction | Automated QA scoring research 2025 |
| Compliance incidents within first year of AI QA | Baseline | 40-50% decrease | Automated QA scoring research 2025 |
The jump from 2-5% to 100% coverage is structural, not incremental. It changes what QA data can tell you: instead of a sample that may or may not represent your tail risk, you have a complete record. That changes both the coaching use case and the compliance use case materially.
The 25-30% QA cost reduction reflects primarily the reduction in analyst time spent on manual review. That labor shifts (in well-run programs) toward calibration, coaching design, and escalation handling rather than basic transcription review.
The 12-18% CSAT improvement figure comes from implementations where AI QA was paired with structured coaching loops, not from AI QA alone. Coverage without action on what the coverage reveals does not move CSAT.
Impact on CSAT and first-contact resolution
QA programs are justified organizationally on the basis that better agent performance drives better customer outcomes. The data on that relationship is directional, not deterministic.
| QA-Outcome Metric | Figure | Source |
|---|---|---|
| Organizations that believe QA reviews can improve CSAT | 75% | Klaus Customer Service Quality Benchmark Report 2023 |
| CSAT improvement (Blueground, after AI QA implementation) | 77% to 82% YoY | Zendesk/Klaus case data |
| Agent-driven dissatisfaction reduction (Welcome Pickups) | 50% to 39% within two months | Zendesk/Klaus case data |
| QA coverage increase (Blueground) | 3% to 5.5% with AI QA | Zendesk/Klaus case data |
| Weekly QA time saved (Blueground, 70 agents) | 40+ hours per week | Zendesk/Klaus case data |
| First-contact resolution average across industries (2025) | 70% | SQM Group FCR Benchmark 2025 |
| FCR range across call centers | 50%-90% | SQM Group FCR Benchmark 2024 |
| Tracking of FCR among service pros (2024) | 80% | Salesforce State of Service 2024 |
| FCR improvement from regular agent training linked to QA | Up to 25% | Industry QA studies 2025 |
| Agent ramp-up time improvement with QA-linked coaching | 28% faster | SQM Group / call center QA research |
The Blueground case (70 agents, approximately 19,000 tickets per month) is one of the more cited examples in the QA tooling space because it shows a concrete before-and-after: 40+ hours per week saved on QA administration, coverage nearly doubling from 3% to 5.5%, and a 5-point CSAT gain year-over-year. The coverage improvement is modest in absolute terms, which illustrates that even purpose-built AI tools require calibration time and human review of flagged interactions to close the gap to full coverage.
The 70% FCR average from SQM Group's 2025 research represents the aggregated cross-industry benchmark. Teams running structured QA with regular coaching loops consistently report FCR in the 80-90% range. The 25% FCR improvement from training linked to QA reviews is not from a single study but represents a consistent directional finding across call center research over several years.
QA tool landscape: Klaus (Zendesk QA), MaestroQA, and others
The QA tooling market has consolidated since 2023. The dominant platforms differ mainly in how much they lean on AI automation vs. structured manual workflows.
Zendesk QA (formerly Klaus)
Klaus was acquired by Zendesk and rebranded as Zendesk QA. The platform provides AI-powered conversation review, IQS tracking, sentiment filtering, and integration with Zendesk support tickets. It generates the IQS metric used as the industry benchmark (88%) in the Klaus benchmark reports. The platform filters for conversations with positive or negative sentiment, identifies interactions most in need of review, and supports both manual scoring and AutoQA workflows.
MaestroQA
MaestroQA is built around customizable scorecards with standard, bonus, and auto-fail section types. The platform introduced AutoQA to extend coverage to 100% of conversations alongside targeted manual review. It is used by support operations at companies including Monday.com and ClassPass. Published case data shows QA scores moving from approximately 70% to 90% over 6-8 months of structured use. The platform integrates with Zendesk and includes coaching workflow features that surface coachable moments from conversation data.
Intercom QA
Intercom's native QA features are built into its support platform, with review workflows tied to its conversation data. Organizations using Intercom as their primary support tool often start with its built-in QA capabilities and layer on specialist tools as their programs mature.
AI-native QA platforms
Solidroad, Intryc, and Crescendo focus on automated QA coverage rather than manual workflow management. These tools score all conversations automatically, surface coaching nudges in near real-time, and run compliance monitoring without requiring a dedicated QA analyst to queue interactions.
Where QA programs break down
Three failure modes show up across the adoption and outcome data consistently.
The sampling illusion. A team reviewing 3% of interactions can produce QA scores, calibration sessions, and coaching plans based on a sample that systematically underrepresents certain agents, interaction types, or channels. The scored interactions may show 87% IQS. The unscored 97% may contain most of the compliance risk and CSAT damage.
