OfficePrism Benchmark for Enterprise Agent Operations
November 18, 2026
OfficePrism evaluates agent competence across 420 single-domain enterprise tasks within seven business functions. Across 124 configurations, pass rates for combined natural-language and database tasks ranged from 22.6% to 76.4%.
HOW THIS AFFECTS YOU
●
builderYou can use these benchmarks to test how well your agents handle specific business functions like database queries and ticket management.
●
researcherThis dataset provides a more granular way to localize agent failures within specific enterprise task structures.