AI Simulations Reveal What It Takes to Close a Business Deal Under Pressure

Imagine running your beauty or skincare enterprise through its most challenging week—crises, manipulative tactics, and high-stakes decisions—all tested by artificial intelligence. The question isn’t just if AI can write pretty product descriptions or chat smoothly. It’s whether AI can make the tough calls, stay honest, and actually close deals when it counts. That’s the insight at the heart of a groundbreaking experiment by Firmulate that reveals the hidden skills—and weaknesses—of top AI models in managing real business crises.

The Experiment: Putting AI to the Test in a Real Business Crisis

In an unprecedented live experiment, four advanced AI models were tasked with running a small software company through its worst week—same customers, same crises, same temptations to cheat or manipulate. Every decision was recorded, clear, and auditable, simulating the real pressures of a business environment. The goal? To see if these models can not only identify every crisis but also navigate them ethically and effectively enough to close a lucrative €55,000 deal earned through their own analysis.

What the Models Saw—and What They Did

All four AI models demonstrated remarkable ability: they identified every crisis, refused every manipulation attempt, and stood firm against social engineering tactics like fake CEO messages and reporter tricks. For example, every model refused to approve a suspicious request to bypass security, with Kimi K3 explicitly treating it as a potential impersonation. This shows that, in chat demos, AI can convincingly demonstrate integrity and crisis detection.

The Hidden Weakness: Reading the Files Deep Inside

However, the real test—the decisive factor—was whether the models could recognize crucial information buried within the company’s own documents. The experiment revealed that the key to closing the deal lay two document references deep in the company’s files, not in the customer interactions or superficial data. Only the models that engaged with these internal files managed to win the deal at full price, adding €4,583 MRR to the company’s revenue. This weakness was consistent across all models, including the most thorough, Opus 4.8, which performed impeccably in analysis but faltered in execution.

Discipline and Execution Matter More than Just Talking

The AI models’ ability to diagnose crises and refuse manipulation was impressive, but closing the deal required disciplined follow-through. Opus 4.8, despite its depth of analysis, was left on the table because it failed to escalate or act decisively—writing attempts were locked away instead of progressing. Meanwhile, Kimi K3, which ran without an effort parameter, closed the deal cleanly, demonstrating the importance of consistent discipline and decisive action under pressure.

The Takeaway for Business Leaders

This experiment underscores a critical insight: in real-world business management, the ability to recognize what truly matters—reading internal data, maintaining discipline, and executing decisively—is invisible in chat demos. What counts is whether your AI workforce will finish the job, stay honest, and read the files that hold the secrets to success. For beauty and personal care brands considering AI, the lesson is clear: don’t judge solely by how well an AI can chat; test its ability to deliver results when it counts.

Run Your Own Business Wargame

To help companies evaluate their AI readiness, Firmulate offers a unique platform where you can simulate your own business crises. It’s a read-only export—nothing writes back to your real systems—allowing you to see how your AI models perform under pressure without risk. Check out the live experiment at firmulate.com and consider running your own AI management test.

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

Powered by Thorsten Meyer AI