Turn guesses into odds.
Most plans hide their guesses inside single numbers. In Hunch every uncertain number is a range you're 90% sure about. Hunch simulates thousands of possible futures and tells you the odds, what drives them, and which unknown is worth checking before you decide.
Below is an example. Change any number, or start from another example under Models. Nothing you type leaves this device.
How Hunch works
Hunch is a calculator for numbers you don't know exactly. Instead of pretending that next month's sales will be 400, you say "I'm 90% sure it will be between 150 and 900". Hunch then plays out 20,000 possible futures, each with its own combination of values, and shows what happens across all of them.
Kinds of numbers
- 90% range
- Low and high values you are 90% sure the truth lies between: there should be only a 5% chance it's below your low and 5% that it's above your high. Auto shape uses a symmetric bell curve, or a right-skewed curve when the range is positive and the high is at least 3× the low (sales, delays and costs often behave that way). You can choose Symmetric, Skewed or Flat (every value equally likely) yourself. A range whose low is 0 or more never produces negative values.
- 3-point
- Minimum, most likely and maximum (a triangular distribution). Useful when you have a clear "most likely" value.
- Chance
- Something that happens or doesn't, with a probability. It is 1 in futures where it happens and 0 where it doesn't, so
rain_delay * rain_daysadds the delay only when it happens. - Exact
- A number you know, such as a price you set.
- Formula
- Calculated from the others, futures by future.
You can type 1.5L, 2 cr, 25k or 1,50,000 in number boxes.
Formulas
Use the short names shown under each number (tap them to insert). Operators: + - * / ^, comparisons > < >= <= == !=
(they give 1 for true, 0 for false), and, or, not, and percentages like 18%.
Functions: min(a, b, …), max(…), sum(…), avg(…), if(condition, then, otherwise),
round(x, digits), floor, ceil, abs, sqrt, pow(x, y), exp, ln, log10, clamp(x, low, high).
A formula that is a comparison, like needed > prepared, shows the chance that it is true.
Reading the results
The chart shows how often each value came up across the simulated futures. The 90% range runs from the 5th to the 95th percentile; the median is the middle value; the average is what you'd get per attempt over many repeats. Set a line (for example 0 for profit, or a deadline) to see the chance of landing above or below it.
What drives it shows, for each uncertain input, the average result when that input is in its lowest 10% and its highest 10%, and how much of the spread it explains on its own. Long bars are the inputs that matter.
Decisions and what's worth finding out
Turn on Treat this as a decision when you are choosing between going ahead (you get the result) and not (you get the line's value, such as 0). Hunch tells you which choice is better on average with what you know now, and how often the other choice would have turned out better.
It also computes the value of information: how much better your decision would be, on average, if you knew a number exactly before deciding. That is the most it makes sense to spend — in money, time or effort — on finding it out (a survey, a trial run, a quote). If it's tiny, stop researching that number; if one input dominates, find that out first. This is the method health-economics agencies use to decide which research to fund (expected value of partial perfect information, estimated from the simulation itself with the regression approach of Strong, Oakley & Brennan, 2014).
Track record
Studies of 90% ranges find that people's ranges usually catch the truth far less than 90% of the time — often only around half. When you later learn the real value of an estimate, enter it under Record the real value (or add any prediction on the Track record page). Hunch shows how often your ranges were right and how much wider they would have needed to be. You can let Hunch widen your ranges automatically by that amount. The adjustment is cautious when you have few records.
Limits
- Inputs are treated as independent. If two inputs move together (price and sales, say), write that into a formula, e.g.
units = base_units * (300 / price). - Results are only as good as the ranges you enter. The value of an example is in the questions it makes you ask, not in false precision.
- Hunch is a thinking tool, not financial, legal or medical advice.
Privacy
Models, settings and your track record are stored only in this browser. There is no account and no server storage. A share link contains the model itself (compressed in the link), not a copy on our server; real values you recorded are left out of shared links. Use Models → Back up to save everything to a file.