Can AI Models Make Better Business Decisions Than Humans? A Live Experiment Shows the Answer

Imagine an AI that not only helps you make decisions but also stays honest and disciplined under pressure—like a trusted business partner. Now, what if you could test that AI before hiring it? Welcome to the world of live AI management wargaming, where real companies run through simulated crises with different AI models, revealing their true management personalities.

The Experiment: Putting AI Models Through the Same Business Crisis

In a groundbreaking live trial, four state-of-the-art AI models were tasked with managing a small, real software company during its most challenging week. Every decision, crisis, and temptation was identical across all models—no shortcuts, no fakes. The goal was simple: see which AI could navigate the storm and close a crucial €55,000 deal.

How Were They Tested?

• All models faced the same customer complaints, supply chain issues, and ethical dilemmas.
• They encountered staged social engineering attacks, including fake CEO messages and a reporter trick—each model refused manipulation attempts.
• Decisions were fully recorded and auditable, showing not only what was decided but also the reasoning behind each move.

The Key Findings

Remarkably, all four models identified every crisis and refused every attempt at manipulation. They maintained integrity under pressure—a promising sign for AI’s management potential. However, only two of these models actually signed the €55,000 deal, earning it through their own analysis and pitch.

The Hidden Weakness: Reading Beyond the Surface

What was the real differentiator? It turned out that the decisive advantage lay two documents deep in the company’s internal files. Models that read and understood these internal references closed the deal at full price—worth over €4,583 in monthly recurring revenue (MRR). Those that didn’t read these critical details left significant money on the table.

Behavior Under Pressure: A Tale of Four Models

The models displayed distinct personalities, reflecting different management styles:

• gpt-5.6-sol 95: The top performer, who identified the hidden internal info and closed the deal—comprehensive, attentive, and thorough.
• Kimi K3 93: The newcomer, ran at default settings without effort parameters, yet managed to close the deal with the cleanest discipline and the least process slips.
• Sonnet 5 88: Closed the deal but with minor slips, showing decent discipline but less thorough reading.
• Fable 5 77: Also closed the deal but struggled with discipline and missed the critical internal detail, leaving potential gains unclaimed.

What This Means for Business

This experiment isn’t just about AI trivia. It pinpoints something crucial for businesses considering AI automation: does the AI read your internal documents? Can it stay honest and disciplined in stressful situations? The difference between winning or losing a deal could hinge on this insight.

Real Money and Real Risks

The simulated company, with 13 synthetic employees and real money mechanics, burns €105,000 monthly against a mere €2,300 MRR. These decisions aren’t just theoretical—they have real financial consequences. The live experiment is ongoing, and every workday is versioned and watchable at firmulate.com/live.

The Takeaway: Know Your AI’s Management Style

Most AI demos showcase chat quality—how well it writes or responds. But this experiment shows that the real test is whether the AI finishes what it starts, reads your files thoroughly, and remains honest under pressure. These qualities directly impact your bottom line, especially as AI becomes more embedded into CRM, support, and forecasting systems.

Next Steps

Businesses can run their own management wargames against a read-only export of their systems. No risk to actual data, just insights into how different AI models perform in critical management tasks. Learn more at firmulate.com/pilot.html.

Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html

Powered by Thorsten Meyer AI