Workflows and controlAI RADAR · STUDY ·

01

The key finding

How reliably does AI complete payment tasks?

40%

All three attempts successful (compared with: At least one attempt failed 60%)

AI can fail again on the next attempt

In BizBench, GPT-6 Sol passed 20 of 50 tasks in all three attempts. Tasks reconstruct payment workflows. Passing requires both reaching the goal and meeting the test rules.

02

The study

How reliably does AI complete payment tasks?

All three attempts successful40%
At least one attempt failed60%

An AI system attempted 50 payment tasks - three times per task.

What the study shows

  • BizBench tests AI agents in reconstructed payment workflows.
  • GPT-6 Sol passed 20 of 50 tasks in all three attempts.
  • Passing requires both reaching the goal and meeting the test rules.
  • The test shows that one successful attempt is not enough to assess an AI system’s reliability.

How solid is this?

Method

Controlled evaluation of 50 tasks, 789 documents and 601 available tools. A simulated customer answers questions.

Limitations

  • Not yet peer-reviewed.
  • Workflows come from one payments business, not a field trial in mid-sized companies.
  • Repeated runs of the same 50 tasks are not 150 independent tasks.
03

What we take from it

Test important AI tasks repeatedly before approving deployment.

Your 10-minute check

Tick the boxes and go through them with your team. Fewer than three ticks? Then it is worth a closer look.

Our interpretation: Deployment needs evidence of repeatability and compliance, not just one successful attempt. The 40% figure is not a general enterprise AI success rate.

What does this mean for your company? Book a conversation ↗