Like a lot of people, I filled out an NCAA tournament bracket this year.
Unlike a lot of people, I did it with AI.
Not in the lazy sense of asking for “the best picks,” but in a more structured way: using AI to separate the most likely outcomes from the most valuable contrarian plays, think through the logic behind each choice, and pressure-test where the bracket was most vulnerable to surprise.
The results were pretty good. The model got 13 of the Sweet 16 teams right, which is strong by any normal standard. But it also missed some important upsets. A few favorites I trusted lost anyway. The process was better than intuition alone, but it was still far from perfect.
That turns out to be a useful lesson for anyone thinking about using AI in business.
Because the real value of AI is not that it magically tells you the future. It is that it can help you think more clearly about uncertainty, tradeoffs, and risk.
That is a much bigger deal than it sounds.
AI is best used as a thinking partner, not an oracle
A lot of the marketing around AI still encourages a kind of fantasy. Ask a smart machine a question, get a smart answer, move forward with confidence.
That is not how serious decision-making works.
In the bracket exercise, AI was genuinely helpful in a few specific ways. It helped structure the decision. It helped separate “highest probability” from “best expected value.” It helped identify relevant variables, compare paths, and make the assumptions more explicit.
But it did not eliminate chaos. It did not make basketball predictable. It did not prevent one-possession games from breaking the wrong way.
That is exactly how executives and investors should think about AI in planning. Its strength is not certainty. Its strength is better process.
If you use AI to sharpen your assumptions, widen your field of view, and stress-test your plans, it can be extremely valuable. If you use it to manufacture confidence, you are likely to become more confidently wrong.
The most likely answer is not always the best decision
One of the most useful parts of the exercise was separating two different goals.
The first bracket was built around the teams most likely to win. The second was built around expected value in a pool with lots of entrants. Those are not the same thing. The “best” pick depends on the context.
That distinction matters far beyond sports.
In business, leaders often talk as if there is one correct strategy. Usually there isn’t. There is the most likely outcome, and then there is the best move given competition, payoff structure, timing, and downside risk.
Investors understand this instinctively. The company most likely to survive is not always the best investment. The market leader is not always the best opportunity. A safe-looking plan is not always the plan that creates the most value.
AI can be very useful here because it can help teams model different objectives instead of collapsing everything into a single recommendation. It can help answer questions like: what is the base case, what is the high-upside case, where are we underestimating risk, and what would a smarter rival do with the same information?
That is much more useful than asking for “the best strategy.”
AI’s hidden value is that it makes assumptions visible
This may be the most underrated thing AI does well.
Most planning processes are full of hidden assumptions. Teams may broadly agree on a market opportunity or product roadmap without ever clearly articulating what has to be true for the plan to work. Growth assumptions go unexamined. Competitive responses get ignored. Fragile dependencies stay buried.
AI can help pull those assumptions into the open.
Used well, it forces more explicit thinking:
What variables matter most? Which ones are stable? Which ones are noisy? What are we assuming about customers, timing, pricing, competition, regulation, talent, or capital markets? Where are we overconfident? What would break this model?
Even when the final output is imperfect, that kind of clarity is valuable. In many cases, the biggest gain is not that AI gives you a better answer. It is that it helps the organization ask better questions.
Good prompts are often just good management
When people say AI gave them a generic answer, the problem is often not the model. It is the question.
If you ask for a single polished recommendation, you will often get something that sounds authoritative and tidy. That may feel useful, but it can hide the real complexity of the decision.
The better approach is to ask for structure. Ask for multiple scenarios. Ask for the strongest case against the plan. Ask for variables most likely to change the outcome. Ask what the analysis would look like if your core assumption is wrong. Ask where the forecast is most fragile.
That is not just prompt engineering. That is management discipline.
The executives who get the most out of AI will probably not be the ones who treat it like magic. They will be the ones who use it to impose rigor on their own thinking.
Every plan needs a volatility layer
One reason the bracket model missed some upsets is that it did a decent job measuring strength but a weaker job measuring fragility.
That happens in business all the time.
Companies build plans around average outcomes and then act surprised when variance shows up. A product launch depends on one partner. A sales forecast depends on a handful of accounts. A supply chain has no margin for disruption. A pricing strategy assumes customers will behave rationally. A budget assumes capital will remain available.
The base-case plan may look fine right up until it doesn’t.
AI can help here too, but only if you ask it to. Instead of just modeling the expected path, use it to identify the biggest sources of volatility. Where are the swing factors? Which assumptions have the widest range of outcomes? What are the hidden correlations? Which small failures could cascade into larger ones?
A lot of strategic mistakes are not caused by bad core logic. They are caused by underestimating variance.
Keep score, or don’t bother
The most useful part of the bracket exercise was not making the picks. It was reviewing them afterward.
What did the model get right? What did it miss? Were the misses random, or did they expose a pattern? Was the process strong but slightly too conservative? Did it underestimate volatility? Did it overrate a few favorites just because they looked clean on paper?
Businesses need the same feedback loop.
If companies use AI to generate strategy memos, forecasts, or operating plans but never compare those outputs to reality, they will learn almost nothing. AI will become a presentation layer instead of a decision tool.
The organizations that benefit most will be the ones that treat AI like a process they can improve. They will look at misses, recalibrate assumptions, and build better questions over time.
The real opportunity
The big opportunity with AI is not that it will replace leadership judgment.
It is that it can make leadership judgment more disciplined, more explicit, and more probabilistic.
That matters because many bad decisions are not caused by a lack of intelligence. They are caused by fuzzy assumptions, unchallenged consensus, and overconfidence dressed up as planning.
AI can help with all of that. It can broaden the analysis, accelerate scenario work, stress-test ideas, and surface blind spots. But it still needs people to decide what matters, what is noise, and what deserves conviction.
That was the real lesson from the bracket.
The model did well. Reality was messier. And that is exactly why AI is useful.