How AI Read Your Files Before Closing the Deal — And Why It Matters for Business Trusts

Imagine an AI that doesn’t just respond to your questions but actually digs into your company’s files—reading context that’s buried two layers deep—to make smarter, more trustworthy decisions. In the fast-paced world of business, where trust and accuracy can make or break a deal, this kind of AI capability is no longer a future fantasy but a present-day reality. For leaders who depend on AI to support customer relationships, manage crises, or even close multimillion-euro deals, understanding what these systems can truly do is essential.

The Experiment That Changed the Game

Recently, a groundbreaking experiment by the firmulate team pitted four leading AI models against each other in a simulated environment mirroring a typical small software company’s worst week. Every AI faced the same set of crises, customer conflicts, and ethical temptations, with each decision fully recorded and auditable. The goal? To see which AI could best navigate real-world business challenges—not just produce convincing chat responses.

Uncovering Hidden Weaknesses

All four models demonstrated impressive capabilities—they identified every crisis and refused every manipulation attempt. Yet, only two of these models managed to close the deal worth €55,000. The key difference? The AI that succeeded had read beyond the surface, digging two document references deep into the company’s own files. This buried fact was critical—it was the decisive piece needed to finalize the sale at full price.

The Power of Deep Reading in AI

Most AI systems, especially those designed for chat or customer support, operate within a limited context—often just the conversation history or immediate prompts. However, the firmulate experiment revealed a vital insight: AI that can access and interpret information layered within internal documents has a distinct advantage. In this case, the model that read the company’s files first—and thus uncovered the buried fact—secured the deal, adding an extra €4,583 in monthly recurring revenue.

Trust and Integrity Under Pressure

Beyond mere fact-finding, the experiment tested how AI handled social engineering attempts—fake CEO messages escalating over multiple stages, or covert reporter tricks. All four models refused to be manipulated, demonstrating a robust understanding of security and ethics. For example, when faced with a staged approval-bypass request, the models correctly treated it as a potential impersonation, refusing to act without proper verification.

What This Means for Business Leaders

As AI becomes more integrated into daily business operations—from customer support to strategic decision-making—the question isn’t just about how well it writes or mimics human conversation. Instead, it’s about whether AI can finish what it starts, read your information thoroughly, and stay honest under pressure. In the experiment, the difference between success and failure boiled down to whether the AI looked beneath the surface.

The Real-World Application

Firmulate’s live platform simulates this exact scenario—hosting real companies with real money mechanics. In one ongoing demonstration, a simulated business burns €105,000 monthly against a revenue of just €2,300, navigating crises and temptations with AI-guided decision-making. The takeaway? Leaders can run their own ‘wargames’ against their business models before deploying AI systems in critical roles. These tests reveal whether AI will truly support, rather than undermine, their operations.

What the Scores Say

According to the latest standings from the Crucible League, the top AI model, gpt-5.6-sol, scored 95 out of 100, having uncovered the buried fact and signed the full deal. Kimi K3 followed closely with a 93, demonstrating the cleanest discipline, while Sonnet 5 and Fable 5 scored 88 and 77 respectively, often leaving opportunities on the table or slipping in process discipline. This league table underscores that even slight differences in reading depth and process rigor can have massive financial consequences.

Why This Matters for Your Business

Whether you’re managing customer relationships, supporting sales, or navigating complex crises, the ability of your AI tools to read and interpret your internal information deeply is a critical factor. The experiment shows that AI systems capable of uncovering hidden insights—and resisting social engineering—are more likely to close deals and maintain integrity under pressure.

Next Steps for Leaders

Leaders should consider running ‘wargames’ against their own business models, using tools like Firmulate’s platform, to identify weaknesses before deploying AI at scale. Understanding an AI’s capacity to read deeply and stay honest isn’t just a technical detail; it’s a strategic imperative for trust, security, and profitability in an increasingly automated world.

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

Powered by Thorsten Meyer AI