AI Score Calibration Guide
An unanchored 1-to-10 scale is a survey question wearing a spreadsheet. Work through all six dimensions one at a time, check every score against its full rubric, then generate a report of what you scored and why.
Listen to this briefing
Replace AI Vibes With Objective Evidence
How the calibration works
A score only means something when two people would give the same number for the same evidence. Each of the six dimensions below has ten written anchors, one per score. Read the anchors, pick the one that matches what you can show today, and the guidance in your report follows from the band that score falls in.
Every number has a written description, and three bands set the guidance
Strategic Fit, one of six dimensions
- 2Solves a problem nobody above your level is trackingBand reading: Below the line of sight of anyone who allocates budget
- 5Supported by one department head, not yet on a leadership scorecardBand reading: A sponsor exists, institutional weight does not yet
- 8Named in this year's leadership priorities or board deckBand reading: Attached to something leadership already measures
The same three bands on all six dimensions
Strategic Fit · Data Readiness · Ease of Implementation · Expected ROI · Speed to Value · Autonomy Readiness
Pick the anchor that matches the evidence you can show today. The guidance in your report follows from the band the score falls in.
Work through all six dimensions one at a time. For each one, move the slider to where the evidence sits, or click the rubric row that describes your situation, then move on. At the end, get a report of what you scored and why.
0 of 6 dimensions scored
Strategic Fit
How well the use case aligns with a priority your leadership is already measuring.
Not scored yet. Move the slider or pick the rubric row that matches your evidence.
| Score | What it looks like |
|---|---|
| 1 | Nobody outside your own team has ever raised this problem |
| 2 | Solves a problem nobody above your level is tracking |
| 3 | One peer team feels the same pain, but it has never been raised upward |
| 4 | A manager has mentioned it in conversation, but it exists nowhere in writing |
| 5 | Supported by one department head, not yet on a leadership scorecard |
| 6 | Referenced in a departmental plan, but not yet in a company-wide one |
| 7 | Two or more department heads have independently raised the same priority |
| 8 | Named in this year's leadership priorities or board deck |
| 9 | Tied to a metric the CEO or board reviews on a recurring cadence |
| 10 | The problem this solves is the stated reason the current budget cycle exists |
Score this dimension to continue.
Data Readiness
Whether the data this use case needs exists, is accessible, is clean, and is sufficient in volume.
Not scored yet. Move the slider or pick the rubric row that matches your evidence.
| Score | What it looks like |
|---|---|
| 1 | Nobody has checked whether this data exists |
| 2 | The data does not exist yet, or lives in someone's inbox |
| 3 | The data exists somewhere, but no one has located or accessed it yet |
| 4 | Data has been located, but it is scattered across systems with no clear owner |
| 5 | Data exists but needs real cleanup, consolidation, or access work first |
| 6 | Data is consolidated in one place, but still needs a meaningful cleanup pass |
| 7 | Data is clean and accessible, but not yet connected to any reporting |
| 8 | Data is already accessible, structured, and used for reporting today |
| 9 | Data is accessible in real time and already feeds an operational system |
| 10 | Data is production grade: monitored, governed, and already trusted for decisions today |
Score this dimension to continue.
Ease of Implementation
Technical complexity, integration requirements, change management load, and vendor dependency.
Not scored yet. Move the slider or pick the rubric row that matches your evidence.
| Score | What it looks like |
|---|---|
| 1 | Custom model work, several system integrations, and a full redesign of how the team operates |
| 2 | Custom model work, multiple system integrations, heavy change management |
| 3 | Custom model work, but only one system integration required |
| 4 | No custom model work, but two or more integrations and real change management |
| 5 | One clear integration and a moderate change to how a team works |
| 6 | One integration, light retraining, most of the workflow stays the same |
| 7 | Off-the-shelf tool, but still needs one non-trivial integration |
| 8 | Largely off-the-shelf, one system, minimal retraining needed |
| 9 | Fully off-the-shelf, single sign-on into an existing system, no retraining beyond a short walkthrough |
| 10 | Already installed in a tool the team uses today. Turning it on is the entire implementation |
Score this dimension to continue.
Expected ROI
Direct financial return plus indirect value: risk reduced, quality improved, capacity freed.
Not scored yet. Move the slider or pick the rubric row that matches your evidence.
| Score | What it looks like |
|---|---|
| 1 | Nobody can describe what the value would even look like if this worked |
| 2 | Value is theoretical, and nobody has tried to size it |
| 3 | Value is theoretical, but at least one comparable case exists elsewhere |
| 4 | A rough estimate exists, built from a single untested assumption |
| 5 | A rough estimate exists but rests on assumptions nobody has tested |
| 6 | A rough estimate exists, and one key assumption has been tested or benchmarked |
| 7 | A calculated estimate exists, but it depends on an adoption rate not yet observed |
| 8 | A specific, defensible number, ideally run through an ROI calculation |
| 9 | A calculated number, validated against a pilot or a comparable deployment |
| 10 | A number already being realised in production, tracked against a baseline |
Score this dimension to continue.
Speed to Value
How quickly the use case can demonstrate a measurable result, distinct from how hard it is to build.
Not scored yet. Move the slider or pick the rubric row that matches your evidence.
| Score | What it looks like |
|---|---|
| 1 | Over nine months before any measurable result, if one is possible at all |
| 2 | Over six months before any measurable result is visible |
| 3 | Four to six months before any measurable result |
| 4 | Twelve to sixteen weeks before a measurable result |
| 5 | A visible result in eight to twelve weeks |
| 6 | A visible result in six to eight weeks |
| 7 | A visible result in four to six weeks |
| 8 | A measurable result within four weeks of go-live |
| 9 | A measurable result within two weeks of go-live |
| 10 | A measurable result is visible on day one, or within days of go-live |
Score this dimension to continue.
Autonomy Readiness
Whether governance controls are proportionate to the level of autonomous action this use case will exercise.
Not scored yet. Move the slider or pick the rubric row that matches your evidence.
| Score | What it looks like |
|---|---|
| 1 | It would act on real systems today with no logging, no approval step, and no way to even notice a bad action afterward |
| 2 | It could act on real systems today with no logging, no approval step, and no way to catch a bad action fast |
| 3 | Logging exists, but nobody reviews it, and there is still no approval step before action |
| 4 | Logging exists and someone reviews it periodically, but not before each action |
| 5 | Logging exists; a person reviews output before anything reaches a customer or a system of record |
| 6 | Logging and a review step exist, and the review step has already caught at least one real issue |
| 7 | An approval point exists and has been tested against a deliberately induced failure |
| 8 | Guardrails, logging, and an approval point are already built and tested for this exact level of autonomy |
| 9 | Guardrails, logging, and an approval point are tested and have already caught a real incident before it reached a customer |
| 10 | A full audit trail, a tested approval point, and a track record across multiple production incidents caught before impact |
Score this dimension to continue.
Every score you set, next to the evidence that justifies it and the professional guidance for a score at that level. Revisit any dimension before generating your report.
Next: open the printable report to keep a record of these scores, then score a real initiative with the same six dimensions. Revisit any dimension from the dock above.
I built this after the same argument kept happening in different rooms: one person calls a project a 7, another calls it a 4, and neither is lying. They just never agreed on what a 7 requires. This page is the answer to 'what does that number mean,' dimension by dimension, so a score is something you can defend in a room.
Last updated 2 Oct 2026
More in Reference
- The Discipline GapWhy do most AI transformations stall before they scale?
- The Agentic AI Pattern CatalogueWhich design pattern fits the agent we are building?
