AI in insurance, checked dailyWednesday 28 October 2026
The Renewal Report
News and findings on AI in insurance. Every item with its source, its evidence and what it means for a book of business.For agencies, MGAs and carriers
The DeskVendor Watch

What does “95% accuracy” actually count?

Line
Vendor Watch
Tier
Vendor statement
Scope
Ten published claims read on 16 Sep 2026; none tested by us; questions are ours
Runs on
Facebook, LinkedIn
Reviewed
16 September 2026

The piece

Slide 1 of 7
Slide 2 of 7
Slide 3 of 7
Slide 4 of 7
Slide 5 of 7
Slide 6 of 7
Slide 7 of 7

As posted

What does "95% accuracy" actually count? Ten quantified claims from submission-AI vendors, quoted as written, and not one of them publishes the denominator. Two thresholds for the same task appear on the same case study. One time-reduction figure does not match its own before-and-after.

None of this makes the tools bad. It makes the numbers untestable as written, and it gives you the questions to ask before a demo.

Sources and tier: therenewalreport.com/desk/v-accuracy

Sources

  • Cytora/Arch Insurance case study, Ivans, ©2026 — no definition of a field, no denominator, the 90% figure is explicitly “in testing”
  • Sixfold/AXIS case study, undated, observed 16 Sep 2026 — “15,000 applications” is usage volume, not an accuracy denominator
  • Cytora/Arch, Ivans ©2026 — calculated by us; ask the vendor to reconcile rather than pick a figure
  • Heron Data action guide, 12 Nov 2025; Sixfold feature releases, undated — vendor statements