Use-case scoring rubric
You have several ideas and need to rank them honestly.
Five criteria, scored 1 to 5, with a few knock-outs that stop a candidate regardless of score. Score each candidate against its AI Opportunity Map, not against enthusiasm. The score informs the decision. The sponsor still makes it.
Before you score
- Every candidate needs sections 01, 02 and 04 of its Opportunity Map filled in.
- Score as a group of two or three, independently first, then compare. Big disagreements are the useful part: they point at an assumption someone hasn't said out loud.
- Write one line of evidence for every score. A score without a reason doesn't count.
Knock-outs
Stop the candidate, or change it until none apply.
- It makes or changes a decision about a person (credit, hiring, safety, eligibility) with no human review.
- It needs data the organization isn't allowed to use this way.
- Nobody will own it after the pilot.
- A wrong output could cause physical harm, and there is no way to catch it before it does.
- Success can't be measured, even roughly.
The criteria
| Criterion | 1 | 3 | 5 | Weight |
|---|---|---|---|---|
| Value. Time, money or risk removed, at today's volume | Minor annoyance, a few hours a month | A few hours a week, or a visible error source | A day or more a week, or a material risk or revenue effect | x3 |
| Feasibility. Data, tools and skills available now | Data doesn't exist or can't be reached | Data exists but needs cleanup or an integration | Data and tools are in reach; similar solutions exist | x2 |
| Risk. What a wrong output costs, and who sees it | Customer-facing or regulated harm, hard to catch | Internal, caught by a normal review | Low stakes, easy to catch and correct | x2 |
| Reversibility. How easily you can undo it | Hard to undo: contracts, migrations, retraining | Undoable with some effort | Switch it off and nothing is lost | x1 |
| Time to learn. How soon you'd know if it works | Months | Weeks | Days | x1 |
Weighted total = (Value x 3) + (Feasibility x 2) + (Risk x 2) + Reversibility + Time to learn. Maximum 45.
Score sheet
| Candidate | Knock-out? | Value | Feasibility | Risk | Reversibility | Time to learn | Weighted total | Evidence (one line each) |
|---|---|---|---|---|---|---|---|---|
Reading the result
- 36 or more: a strong candidate. Prototype it.
- 25 to 35: worth it if the Value score is 4 or 5. Otherwise look for the simpler version.
- Below 25: park it and write down why. It may come back when the data or the tools change.
- Ties: pick the one that is easier to reverse and faster to learn from.
The rubric is a tool for an honest conversation. If the result feels wrong, find out which score is wrong before overriding it.
Use and reuse. © 2026 Maple Spark Labs Inc. You may use and adapt this material within your organization. Keep this notice. Please ask before republishing, redistributing or selling it.