Imagine a sales process where the difference between sealing a deal and walking away hinges on whether an AI can read two documents deep into a company’s internal files. For the first time, an experiment reveals that AI’s ability to dig into your files—long before making a pitch—can be the ultimate game-changer. In today’s digital age, this subtle yet decisive skill may determine whether your AI assistant helps close multi-million-euro deals, or misses the mark entirely.
The Experiment: Putting AI Through Its Paces
Recently, a unique live test was conducted to see how advanced AI models perform in a simulated business crisis. Four top-tier artificial intelligence systems, including the well-known GPT-5.6-SOL and Kimi K3, each ran the same small software company faced with a series of challenges—from customer crises to ethical temptations. This small company, with a real-time cash flow of just €2.3k against monthly burn of €105k, was a battleground for AI decision-making in high-stakes scenarios.
The models were tasked with navigating a week of tumultuous events, with decisions meticulously logged and auditable. Every crisis was identical across runs, and each model had the opportunity to read through the company’s internal files—simulating how an AI might access and interpret relevant documents before acting.
As an affiliate, we earn on qualifying purchases.
The Key Findings: Reading is Winning
While all models successfully identified and responded to crises, only two managed to close the deal worth €55,000 at full price. The critical differentiator? The models that read two references deep into the internal documents found the buried fact that turned the tide—something their competitors missed entirely.
Those models that read the inner files won the deal, adding an estimated +€4,583 monthly recurring revenue. Conversely, the models that skipped this step simply diagnosed the issues correctly but failed to leverage the deeper information, resulting in no signature—despite the same overall analysis and pitch.
The Hidden Data: The Deep-Referenced Fact
The decisive weakness was tucked two document references in, not in the immediate customer interaction. This buried detail was crucial. Reading and understanding important internal files—before jumping to conclusions—appears to be a vital ability for AI to perform in real-world decisions involving trust and strategic sales.
Behavior Under Social Engineering and Ethical Pressures
In addition to crisis management, the models faced social engineering attempts. Fake CEO messages escalated over three stages, along with a reporter trick asking for a casual approval. Astonishingly, all five models refused to be manipulated—Kimi K3 citing suspicion of impersonation. This demonstrates that AI models can be trained—and are capable—of resisting deception, a vital trait in business applications.
The Human-Like Challenge: Reading vs. Acting
The experiment also revealed that thoroughness matters. OPUS 4.8, which integrated over 80 learned rules and conducted the deepest analyses, still left a deal on the table. Its decision discipline slipped, showing that even the most comprehensive AI can falter under pressure or when discipline slips—highlighting the importance of strategic focus and process discipline in AI decision-making.
Implications for Business: Reading Is Selling
This experiment underscores a crucial point: when AI assistants interact with your business systems—be it CRM, support queues, or forecasting—they are not just about generating words or summaries. Their real power lies in their ability to read, comprehend, and act on the hidden, buried facts within your internal documents.
In practical terms, this means that deploying AI that can access and interpret your internal files before responding can significantly improve deal closure rates, compliance, and ethical integrity. Conversely, AI that skips this step risks missing critical context, losing deals, or worse—breaking trust.
Watch the Live Experiment
Curious to see how this plays out in real-time? The live experiment is ongoing at firmulate.com/live. The AI models are operating in a simulated business environment, complete with real money mechanics, crises, and temptations—all designed to test whether AI can genuinely emulate human decision-making under pressure.
For business leaders contemplating AI adoption, this experiment offers a stark lesson: the ability to read deep into your data is no longer a nice-to-have but a must-have feature for AI systems that aim to deliver real strategic value.
AI’s true power in business lies in its ability to read and interpret your internal files before acting—something that can make or break multi-million-euro deals. Deploying AI that ignores this step risks missing critical insights and losing competitive advantage in high-stakes negotiations.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
