Scoring Fit with AI
Write a scoring rubric, have AI apply it with reasons, and keep the scoring away from anything that decides a person's eligibility.
What you will be able to do
Write a fit-scoring rubric, turn it into an AI prompt that returns a score with reasons, and name the kinds of data a GTM score must never use.
What a fit score is
A fit score is a number from 0 to 100 that says how well an organization matches your ICP. It helps you decide who to contact first, and who not to contact at all. In the Temerarii method, only rows above a threshold you choose, often somewhere around 60 to 70, get a message. Everyone else waits or is skipped.
Start with a rubric, not with AI
A rubric is a written list of what earns points. Write it before you open any AI tool. If you cannot write the rubric, the AI cannot score fairly either.
| Criterion | Points | Example evidence |
|---|---|---|
| Matches the "Who" in the ICP | 30 | It is a public school, library, or youth center |
| Has a fresh signal | 25 | A grant award in the last 30 days |
| Meets the fit facts | 25 | Has a room and adult staff after school |
| Within our reach | 20 | Within one hour of our team |
| On the "Not a fit" list | Score is 0 | Charges families; unsafe to reach |
Let AI apply the rubric
AI is good at reading a messy row of facts and applying a rubric to it. Ask it to return the score, the points for each criterion, and one sentence of evidence for each. Reasons make the score checkable. A score with no reasons is a guess you cannot review.
Fit Scoring Lab
Build a prompt that scores one organization against your rubric and shows its reasons.
Your prompt
Act as a careful analyst who scores organizations against a written rubric and never invents facts. Format: Return JSON with: score (0-100), points per criterion, one sentence of evidence per criterion, and a list of facts that were missing. If a fact is missing, give 0 points for that criterion and say so.
Copy your prompt and test it on three made-up rows. Does it give 0 points when a fact is missing? Does it ever invent a fact? If it invents, tighten the prompt or drop the AI step.
Intent, not eligibility
This is the most important rule in the module.
A GTM score decides who to talk to. It must never decide who gets something: a loan, a job, housing, insurance, benefits, or services. Decisions like those fall under strict laws. One is the Fair Credit Reporting Act (FCRA), a federal law about reports used to judge people for credit, jobs, insurance, and housing.
The Temerarii engine enforces this in code. Its checker blocks outreach plans that touch restricted patterns such as credit, bankruptcy, payment history, collections, employment, rental history, criminal record, and eligibility decisions. A plan that touches any of them fails and does not run. The motto is short: signals, not scores; intent, not eligibility.
For you, that means a fit score uses facts about organizations and their public actions. It never uses a private person's money, health, record, race, religion, or family.
Watch for bias
Scoring can repeat old unfairness. In 2018, Reuters reported that Amazon dropped an experimental hiring tool after finding it rated resumes that mentioned the word "women's" lower. It had learned from years of past hiring data. Your rubric can do the same thing quietly. If "has worked with us before" earns points, neighborhoods you never served will keep scoring low. Read your top and bottom 20 rows by hand and ask who is missing.
Go deeper: AI, Minus the Hype: A Practical Guide to Solving Real Business Problems
Discussion
Ask. Answer. Build together.
Loading…