The review-without-action gap. 81% of conversations never reviewed is the obvious coverage problem, but reviews that are completed without connecting to any structured feedback loop cause a quieter version of the same issue. AmplifAI's research showing 79% of agents find QA feedback helpful alongside 85% of contact centers struggling to find time for QA suggests the bottleneck is often in the calibration and coaching work downstream of review, not the review itself.
Scorecard drift. Scorecards built for one product configuration, compliance requirement, or interaction type become misaligned as operations change. Zendesk QA recommends reviewing scorecards at least quarterly. Teams that skip this end up scoring interactions against criteria that no longer reflect their actual standard.
QA program benchmarks by team size
| Team Size | Recommended Coverage Target | Practical Coverage (Manual) | AI QA Coverage Potential |
|---|---|---|---|
| 1-10 agents | 15-25% of interactions | 10-20% achievable | 100% |
| 11-50 agents | 5-15% of interactions | 3-7% typical | 100% |
| 51-200 agents | 3-5% of interactions | 2-4% typical | 100% |
| 200+ agents | 1-3% of interactions | 1-2% typical | 100% |
These ranges reflect general industry practice rather than prescribed standards. The recommended targets for smaller teams are achievable with manual review because conversation volume is lower relative to available reviewer time. At larger team sizes, the gap between recommended and actual coverage widens considerably under manual-only approaches.
Connecting QA to business outcomes
Review consistency and downstream metrics are closely linked across the data sources below.
CSAT Score Benchmarks by Industry 2026 shows that the top-performing support operations by CSAT score consistently run structured QA programs with calibrated scorecards and coaching loops. The cross-industry CSAT average is 78/100. Teams with mature QA infrastructure tend to cluster at 82-88.
Customer Support Cost Per Ticket Benchmarks 2026 includes data on how repeat-contact rates and escalation rates drive per-ticket cost. QA programs focused on first-contact resolution directly reduce these costs by identifying the interaction patterns and agent behaviors that generate avoidable repeat contacts.
Customer Support Automation Statistics 2026 covers the deflection rates and cost structures of automated support. AI-assisted QA and automated support intersect meaningfully: automated interactions need quality review too, and the AI tools that handle QA often share infrastructure with those that handle automated response.
Customer Support Agent Turnover Statistics 2026 covers the structural cost of high agent churn. QA-linked coaching is one of the more consistently cited factors in agent retention research: agents who receive structured feedback report higher job satisfaction and stay longer. The 28% faster ramp-up from QA-linked coaching also reduces the cost of new-hire onboarding.
Conclusion
92% QA adoption coexisting with 81% of conversations never reviewed and only 33% of teams tracking IQS is not a paradox. It is what happens when programs are treated as checkboxes rather than operational infrastructure.
The practical divide is between QA programs that stop at having a scorecard and those that close the loop: coverage reviewed, findings coached, scorecards updated when the product or compliance requirements change. FCR and CSAT separate along those lines more reliably than along almost any other operational variable.
AI-assisted QA has made 100% coverage achievable for teams that invest in it. The case data from Klaus/Zendesk and MaestroQA shows consistent patterns: 40+ hours per week recovered from manual review administration, QA scores moving from 70% to 90% over 6-8 months, and CSAT gains of 5+ points with sustained execution. The limiting factor is no longer whether you can review all your conversations. It is whether your team acts on what the review surfaces.
In-house manual QA works at under 50 agents. Above that, the math on analyst time versus inbound volume makes 100% coverage impossible without tooling. Zendesk QA, MaestroQA, or one of the AI-native platforms becomes the practical path to meaningful coverage.
If you are building out a support team and need help structuring QA programs alongside hiring, Stealth Agents provides dedicated customer support staffing with QA-ready onboarding built into the engagement model. Book a consultation or view pricing to see what that looks like for your team size.
Methodology and sources
Statistics in this article were drawn from the following primary sources. Where figures varied across sources, the range is noted or the most methodologically rigorous source is cited.
- Klaus Customer Service Quality Benchmark Report 2023 (survey of 4,000+ CS professionals, 98 countries, conducted with Aircall and Support Driven)
- Zendesk QA product documentation and IQS benchmark guidance, 2025-2026
- MaestroQA customer case studies and scorecard documentation, 2024-2025
- AmplifAI Customer Service Statistics 2026 (135+ statistics compilation)
- SQM Group FCR Benchmark Report 2024 and 2025
- Salesforce State of Service 2024
- Lorikeet CX AI QA Tools Analysis 2025
- Solidroad Call Center Quality Assurance Software data, 2025
- Intryc Customer Support QA Guide 2026
- Crescendo AI Automated Quality Assurance analysis, 2026
- NICE AI-Driven QA in Customer Service research
- Enthu.AI QA Scorecard Guide 2026
- Freshworks Customer Service Benchmark Report 2025
- Zendesk Customer Experience Trends Report 2025
- Blueground and Welcome Pickups case data (via Zendesk QA/Klaus)
- The Level AI Customer Support QA Tools analysis, 2025