Foldback turns an argument about a person into a disagreement about a number. You lay out the options, what could happen after each one, and what each path costs or earns. The app works out which option leads, the single number that would change your mind, and what is worth finding out before you decide.

Why the name? Folding back is the classic way to solve a decision tree: start at the outcomes on the right, work back to the decision, average over chance at each circle and keep the best option at each choice.

▶ Watch the 4-minute demo: the tool in brief, then “Build it with me” on a real case.

Reading the tree

An option's certain value is what choosing it costs (−) or earns (+) for sure. An outcome's value if it happens is added only on that path. Costs are negative everywhere, so every formula is a plain sum. The app then folds the tree back from the right: each chance circle is worth the probability-weighted average of what follows it, and the option worth most is marked yellow.

Worked example: should we replace Sam?

Load the example and follow along in the panel on the right:

1. Where it stands. Exit now leads by $4,000 on the likely values. Because the key numbers are ranges, the app also plays out 10,000 plausible worlds. Exit comes out best in only about six worlds in ten, so this is far from settled. The chart below it shows every option's likely value and how often each comes out best.
2. What would change your mind. If the chance coaching works rises above 43.3% (now 30%), coaching wins. That value sits inside the range the team gave, so this is the number the discussion should be about. Likewise, if Sam's gap is less than about $10k a year (now $16k), keeping Sam wins.
3. Try it. In Shared numbers, change the Likely value of chance_coaching_works from 30% to 45%. Coaching takes the lead immediately. Change it back, or press Undo.
4. What is worth finding out (under Go deeper). Pinning down Sam's real gap is worth the most, a couple of thousand dollars at most. Pinning down the exit cost is worth almost nothing, so don't spend time arguing about it. (The article ranks the chance coaching works first; its figures differ a little because it treats the ranges differently.)
5. Is a diagnostic month worth it? Tick Price one. A month of coaching that reads right 75% of the time is worth about $2,000 as information, but costs about $2,900 (fees plus another month of the gap). Decide now, or find a sharper test.
6. Meeting note. Copy a plain-text summary, including what each number rests on, into minutes or an email.

Building your own

• New to decision trees? Click Build it with me. It asks one question at a time (your options, what could happen, how likely, what each costs or brings) and builds the tree for you. You can change anything on the tree afterwards.
• Click New blank tree, type the decision as the title, and name your options. Press Enter in a name to add the next one.
• Measure every option against the same baseline, such as an average performer in the seat. If every option comes out negative, the best one is the smallest loss.
• Click + what could happen? on an option to split it into chance outcomes, and + Add an outcome below them for more. Hover over the round chance circle and click its ✕ to remove the whole group at once (on a touch screen, tap it twice). Any outcome can split again. Leave one outcome's chance empty and it takes the rest (1 minus the others).
• If an option ends in a hire, the hire is a chance too: will they stay, will they perform? Give it its own outcomes.
• Put anything you use in more than one place, or want to argue about, in Shared numbers, and type its name in the tree. Fill Low and High to make it a range. Give it a plain description, such as “Sam's yearly gap vs an average performer”: the results use it instead of the name.
• Tag each shared number as data, expert judgement or a guess. Tags are a record only and never change a number.
• Drag the background to move around a big tree. Fit, the zoom buttons or Ctrl + scroll resize it.

What you can type in a field

You type	Meaning
30%, 0.3	a probability
-16k, 2.5m	a value (k = thousand, m = million)
-37k to -25k	a range, low to high, likely value in the middle
-37k to -25k ~-30k	a range with the likely value given
cost_of_exit	a shared number
cost_of_coaching + sam_vs_average_per_year / 2	a formula (min, max, if and more also work)

Keep in mind

• A confident, narrow, invented range gives a confident, narrow, invented answer. Build the tree with the decision-maker, using their numbers.
• Options are compared at your likely values. If the average over all the plausible worlds would put a different option ahead, which can happen when a range leans to one side, the panel says so.
• “Fairly safe” means the leader comes out best in at least 90% of plausible worlds, “leads, but not safely” 70% to 90%, and below that “far from settled”. These cut-offs are rules of thumb, not statistical tests.
• Ranges are treated as independent, there is no attitude to risk (averages only), and nothing is discounted over time.
• The tree prices only what's in it. Fairness, legal exposure and the effect on people and teams count only if you add them.