Tool · XLSX · version 1.0
Ranking an AI portfolio by what it costs to find out
A ranking model that puts a modest prize you can settle in six weeks above a large one that takes nine months and two million to test.
Free. No sign-up, no email required, no interstitial. Share it with whoever needs it.
Who it is for
Anyone holding a portfolio of AI pilots that nobody can rank, sponsored by people who cannot compare them.
What is in it
- A scoring model: value at stake, discounted by confidence and data readiness, divided by what it costs to find out
- Blocking rules: no decision owner, no falsifier, or no data, and the row scores zero
- A kill gate with six triggers, to be run monthly
- A worked example of eight candidates in which the largest prize ranks last of those that score at all, and two more are blocked outright
What it will not do
The worked example is illustrative. This workbook rests on method rather than on a public dataset, which is a weaker kind of proof than our telecom analyses and a different one.
Sources
Method. The figures in the worked example are invented to show the ranking behaviour and are not a benchmark.
If you use this and find an error, tell us at hello@blueprysm.com and we will correct it and say what changed. Last updated 29 September 2026.