The Best Forecasters Are Not the Most Certain

Russ Powell

49 minutes ago

6 min read

How good forecasting habits can help managers make better decisions under uncertainty

The Best Forecasters Aren’t the Most Certain

It's prediction season.

Football is underway. Baseball is heading toward October. The November 3 midterm elections are approaching. Everywhere you look, someone is offering odds or probabilities about what happens next. Increasingly, some of those forecasts aren't coming from people at all.

Managers are in the prediction business too. We just don't usually call it that. No top hat, no crystal ball. We just say it with a straight face:

"We'll ship by the end of March."

"That customer will renew."

"Trust me. She'll be successful in this role."

Plans, budgets, hiring decisions, strategy, and many everyday management calls depend on judgments about a future we cannot know yet. So how do you help a team make good decisions when no one can know for certain what happens next?

What Great Forecasters Figured Out

A few years ago, I wrote about research into superforecasters—people unusually good at predicting future events.

They tended to be curious, humble, and open-minded. They thought in probabilities, updated their forecasts, and kept track of how well they did.

Those habits hold up. But something else stands out to me now: they're also habits of good leadership—and especially useful habits for managers. A manager's job is not simply to arrive at the right answer personally. It's also to improve the quality of the thinking happening around them.

Make Uncertainty Discussable

Consider the difference:

"We'll ship by March 31."

"I'm about 70% confident we'll ship by March 31."

The second sounds less decisive, but it gives the team much more to work with. Why 70% rather than 90%? What assumptions are behind it? What could derail us? What would change our confidence?

Putting a number on your confidence doesn't make the forecast accurate. It makes the judgment visible enough for the team to examine. That's one of the most useful things a manager can do with uncertainty: bring it into the conversation instead of slathering it with certainty.

That's harder in organizations that reward certainty. If managers get punished for saying, "I'm 70% confident," they'll learn to say, "Yes, we'll make the date." Senior managers, executives, and boards can insist on rigor without demanding false certainty: What's your best judgment? How confident are you? What would change your mind?

Make Disagreement Useful

Good forecasters don't just look for evidence that supports what they believe. They actively look for information that might prove them wrong. That's also a useful stance for a manager.

Compare "Here's what's happening" with "Here's how I'm seeing it. What am I missing?" When the person with more authority sounds certain, others have to decide whether challenging that certainty is worth the effort. Some will. Others may keep concerns to themselves.

A manager who treats a conclusion as a working hypothesis sends a different signal: contrary information is useful here. This is one reason I encourage the managers and management teams to work with to watch out for the Knower Trap. We fall into it when we become so certain we understand a problem that we stop looking for—and sometimes stop hearing—evidence that suggests otherwise.

Make Changing Your Mind Respectable

Good forecasters update their thinking when new information arrives. Teams don't always work that way. Once we've committed to a plan, advocated for a hire, or argued strongly for a strategy, changing our minds can feel like admitting we were wrong.

Good managers can make changing course respectable:

“Given what we knew then, that was our best judgment. We know more now, so we're changing it.”

That's not indecision. It's learning, and sometimes wisdom.

A team that treats changing your mind as losing will eventually defend some very bad decisions.

AI as Partner, Not Oracle

Here's what's changed dramatically since that first post: we now have another powerful forecaster at the table.

In a pre-registered study of 991 people, participants using more capable LLM forecasting assistants improved their accuracy by 24–28% compared with a control group using a less capable model.[1] The results were unusually sensitive to one forecasting question, so I'd treat this as one promising study rather than a settled rule.

And on ForecastBench, several AI systems have recently performed at levels statistically indistinguishable from a benchmark based on elite human superforecasters’ forecasts collected in 2024.[2]

For managers, though, the interesting question isn't whether AI can beat a human forecaster. It's whether AI can help a team think better—by finding relevant base rates, surfacing contrary evidence, and identifying assumptions people may have missed.

AI doesn't know which question matters most to your business. It doesn't own the tradeoffs or the consequences.

Use AI as a partner in judgment, not an oracle.

Treat the Number as Evidence, Not Truth

I'm no prediction-market expert, but the scale caught my attention. In August, combined trading volume on Kalshi, Polymarket, and Polymarket US was about $45 billion.[3] For rough context, U.S. legal sportsbooks handled about $167 billion during all of 2025.[4]

Those aren't apples-to-apples measures: the prediction-market figure is global, and contracts can trade repeatedly, while the sportsbook figure is U.S.-only. Still, the scale is striking.

Between football season, baseball's postseason, and the midterms, we'll see precise-looking probabilities throughout the fall. They can be useful, but precision can exceed the quality of the underlying evidence. The same is true of AI forecasts, expert forecasts, and our own:

Treat the number as evidence to examine, not an answer to obey.

Make Learning Routine

One of the less glamorous habits of good forecasters is that they keep track. In my experience, most teams don't. It's easier to remember the prediction we nailed than the confident calls that were quietly off.

After the fact, we also confuse outcomes with decision quality. A successful project can make the original decision look brilliant; a failure can make it look foolish. But good outcomes don't always mean good decisions, and bad outcomes don't always mean bad decisions.

One simple way to learn is to keep a forecast log. You might use headers like these:

"70% chance this customer renews by December 15."

Then go back and look. Were you too confident? Too cautious? Which assumptions held up? Which warning signs did you miss?

The goal isn't to catch people being wrong. It's to help the team get better at making judgments under uncertainty.

Try Three

Tomorrow, write down three forecasts that will resolve before year-end: one about work or business, one personal, and one outside either—sports, the economy, technology, public affairs, whatever interests you.

Make each specific enough to resolve cleanly. Assign a probability, note what would cause you to change it, and give it a resolution date.

Better yet, try it at your next management-team meeting. Pick one important assumption behind a current plan—for example, "There's a 70% chance we'll hit the March 31 release date." Have everyone estimate the probability independently, compare the numbers, and ask what explains the differences.

I'll come back to this exercise near the end of the year and look at how to score the forecasts and, more importantly, what we can learn from them.

Managing well rarely means knowing what comes next. More often, it means helping a group think clearly about what might happen so they can make better decisions together before the answer is known.

Notes & Sources

[1] Philipp Schoenegger, Peter S. Park, Ezra Karger, Sean Trott, and Philip E. Tetlock, “AI-Augmented Predictions: LLM Assistants Improve Human Forecasting Accuracy,” ACM Transactions on Interactive Intelligent Systems, 2025.

[2] Forecasting Research Institute / ForecastBench, 2026 reporting on AI forecasting performance relative to its recorded human-superforecaster benchmark.

[3] The Block, September 2026, reporting approximately $45 billion in August trading volume across Kalshi, Polymarket, and Polymarket US.

[4] American Gaming Association, “Commercial Gaming Revenue Hits $78.7 Billion in 2025,” February 26, 2026. U.S. state-regulated sportsbooks reported a total handle of $166.94 billion in 2025.