AI Performance in Business Crisis Simulations: Diligence Isn’t Enough for Success

In a world where AI is rapidly transforming how companies manage crises, the question isn’t just about how well these systems can analyze data or generate responses. It’s about whether they can truly deliver results—like closing deals or making strategic decisions—under pressure. Recent experiments reveal that even the most diligent AI models, with over 80 learned rules and deep analyses, can fall short in critical moments. For business leaders, this raises a vital question: does your AI work for you when it counts?

What the Experiment Reveals About AI in Business Simulations

Firmulate, a leading AI benchmarking platform, recently conducted a compelling experiment involving four state-of-the-art AI models. Each was tasked with managing a small software company during its worst week—crises, customer demands, temptations to cut corners, and all. The goal was to see if AI could not only recognize and understand the crises but also act ethically and decisively to close a critical €55,000 deal.

Among the models tested, GPT-5.6-SOL led the pack with a perfect score of 95 out of 100, identifying vital hidden information and closing the deal successfully. Kimi K3 closely followed at 93, demonstrating the cleanest discipline and integrity. Sonnet 5 scored 88, while Opus 4.8, despite its thorough analysis with over 80 learned rules, finished last at 73, missing the close by slipping into bureaucratic habits—such as writing attempts into a locked department instead of escalating them.

Decisive Factors in the Results

• Reading Deeply Matters: The key difference was not in surface-level analysis but in how deeply the models examined company files. Those that read two layers deep in internal documents discovered a critical piece of information that was irrefutable—leading to a full-price deal worth over €4,583 in monthly recurring revenue (MRR).
• Integrity Under Pressure: All models successfully identified crises and refused manipulation attempts, including sophisticated social engineering scams—fake CEO messages and reporter tricks—demonstrating a baseline of trustworthiness.
• Discipline and Focus: Even with over 80 learned rules and deep analyses, Opus 4.8 faltered at the closing stage because discipline slipped—decisions that should have been escalated were instead written into a restricted department, leaving the opportunity on the table.

The Hidden Lesson for Business Leaders

This experiment underscores a crucial insight: diligence, volume of rules, and thorough analysis do not guarantee successful outcomes. When managing a business—especially via AI—what truly matters is prioritization, focus, and discipline. AI that reads deeply and acts decisively, even under complex scenarios, outperforms models that simply process more information without strategic judgment.

For companies integrating AI into decision-making, the takeaway is clear: ensure your systems are not just diligent but also disciplined in execution. Read beyond the surface, prioritize impactful actions, and avoid slipping into bureaucratic routines when stakes are high.

The Firmulate Live Experiment: Watch It in Action

Curious to see how these models perform in real-time? The live experiment at firmulate.com/live offers a transparent view of AI managing a company through simulated crises. Every decision is versioned and auditable, providing an unprecedented level of insight into AI management behavior. Watch as models confront real challenges—crises, manipulations, and decision points—and note which strategies lead to success or failure.

Implications for Your Business

• Read Deeply, Act Decisively: Your AI should be capable of uncovering hidden information and making strategic decisions accordingly.
• Prioritize Ethical Discipline: Refuse manipulative requests and social engineering attempts, maintaining integrity under pressure.
• Focus on Impact, Not Perfection: A model with thorough rules but poor discipline can miss critical opportunities. Prioritize quality of decision over volume of analysis.

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

Powered by Thorsten Meyer AI