Your competitors don't share their playbook. Neither does the real world. AI just learned to win anyway.
For decades, the benchmark for AI greatness was games where every piece was visible. Chess. Go. Checkers. Give the machine a complete picture of the board and enough compute, and it eventually crushes every human alive. Impressive, sure. But those games were never a fair analogy for running a business. In business, you never see the whole board.
That just changed. And the implications for small and mid-sized operators are bigger than most tech headlines are letting on.
The Game That Broke AI's Comfort Zone
Stratego is a two-player board game where half the pieces are always face-down. You never know exactly what your opponent is holding. You make moves based on inference, pattern recognition, and calculated risk—not perfect information. Sound familiar? It should. That's Tuesday in most operations.
Researchers recently built an AI system that beat the all-time best human Stratego player in recorded history—and did it on a limited compute budget. This wasn't a moonshot project backed by billions in GPU clusters. It was a targeted, efficient system designed to reason under uncertainty.
The technique behind it involves what researchers call imperfect-information game solving: the AI doesn't wait for a complete picture before acting. It builds probabilistic models of what it can't see, updates those models as new signals arrive, and commits to the highest-expected-value move given everything it knows right now. It plays the hand it has, not the hand it wishes it had.
That's not a party trick. That's a decision architecture your business can use today.
Why "We Don't Have Clean Data" Is No Longer an Excuse
The old argument against deploying AI in real operations went something like this: "Our data is messy. We don't have complete records. AI needs perfect inputs to be useful." That argument just expired.
Think about the incomplete-information problems you face every single week:
- You don't know if a supplier is going to deliver on time—or at all—until it's almost too late to react.
- You don't know which customer in your pipeline is three clicks away from churning versus which one is about to place a repeat order.
- You don't know which inbound lead is worth a callback today and which one will ghost you after 40 minutes on the phone.
- You don't know whether that freight quote is a fair market rate or whether you're leaving money on the table.
In each of those cases, you're currently relying on gut feel, experience, or whoever happens to be paying attention that day. That's not a knock—that's just the reality of running a lean operation. But the same class of models that cracked Stratego can now power decision agents that triage these exact problems, flag the right signals, and route actions—even when the picture isn't complete.
The key word is agent. Not a dashboard. Not a report. An active system that reads your incoming signals, reasons about what's likely happening behind the curtain, and surfaces a prioritized action for you or your team to execute.
What This Actually Looks Like Inside a Real Operation
I run an e-commerce and import operation. I've automated most of it myself using AI agents, large language models, and workflow tools like n8n. So when I talk about hidden-information decision agents, I'm not describing a whitepaper concept—I'm describing something I've actually built and use.
Here's a concrete example. Supplier risk flagging: my operation monitors order confirmation windows, lead time variances, and communication patterns from each supplier. When the signals start drifting—a reply comes two days late, a shipment estimate shifts by a week—the agent doesn't wait for a crisis. It surfaces a risk flag with a suggested action before the delay hits my customers. No analyst on staff. No manual monitoring. The agent infers risk from incomplete signals, exactly like an AI playing Stratego infers where the high-value pieces are hidden.
You can build the same logic around:
- Inbound lead triage — score and route leads based on behavioral signals, not just form fills
- Customer churn prediction — catch disengagement patterns before a customer goes quiet
- Inventory reorder timing — factor in supplier reliability history, not just reorder points
- Customer support routing — identify high-risk or high-value tickets before a human even opens them
None of these require a data science team. None of them require perfectly structured data. They require the right agent architecture and someone who knows how to build it for your specific operation.
The Budget Question
The Stratego breakthrough matters partly because of the budget angle. This wasn't a resource-unlimited research lab flexing. It was a lean, efficient system that achieved top-of-the-world performance under real constraints. That's the model for how AI gets deployed in small and mid-sized businesses—not enterprise-scale infrastructure, but targeted agents solving high-value problems at a cost that makes sense for your margins.
The window where only large companies could afford this kind of decision intelligence is closing fast. Operators who move now build a compounding advantage. The ones who wait until it's obvious are paying a catch-up premium.
The AI that beat the world's best Stratego player didn't have more information than its opponent. It just made better decisions with what it had. That's the capability now available to your operation.
At Maqia, we build exactly these kinds of decision agents for operators like you—no code on your end, no data science team required. If you want to see what a hidden-information agent looks like inside a real operation similar to yours, the next step is simple: book a call with us at maqia.co. We'll map out where the highest-value incomplete-information problems are in your business and show you what an agent built to handle them actually looks like in practice.