AI agents resolve 70% of ecommerce support tickets, median 48% in new benchmark

A public benchmark released by $100 M ARR AI CX leader Gorgias tested 13 vendors on 212 live, mid-market ecommerce stores, using real inventory and blind scoring of responses. The study, reported by SaaStr, found the best AI agents fully resolve about 70% of conversations, while the median vendor resolves 48% link.

Top five performers – Rep AI (75%), Decagon (73%), Yuma (71%), Ada (68%) and Gorgias (64%) – average roughly 70% automation.

All vendors and their automation rates (conversation counts in parentheses):

• Rep AI: 75% (120)
• Decagon: 73% (34)
• Yuma: 71% (118)
• Ada: 68% (420)
• Gorgias: 64% (388)
• DigitalGenius: 50% (561)
• Sierra: 48% (408)
• Siena: 46% (300)
• Kodif: 44% (188)
• Intercom: 42% (307)
• Zendesk: 37% (467)
• Envive: 22% (499)
• Klaviyo: 21% (309)

The top five average 70%, the median is 48%, and the field resolves 46% when weighted by conversation volume.

According to the benchmark’s author, Zendesk (37%), Intercom (42%) and Klaviyo (21%) each resolve no more than 42% in this benchmark, which the report describes as one of the weakest options available.

Automation vs. quality – Yuma (71% automation, 72 quality), Decagon (73% automation, 70 quality) and Gorgias (64% automation, 66 quality) score strongly on both dimensions. Rep AI (75% automation, 55 quality) and Ada (68% automation, 39 quality) achieve high automation but low quality. Sierra (48% automation, 72 quality) and Intercom (42% automation, 66 quality) show the opposite pattern.

Trend over the last four weeks – Ten vendors have enough history to show a four-week trend. On automation, nine declined: Siena -23, Kodif -13, Klaviyo -11, Sierra -11, DigitalGenius -10, Ada -9, Gorgias -6, Intercom -6, Zendesk -4. Envive improved by +8. Quality moved similarly, with nine of ten down; only Siena rose +2.

Definition of “resolved” – The benchmark counts a conversation as automated only when the AI handled it with zero human touch, no handover, and no deflection out of the channel. In-product analytics often use looser rules; for example, Gorgias’s own reporting marks a ticket resolved after 72 hours without human involvement, which can inflate its resolution rate.

Topic-level quality – Across the field, returns policy scores 77.4, shipping policy 74.4, modify/cancel 63.5, order tracking 60.1, damaged item 59.7, shopper unsure 49.3. Order tracking, a common post-sale ticket type, sits near the bottom.

Common failure modes – The most frequent failures were authentication walls, repeated requests for an order number, clarification loops, and handoffs to a human. Judges rarely observed agents fabricating facts but often saw agents that would not engage.

Speed – Mean time to a complete reply ranges from 5.4 seconds (Envive) to 16.5 seconds (Yuma). The benchmark weights support 50% automation, 40% quality and 10% speed, reflecting lower tolerance for latency in pre-sale contexts.

Planning guidance – The report suggests building a business case on 65%-70% automation for a top-five vendor and 20%-42% for bundled help-desk options. Shortlist vendors that meet both automation and quality thresholds, weight results by your own ticket mix, run a cold test on reference stores, and re-test monthly.

Our read: The benchmark confirms that only a handful of AI agents consistently deliver high-quality, fully automated support. Brands should prioritize vendors that clear both automation and quality bars, verify performance against their own ticket composition, and expect volatility as models and store configurations evolve.