Research library
AI support and human handoffs research
Source data and original analysis for ecommerce teams. Start with the featured study, inspect the definitions and calculations, and follow each finding to its source.
13research pages
290source citations
26research exhibits
November 2026library updated
Featured study
Retail AI Support Benchmark Statistics for Order-Changing Tasks
See which retail AI benchmark cases require record changes, identity lookups, multiple orders, repeated writes or transfers, and how to test those outcomes.
154 citations · 6 charts · Updated November 2026 Read the study- 104 of 114 English retail benchmark cases expect a write.
- 44 of those benchmark cases expect multiple writes.
- 58.1% of return-expected benchmark cases have repeated order lookups.
- 61.0% of pending-edit benchmark cases expect multiple writes.
Research library
Explore the studies and their source data.
- Complaint handoff statistics for forms shoppers have to repeatCarrying the complaint forward raised form return by 26 percentage points, while the study did not establish better resolution satisfaction.9 citations · 1 chartsThe simplified process returned 48% of complaint forms, versus 22% in the old process.
- Service apology statistics and the difference between withdrawal and resolutionThe randomized messages changed feedback withdrawal; the experiment did not measure whether the underlying problem was resolved.10 citations · 1 charts44.8% withdrew feedback after the apology, versus 21.1% across the compensation arms.
- Pre-sale AI support statistics across four service comparisonsFour randomized retail comparisons show how AI support results change when the alternative is no live help, a human, or a human escalation path.37 citations · 1 chartsThe no-live-help comparison produced a 16.25% sales estimate.
- Chatbot response time statistics for new and experienced usersAn interaction experiment found that a response delay changed novice and experienced users’ perceptions in opposite directions.32 citations · 1 chartsRaw group means moved in opposite directions.
- AI Chatbot Disclosure and Customer Response StatisticsRead the response, immediate-exit and final-decision outcomes side by side in a six-condition disclosure experiment.4 citations · 2 charts
- Human-Like Chatbot Conversation StatisticsSee what a retail buy-back experiment found when it varied social cues, response pacing and humor independently.4 citations · 2 charts
- AI chat handoff statistics for technical problems and shopper frustrationTaobao’s field study separates capability escalations from customer frustration and measures what human agents actually did after taking over.4 citations · 2 charts
- AI customer support escalation statisticsA 5,172-agent field study measured requests for a manager. Use its findings to measure a shopper’s request for help and the actual handoff separately.10 citations · 2 charts
- AI customer support productivity statisticsRead the human-assist field study by resolutions per hour, chat duration, agent tenure and response quality before changing checkout support staffing.9 citations · 2 charts
- AI customer support quality and repeat contact statisticsA randomized Taobao experiment tracked how AI-assisted agents handled issues, how customers rated chats, and whether they came back for the same problem.7 citations · 2 charts
- Peak-season ecommerce support queue statistics by shopper questionEstimate what the holiday queue needs from your team by separating question mix, attributed purchase rates and after-hours coverage.6 citations · 2 charts
- Ecommerce support channel statistics for deciding where to staff chatMatched-cohort data from 10,990 Gorgias brands separates growth in chat tickets from email workload and shows how to evaluate staffing by channel.4 citations · 2 charts
Method
Follow the evidence and its limits
- 01
Identify the population
Every study states who or what was measured, and when.
- 02
Inspect the calculation
Downloads preserve source values, denominators and derived results.
- 03
Keep the scope
Historical data and controlled comparisons retain the conditions that determine where the findings apply.
Study methods distinguish original calculations, reported experimental results and practical interpretation.