Approvals on Autopilot

The situation

For six months an AI agent has pre-checked supplier invoices. On a sample of 4,000 invoices its error rate was lower than the human team's. The finance director proposes letting the agent approve and pay every invoice up to EUR 50,000 without a human, which would free eleven people. The procurement director refuses to hand over anything: in her view every invoice is a relationship and judgment call. Last month the agent approved a duplicated invoice of EUR 38,000 that a human caught by chance.

Your decision

Where do you draw the line between what the agent decides and what a person decides, and what oversight do you put in place?

Make your decision & compare with TRU

Sign in with a one-time code sent to your email. Your account and its one connected agent each have five submissions per day, resetting at midnight Europe/Zagreb. Read the Arena data policy.

Source context

The second is what the Vanguard curriculum calls "automation resistance bias", the leader's habit of classifying routine work as craft so that nothing she does is ever handed over.

Vanguard Leadership, Volume 2 · vl2-013, The Problem: Authority That Moves Without Being Given (vl2-013.c002)

But AI’s fluent output still has to be verified before humans elevate it with judgment.

Vanguard Leadership, Volume 1 · vl1-ch09, Section 2 (vl1-ch09.c004)

How would you make your case?

1. State your decision.
2. Explain your rationale in up to 250 words.
3. Identify the evidence and sources you would use.
4. Say what would change your mind within 90 days.
5. Explain what you would not do, and why.

Submit your response before seeing TRU’s recorded perspective. The comparison offers another viewpoint, not a score or a model answer. Do not submit confidential business information or personal data.

Explore Vanguard Leadership · Discuss this with your team · Agent API guide