OpenAI Five beats humans at Dota 2

In 2018, an artificial intelligence team called OpenAI Five sat down to play Dota 2, one of the most complex team strategy video games in existence, and started beating human players. What’s interesting isn’t just the result, but how it got there: no one explained the optimal rules to the system or gave it a playbook. It learned by playing against itself over and over, making mistakes, adjusting, and improving, until it discovered strategies that not even human experts had considered.

That approach has a name: reinforcement learning. And although it sounds like labs and video games, the underlying idea is deeply practical for any business. It’s about an AI that doesn’t just classify or predict, but decides in changing scenarios, where each action affects what comes next. For a Mexican small business, that opens a different conversation: not “how do I automate a repetitive task,” but “how do I better support the decisions that today depend on one person’s intuition.”

What reinforcement learning really is

Imagine someone new in your warehouse learning to fill orders. At first they make mistakes: long routes, poorly stacked boxes, wasted time. With practice and feedback, they gradually find the best way to move around. Reinforcement learning works with that same logic, but at a scale and speed no person could match: the system tries paths, receives a signal of how well it did, and adjusts its behavior to improve the result.

What’s valuable for a business isn’t the video-game part, but the principle: there are problems that aren’t solved with a fixed rule, but by making good decisions one after another. When to restock, what price to offer, how to assign routes or shifts, in what order to fill orders. These are decisions where the context changes constantly and where accumulated experience is worth its weight in gold.

From the lab to a small business’s operation

You don’t need a research team to take advantage of this way of thinking. What you do need is to identify where your company makes repeated decisions that today depend on someone’s “eye.” That’s where applied AI, built on custom software, can become a reliable copilot:

• Inventory and purchasing: anticipate when and how much to restock so you don’t run short or overload the warehouse.
• Routes and logistics: propose the best delivery order considering traffic, zones, and priorities.
• Service and follow-up: suggest which customer to contact first based on likelihood of closing or risk of losing them.
• Resource allocation: distribute shifts, staff, or equipment where they pay off most at each moment.

The detail that makes the difference: your data and your context

Here’s the point that often gets overlooked. OpenAI Five learned by playing millions of matches because it had a clear environment to practice in and a clear signal of success. In your company, that “environment” is your data: your sales history, your real delivery times, your customers, your particular way of operating. A generic AI knows none of that; a custom solution does.

The advantage isn’t in having the most advanced AI, but in applying it to what only your business knows: your data, your processes, and your history.

That’s why the sensible path isn’t to buy a magic tool and expect miracles, but to build on what you already have. You keep your data, your processes, and your history, and on that foundation you build a system that learns from your real operation and gives you back decisions with reliable information, not hunches.

Start with a decision that matters

The first thing is to bring the topic down from the tech clouds to something concrete and measurable. These steps help:

• Pick a single repeated decision. Look for that point where someone decides “by eye” several times a day or week and where getting it wrong is costly.
• Gather the history you already generate. You don’t need perfect data to start; you need real data on how that decision has gone in the past.
• Start small and measurable. Define from the start how you’ll know if it works: fewer stockouts, faster deliveries, more closings. A small pilot teaches more than a huge project.
• Keep the person in charge. The AI proposes, your team decides. That setup builds trust and improves results instead of replacing human judgment.

OpenAI Five showed that a machine can learn to decide well in very complex scenarios. The good news for small businesses is that you don’t have to play Dota 2 to take advantage of the idea: it’s enough to choose a decision that matters, feed it data, and pair it with software made to your measure. At Normandia Web we help you choose that first decision and build the system that backs it, so technology works in favor of your results.