Practices for making — and improving — calibrated judgments about what comes next.
Most decisions rest on implicit forecasts. Will this hire work out? Will this market move the way we expect? Will this initiative pay back inside three years? Executives forecast constantly and rarely score themselves. The research on what separates accurate forecasters from inaccurate ones is unusually clear, and the practices are learnable. None require a model or a spreadsheet — they require habits of mind, applied before you commit to the call.
Borrowed from Tetlock's superforecasting research. These work even without tracking — they work much better with it.
Break a hard forecast into smaller pieces you can actually estimate. "Will the California commercial casualty market harden over the next 18 months?" decomposes into: How are loss-cost trends moving? How is reinsurance capacity changing? How are competitors pricing? Each sub-question is more tractable than the whole. Fermi estimation is the formal name; the move is to never guess at a big number when smaller numbers will get you there.
"This will probably happen" is unfalsifiable and unhelpful. "I'd put this at 60–70% likely" is a real forecast. The range itself encodes your uncertainty. The discipline of putting numbers on it — even rough ones — surfaces overconfidence faster than any other practice.
Before reasoning from the inside ("here's why our case is different"), reason from the outside ("how have similar cases gone?"). New M&A integrations? Look at base rates for synergy realization. New technology rollouts? Look at base rates for adoption timelines. The outside view is almost always more pessimistic than the inside view, and almost always more accurate.
When new evidence arrives, the question is not "does this change my view?" — it's "by how much, and in which direction?" A Bayesian update moves your estimate by an amount proportional to the evidence's strength. Big shifts on weak evidence and no shifts on strong evidence are both miscalibration; both signal that something other than evidence is driving your view.
The inside view is what you know about this specific case — the deal, the candidate, the project. The outside view is what's true of cases like this one in general. Both matter. The mistake is letting the inside view dominate by default. Ask both questions explicitly and notice when they disagree; the disagreement itself is information.
A coherent story about why something will happen feels like evidence and isn't. Good forecasters notice when their confidence is coming from the elegance of an explanation rather than the strength of the underlying facts. If you can equally well construct the opposite story from the same facts, your story isn't doing the work you think it is.
The practices above raise forecast quality on their own. The thing that compounds — the thing that turns a competent forecaster into a calibrated one — is the feedback loop. You write down your prediction with its probability and date. Time passes. You go back and check.
Without this loop you cannot tell whether your judgment is improving. With it, even briefly, you start to notice patterns: domains where you're systematically overconfident, kinds of questions where you're well-calibrated, the difference between "I felt sure and was right" and "I felt sure and was wrong." That's the actual skill.
The lightest viable version is one line per forecast in a single Word document, OneNote page, or calendar entry. Date, the question, your probability, and one sentence on why. Revisit when the answer is known. That's it.
Before you commit to a call — read this once, then make the forecast.