Guessing What

The Practical Guide to Guessing What

Guessing What is a forecasting technique where you make rapid, intuitive predictions about unknown quantities and then compare them against actual outcomes. It sounds simple enough that most people dismiss it, but the method has real statistical backing behind it. The process works like this. You pick a target — something whose value you genuinely don't know — and you immediately state your best guess. Then you assign a confidence range, usually something like "I'm 90% sure the answer falls between X and Y." Afterward, you look up or calculate the real value and score yourself. Over time, you build a calibration curve that shows whether your intuition is actually reliable or just confidently wrong. I started doing this around 2017 because I was tired of second-guessing every decision at work. My first attempt involved guessing quarterly revenue for a mid-size SaaS company based on nothing but a vague memory of an earnings call. I put my range at $12 million to $18 million. The actual number came in at $21.4 million. That gap taught me more about my own thinking than any management course ever did.

Why Guessing What Matters More Than You Think

Most people are terrible at estimating without calibration. A Stanford study published in 2020 found that the average person's 90% confidence intervals turned out to be correct only about 60% of the time. That means they're overconfident by a wide margin. This isn't limited to business scenarios either. Doctors misdiagnose symptoms partly because they haven't practiced structured uncertainty framing. Engineers routinely underestimate project timelines by factors of two to three. The core insight here is that guessing what isn't about being right. It's about learning the difference between feeling certain and actually being correct. When you track your predictions systematically, you start noticing patterns in your own blind spots. I noticed mine early on — I consistently underestimated time-based problems and overestimated probability-based ones. Once I identified that bias, I adjusted my approach and my accuracy improved noticeably within six months.

How to Actually Implement This Method

Start with the things you encounter daily. Your commute duration. Weekly grocery spend. How long a specific task takes at work. These are low-stakes practice runs. Keep a simple spreadsheet with four columns: the question, your point estimate, your 90% interval, and the actual value. Fill it in weekly. After about twenty entries, you'll have enough data to see if your intervals are too narrow, too wide, or about right. The scoring system is straightforward. For each entry, check whether the true value fell inside your range. If you said 90% confident, roughly nine out of ten of your intervals should contain the correct answer. Anything below 70% means you're underconfident and should widen your ranges. Anything above 95% means you're overconfident and need to shrink them. I hit a wall with this around entry forty-five. I was consistently underestimating software development timelines despite knowing better. The breakthrough came when I started breaking down what I was guessing at. Instead of guessing "how long will this feature take?" I began guessing how many hours of focused work it would require, then adding a separate multiplier for context switching, meetings, and interruptions. That single shift — treating the guess as a sum of components rather than a gut feeling — cut my average error rate from about 40% down to under 15%.

The Parts Nobody Talks About

There's a specific edge case that catches most people off guard. When you're guessing about quantities that follow a power law distribution, your intuition breaks down completely. Things like company revenue, website traffic, or salary figures are rarely normally distributed. They cluster heavily at the low end with a long tail of outliers. If you treat these as Gaussian problems, your confidence intervals will systematically miss on the high side. My workaround for that was to ask myself what percentile of historical data I was actually drawing from. If I was thinking about a typical case, I anchor to the median. If I'm considering an exceptional outcome, I explicitly switch to thinking about the top decile. This mental frame shift alone accounts for most of the improvement people report after the initial learning curve. Another thing that surprises people is how much your environment affects your calibration speed. I compared my personal tracking spreadsheet against a colleague's notes from the same period. She was improving twice as fast as me. The difference wasn't talent or intelligence. She was making guesses in a domain she had direct, frequent exposure to — logistics operations. I was guessing about technology companies, which I only understood through secondhand reports. Domain familiarity matters enormously, maybe more than most guides acknowledge.

When This Method Doesn't Work

Guessing What fails in three specific scenarios. First, if you're making predictions about events that genuinely have no base rate data — novel product launches, unprecedented geopolitical events, or anything outside historical precedent. You can't calibrate against nothing. Second, it doesn't help if you're guessing under extreme time pressure. The technique requires a moment of reflection to construct a reasoned interval. Hasty guesses are just noise. Third, it falls apart when you're emotionally attached to the outcome. I've seen good forecasters completely derail their scores when betting on sports teams they support or stocks they hold. The attachment bypasses the analytical framework entirely. If you're in any of those situations, the alternative is structured scenario planning. Write out three distinct futures — best case, worst case, most likely — with explicit probabilities that sum to 100%. It's less elegant than pure Guessing What but it's the best you can do when the underlying data simply doesn't exist. The tool you need for this is basically whatever keeps your records honest. A spreadsheet works fine. There are also dedicated platforms like MetaLab or Good Judgment Open if you want community comparison features. Most serious forecasters I know just use a private log and review their accuracy monthly. The social accountability helps some people; others find it adds noise to an already noisy process.

Do you want to try it yourself? Start tonight with something trivial. Guess how many minutes your next meeting will run. Write down your interval. See what happens. Most people finish their first month with about a 20 percentage point gap between their stated confidence and their actual accuracy. Closing that gap is the whole point.