AI ITSM deflection + customer-success benchmarks (May 2026)
Source report: /tmp/deflection-benchmarks-2026-05-11-v2.md (569 lines, 46KB).
TL;DR
The vendor field has no normalized metric. Five distinct things get called “deflection”:
- Deflection = ticket never reaches human
- Containment = ticket stays within self-service
- Auto-resolution = ticket closed by AI (may or may not have routed to human first)
- Augmentation = AI assisted but human resolved
- Response automation = first reply from AI
Even within a single vendor (notably Risotto), the definition drifts page-to-page.
Industry baseline (Fixify 2026 / Forrester / Salesforce)
- ~35-45% average at typical AI-ITSM deployment
- 55-70% mature deployment after 6-12 month ramp
- ~41% median per Salesforce industry stats
A 60-70% rate reads as above-market without triggering CIO diligence skepticism; rates above 75% start drawing scrutiny.
Cleanest single number per vendor (after normalization)
| Vendor | Claim | Caveat |
|---|---|---|
| Atomicwork | 65% auto-resolution at Ammex (6-mo ramp from 20%) | Apples-to-apples auto-resolution |
| Serval | 80% auto-resolution at Verkada in 24-hr deploy | Founder ex-Verkada — strongest selection bias |
| Console | Synthesia 25% → 70-75% | Strongest baseline-paired claim in the entire field |
| Risotto | 55% auto-resolution at Gusto (60% per TechCrunch) | Vendor’s own numbers internally inconsistent |
| Treeline | 98% augmented-or-resolved | Augmentation-weighted, NOT deflection-comparable |
| Aisera | 84% at GAP | Was “up to 80%” pre-May 2026 |
| Moveworks | 88% at Broadcom | Multi-year deployment |
Atomicwork “98%” disambiguation
Wiki correctly attributes Atomicwork’s 98% to Pepper Money ESAT (satisfaction 77→98%), NOT to ticket resolution. The “98% augmented or directly resolved” claim belongs to Treeline. The deflection-benchmarks research pass flagged this as a common conflation in external write-ups; the wiki is currently correct on this point.
Vendors with no published rates
- Echelon — pivoted to “ServiceNow build partner” positioning; no rates.
- ClearFeed — no published deflection rate.
- Thrive MSP — no published AI metrics; uses “AI services managed” framing.
- Modern (YC W26, 2 people) — has a 80%+ marketing claim with zero named customers.
Notes
- White space: paired autonomous-resolution + augmentation-effectiveness (handle-time reduction on human-resolved tickets) metrics. No competitor has built a clean “AI co-pilot effectiveness” headline as of May 2026, despite augmentation being most of the field today.
- Single-pass deflection underestimates k-run buyer experience; pass^k consistency reporting is unused across the competitor set (see academic-papers-agent-reliability-2026).
Honest verification gaps
- ASOS-side confirmation of the Edra claim — still open (see edra errata).
- HubSpot-side confirmation of the 150K conversations / 600 KB updates / 12% handoff Edra metric — still open.
- Gartner IT Key Metrics 2025 paywalled — exact tier-1 deflection benchmark not extracted.
Related
- academic-papers-agent-reliability-2026 — pass^k consistency
- pricing-benchmarks-ai-itsm-2026 — outcome-pricing wedge
- community-sentiment-snapshot-2026 — practitioner-side deflection scrutiny
- atomicwork · serval · console · risotto · treeline · aisera · moveworks
- Init Intelligence