Library · Data Practice OS

AI readiness scorecard (workbook v2)

Template · Published · Source reviewed 2026-11-06

This page contains the Markdown guide. The workbook mentioned below remains in the original Data Practice OS bundle.

The file: docs/templates/assets/AI_Readiness_Scoring_v2.xlsx in the Data Practice OS folder. Copy it for each assessment. It implements AI readiness assessment, which is adopted from The source draft model (revised 4 Nov 2026). It is a proposed method, not an agreed decision.

Formulas only, with no macros. Blue-on-yellow cells are inputs and grey cells are formulas. The green row on the Controls sheet is an EXAMPLE that is never counted.

Sheets

Sheet	What you do there
How to use	The steps, the 0-4 scale, the bands, and two assumptions you agree before scoring: the minimum score for a mandatory control (default 3) and the maximum age of gate evidence (default 90 days). Neither default comes from the source draft
Scope	Name the AI use and business process; record the boundary, required context, acceptance thresholds, obligations, evidence plan, assessors and method version. Switch use profiles on (Y) or off. Enter inventory counts for coverage
Controls	54 controls seeded from the draft. Per control: mandatory, applicability and N/A rationale, owner, test, threshold, result, score (0-4 or Unknown), evidence status, link, date, assessor, effort
Capabilities	Calculated: each capability is the mean of its applicable controls
Scorecard	Calculated: each dimension is 5 × the mean of its applicable capabilities. Shows the total, the band, unknowns, coverage, confidence, and the readiness decision brought in from Gates
Gates	Per gate: applies, recorded result, test, evidence link, date and status, approver, approval date. Gives the readiness decision and an optional pilot (conditional) decision
Gap List	Calculated ranking: mandatory blockers first, then points gained per unit of effort

How to use it

Scope first. Fill in the Scope sheet with the business owner. Readiness applies to this scope and version only.
Agree the gates and the mandatory controls before scoring. Gate-linked controls default to mandatory. Change any of them now, not after you have seen the scores.
Score each control against its threshold, with evidence. Leave the score blank or enter Unknown when you can't tell. Never guess a number.
Set the evidence status honestly. Verified keeps the score. Partial caps it at 2. Unverified makes the control unknown.
N/A needs a rationale. An N/A without one is still counted, and the Scorecard flags it.
Record the gates. A gate passes only with a Pass result, Verified evidence dated within the limit, an approver and an approval date, and no linked mandatory control below threshold.
Calibrate. Two assessors score the first assessment independently, then reconcile.
Publish the score and the decision separately, with the unknown count beside the score.

What changed from v1

v1 problem	v2
Fixed 60/40 scan/interview weighting	Removed. Evidence is recorded per control, and any mix of sources supports the same control
One score per capability	One score per control; capabilities and dimensions roll up
"Production-ready" band	Renamed Established. Bands are maturity, not approval
No gates	A Gates sheet; the readiness decision can't be overridden by the score
Unscored counted silently as zero	Unknown is reported, kept in the denominator, and labelled a lower bound
No applicability	Applicable / N/A with a rationale; use profiles switch whole profiles off
No evidence fields	ID, requirement, stage, owner, test, threshold, result, evidence link and date, assessor, evidence status
Hard-coded "20 − scored"	Counts come from the data; coverage and confidence are reported
Gaps ranked by points per effort only	Mandatory blockers first, then points per effort

Verification

The workbook is generated by tools/build_readiness_workbook.py. tools/verify_readiness_workbook.py fills a copy with two scenarios, recalculates it in Excel, and compares every calculated column with an independent model of the rules. The scenarios are a mixed one (unknowns, partial evidence, N/A with and without a rationale, a whole capability N/A, a stale gate, a failed gate) and a perfect one. Last run on 6 Nov 2026: 0 mismatches and 0 formula errors in both. Re-run both scripts if you change the controls.