The short version
2026 editionKey takeaways
Each takeaway links to its canonical data row. Reported figures belong to their cited producer; modeled results are labeled as calculations.
- Support study: increase in issues resolved per hour: 15.2% (Rollout mainly fall 2020–winter 2021; article published February 2025).Brynjolfsson, Li and Raymond
- Support study: reduction in average chat handling time: 8.5% (Rollout mainly fall 2020–winter 2021; article published February 2025).Brynjolfsson, Li and Raymond
- Support study: increase in resolved-chat share, approximately (not statistically significant): 1.3 pp (Rollout mainly fall 2020–winter 2021; article published February 2025).Brynjolfsson, Li and Raymond
- Support study: decrease in customer nps, approximately (not statistically significant): 0.12 (Rollout mainly fall 2020–winter 2021; article published February 2025).Brynjolfsson, Li and Raymond
- Support study: increase in issues resolved per hour for less-skilled and less-experienced agents: 30% (Rollout mainly fall 2020–winter 2021; article published February 2025).Brynjolfsson, Li and Raymond
- Zendesk consumer survey: want the representative to continue from the previous interaction: 81% (2026 report released November 2025; fieldwork dates undisclosed in the release).Zendesk
- Zendesk consumer survey: are frustrated by repeating information: 74% (2026 report released November 2025; fieldwork dates undisclosed in the release).Zendesk
- Zendesk consumer survey: say responsiveness and accuracy strongly affect purchases: 86% (2026 report released November 2025; fieldwork dates undisclosed in the release).Zendesk
- Zendesk consumer survey: would choose text, voice and visuals in one conversation: 76% (2026 report released November 2025; fieldwork dates undisclosed in the release).Zendesk
- Zendesk consumer survey: expect explanations for ai-made decisions: 95% (2026 report released November 2025; fieldwork dates undisclosed in the release).Zendesk
What changed when support agents received AI assistance?
The study found higher resolution productivity and shorter handling time. The average quality measures did not show a statistically significant improvement, so higher productivity must not be relabeled higher customer satisfaction.
Scroll horizontally to compare all columns →
| Measured outcome | Reported change | Interpretation |
|---|---|---|
| Increase in issues resolved per hour | 15.2%Quasi-experimental outcome | Statistically significant |
| Reduction in average chat handling time | 8.5%Quasi-experimental outcome | Statistically significant |
| Increase in resolved-chat share, approximately | 1.3 ppQuasi-experimental outcome | Not statistically significant |
| Decrease in customer NPS, approximately | 0.12Quasi-experimental outcome | Not statistically significant |
| Increase in issues resolved per hour for less-skilled and less-experienced agents | 30%Quasi-experimental outcome | Subgroup result; not the workforce average |
Uses the published article, not its earlier working paper. Relative percentage changes, percentage-point changes and NPS points are different units. A nonsignificant estimate is not proof of zero effect.
Primary source: Brynjolfsson, Li and Raymond — February 4, 2025
What do consumers expect from AI-enabled service?
Consumers report preferences for continuity, responsiveness, channel flexibility and explanations. These are survey expectations, not observed retention rates or experimentally measured support outcomes.
| Consumer response | Respondent share |
|---|---|
| Want the representative to continue from the previous interaction | 81%Survey expectation |
| Are frustrated by repeating information | 74%Survey expectation |
| Expect support tailored to prior interactions | 67%Survey expectation |
| Say responsiveness and accuracy strongly affect purchases | 86%Survey expectation |
| Would choose text, voice and visuals in one conversation | 76%Survey expectation |
| Expect explanations for AI-made decisions | 95%Survey expectation |
The combined participant count must not be used as the denominator for each consumer question.
Primary source: Zendesk — November 18, 2025
Does the study show that autonomous bots can replace agents?
No. It evaluates suggestions delivered to human support agents, who could edit or ignore them. Autonomous resolution, escalation policies and end-to-end chatbot performance require their own evaluation.
Do not relabel assisted-agent productivity as a bot containment rate, headcount reduction or guaranteed cost saving.
Primary source: Brynjolfsson, Li and Raymond — February 4, 2025
Which metrics should an AI support scorecard keep separate?
Track completed resolutions, handling time, repeat contacts, escalation and customer-rated quality independently. Define the denominator for each measure and include the work performed after a bot hands the conversation to a person.
| Metric | Definition for a local evaluation |
|---|---|
| Resolution productivity | Accepted resolutions divided by staffed hours |
| Handling time | Active handling time per conversation, with concurrency rules stated |
| Escalation | Conversations transferred to a human or supervisor divided by eligible conversations |
| Quality | Independent review or customer feedback using the same scale across conditions |
Workspace369 editorial measurement framework; these definitions are not extra empirical findings.
How this report was built
Methodology and limitations
- The measured-results table retains the published paper’s preferred estimates and its null-quality findings. The survey table contains only separately attributed consumer expectations.
- The public author-hosted article was checked directly, including the main-results tables. Only selected factual results are summarized; the paper and its charts are not redistributed.
- The original contribution is the evidence-type comparison and reusable measurement framework, not a new support experiment. No cross-source average is calculated.
What these numbers cannot tell you
- The operational study concerns one company, a historical assistant and a nonrandom full rollout. The identification strategy is stronger than a simple opinion poll but does not guarantee transfer to another team.
- The consumer survey is vendor-sponsored. Stated preferences are not observed purchasing decisions, and its report-edition year is not its fieldwork date.
- No result establishes how Workspace369 performs, proves an autonomous-agent replacement rate, or supports a blanket customer-service ROI claim.
Freshness and corrections
Maintained by the Workspace369 editorial team. Review quarterly and when a cited producer releases a replacement study. Next editorial review: December 2026. Retain historical model versions and observation dates. The edition date changes only when the evidence or content is substantively reviewed; it does not change the underlying observation period.
First edition: . Data extraction, source attribution and arithmetic checked for this edition. No independent peer review is claimed.
Found an error or a newer primary release? Send a correction with the source and affected statistic. Confirmed corrections should be recorded in the revision history before republishing.
Primary sources and provenance
Every reported numeric cell links directly to its producer. The downloads include exact table or workbook locators, observation periods, access dates, formulas and input references.
- Brynjolfsson, Li and RaymondGenerative AI at Work, published Quarterly Journal of Economics article (2025) ↗Published February 4, 2025. Accessed September 12, 2026.Selected factual observations paraphrased with attribution. No source report, chart, participant data or proprietary database is redistributed; source rights remain with its producer.
- Zendesk2026 CX Trends: original publisher research announcement ↗Published November 18, 2025. Accessed September 12, 2026.Selected factual observations paraphrased with attribution. No source report, chart, participant data or proprietary database is redistributed; source rights remain with its producer.
Made to be checked, then cited
How to cite this report
For a source-reported statistic, credit the original publisher and link to the exact row here when using our compilation. For a modeled result, cite Workspace369 and include the assumptions. Linking to this page does not make us the original producer of third-party data.
Workspace369. (2026-09-12). AI Customer Service Performance Statistics. https://workspace369.com/research/ai-customer-service-performance-statistics/. Primary sources and observation periods as listed in the report.
The downloads are English-language reference datasets, including on translated pages. Source rights remain with their producers. Attribute Workspace369’s compilation and calculations; consult each source’s reuse terms.