Meet CARB: the Collections and Accounts Receivable Benchmark
CARB (Collections and Accounts Receivable Benchmark) is a 168-task suite that measures how reliably AI agents handle real B2B finance work like cash application, collections, and deductions, using 969 strict pass/fail criteria graded against synthetic company ledgers. Baseline results show top models passing 80 to 87 percent of tasks with tool access, while revealing which AR workflows are production-ready today and which still need human review.
