Value Confidence Calculator
Two projects on the same roadmap. One projects $2M in new revenue. The other saves $400k a year in labor, already measured.
Five to one, if you rank on the headline. But $2M of revenue at a 25% contribution margin is $500k of actual value, and now it's close. Then account for the fact that the $2M came from a survey and the $400k came from a time study, and it reverses.
Two corrections, both invisible in the raw numbers. The first is units — revenue, margin, and cost savings are not the same dollars. The second is evidence, which is what this tool is about.
A pricing study is strong evidence about price and nearly worthless evidence about adoption. One number can't express that. A product of per-factor multipliers can.
The score is what you rank on. But the more useful thing usually falls out alongside it: which single unknown is deciding the outcome, and what it would cost to resolve. Often that's a three-week test, not a six-month build.
In practice you don’t pick this. A published table does, based on what evidence you can point to. Here you’re picking so you can see what the table would do.
Rung values are v1 — the ordering follows published research on forecast optimism and stated-preference bias; the specific multipliers are chosen spacing, not measurements.
- Raw value
- $1,866,240
- Blended confidence
- 0.95 × 0.50 × 0.70 = 0.33
What it costs
The 15% default is a heuristic — the common rule of thumb for annual software maintenance sits in the 15–20% of build range — not a measurement of anything.
$620,525 adjusted against $839,500 first-year cost. To clear it, the estimate would have to be right 45% of the time; the evidence supports 33%.
Where to buy evidence
Gain from one rung up, per factor
Buy the evidence
The useful output is both numbers. The adjusted value is what you rank on — an estimate you can defend across a portfolio. And alongside it: which unknown decides the outcome and what it costs to resolve. If a paid pilot costs 5% of the build and flips the decision, run the pilot. Arguing about the estimate costs hours from expensive people and settles little; buying the next rung of evidence is cheaper and settles it.
Calibration
The rung values in this tool are v1. The honest way to run this framework is to record claimed value against realized value on shipped work, then replace these ratios with your own observed ones. Until then, treat the multipliers as a shared starting point for argument, which is already more than a bare headline number gives you.
On the numbers
The rung values here are mine, not anyone's research finding. No study says a stated-preference survey is worth 0.5. What the literature supports is the ordering: intent predicts behavior conditionally and unevenly, and estimates made from inside a plan run optimistic. The specific multipliers are plausible spacing chosen to make the model behave sensibly, and they should be replaced with your own measured claim-versus-actual ratios once you have eight or ten shipped projects to compare. Sources below argue for the shape, not the values.
Sources
- Morwitz, V., Steckel, J. & Gupta, A. (2007). “When do purchase intentions predict sales?” International Journal of Forecasting, 23(3), 347–364. — the adoption ladder's existence and ordering
- Flyvbjerg, B., Holm, M.S. & Buhl, S. (2002). “Underestimating Costs in Public Works Projects: Error or Lie?” Journal of the American Planning Association, 68(3), 279–295. — the contingency toggle
- Flyvbjerg, B. (2006). “From Nobel Prize to Project Management: Getting Risks Right.” Project Management Journal. — reference class forecasting; the calibration plan
- Kahneman, D. & Lovallo, D. (1993). “Timid Choices and Bold Forecasts.” Management Science, 39(1). — the outside view
- Hubbard, D. (2014). How to Measure Anything, 3rd ed. — expected value of information: the “buy the evidence” step