AI Strategy and Transformation: Templates
Four working templates: the process inventory row, the candidate score sheet, the benefit baseline record and the refusal-list entry.
Markdown. No sign-up, no email.
Four templates. Copy them, delete the guidance in brackets, keep the structure. They are deliberately short: a template long enough to feel thorough is a template nobody completes.
1. Process inventory row#
One row per recurring process. This is the artefact everything else depends on.
| Field | Guidance |
|---|---|
| Process | [Name it as the people doing it name it, not as the org chart does] |
| Function | [Who owns it] |
| Trigger | [Schedule, or the event that starts it] |
| Instances per year | [Count from system logs. If you cannot count it, you cannot rank it] |
| Minutes per instance | [Observed or logged, never asked. Self-reported figures understate routine work] |
| Annual hours | [Instances multiplied by minutes, divided by 60] |
| People involved | [Distinct roles that touch it] |
| Output reversible? | [Minutes, days, or not at all] |
| Failure visible? | [Immediately, at review, or only when a customer finds it] |
| Regulated? | [Which regulation, or none] |
Rank by annual hours and publish the top 20. That list is the candidate pool for the year.
2. Candidate score sheet#
| Axis | Score 1 to 5 | Evidence |
|---|---|---|
| Volume | [Instances per year] | |
| Structure | [Are inputs consistent and machine-readable?] | |
| Reversibility | [How long to undo a wrong output?] | |
| Tolerance | [What does being wrong once cost?] |
Decision rule: below 3 on reversibility or tolerance, the process does not go to automation whatever its volume. It may go to augmentation, where a person still decides.
| Recommendation | [Automate / Augment / Refuse] |
| Constraining axis | [The lowest score, and why it is what it is] |
| Decided by | [Name. Not a committee] |
| Date |
3. Benefit baseline record#
Complete this before anything changes. A baseline written afterwards is an estimate.
| Field | Entry |
|---|---|
| Process | |
| Measure | [One measure, defined in a sentence, owned by Data and Analytics] |
| Period observed | [Four weeks minimum, or one full cycle] |
| Baseline value | [Median, not mean, if the range is wide] |
| Range | [Minimum to maximum. A wide range is itself the finding] |
| Forecast after change | |
| Forecast made by | |
| Review dates | [90 days and 12 months] |
| Frozen on | [Date. It is not adjusted afterwards] |
4. Refusal-list entry#
| Field | Entry |
|---|---|
| Process | |
| Why refused | [Irreversible outcome / regulated decision / accountability must sit with a person / failure invisible until expensive] |
| What would change this | [Be specific. "Better models" is not a condition; "a reversal path that works within one working day" is] |
| Decided by | |
| Date | |
| Next review | [Quarterly] |
The last two rows are what stop the refusal list becoming a permanent no that nobody revisits. A refusal without a review date is a decision that has quietly become a policy.
Using these with the department's other pages#
The charter states what the function owns, the SOPs say when each of these is run, the KPIs define what the baseline feeds, and the workflows name who receives each output. Filling a template in without the SOP that governs it produces a tidy document and no decision.