What gets better than today, and do people use it?
- 1No better than today, or rarely used
- 3Some measured gain; some people use it
- 5Large measured gain; used every week
0Scale
7Shrink
2Pause
0Kill
Nine pilots, nine proposed calls. None is ready to scale yet. Seven run on a narrower scope while evidence is gathered, and two are on hold until a named blocker clears. First deadline: 23 October.
1
Score four questions
Each question is scored 0 to 5 from evidence. The evidence and its source sit on the pilot's own tab, so anyone can check a score.Is the full cost worth paying, including checking and rework?
- 1Costs more than it saves
- 3Roughly breaks even
- 5Clearly cheaper per task, all costs counted
Is the output good enough, checked by someone independent?
- 1Below the standard
- 3Meets the standard on some tasks
- 5Meets the standard on an independent check
If it goes wrong, is the harm small, and can someone stop it?
- 1Serious harm possible; nobody can stop it
- 3Controls exist but are untested
- 5Tested controls; a named owner can stop it
2
Add two totals and read the call
The two totals place a pilot on the map, and the zone it lands in is the rule-based call. The proposed call is the decision people make, and the dot shows it in its colour.
Value for money=Value+Cost
Readiness to scale=Quality+Risk
Each runs 0 to 10. 6 or more is high.
Set by the zone
Scale
Value for money 6+ and readiness 6+Expand only the tasks and users the evidence supports, with a review date and a stop rule.
Moves to the next stage or the next tier.Shrink
Value for money 6+, readiness 5 or lessCut back tasks, users or authority. Run a bounded test of what is still uncertain.
Stays at its stage with a narrower scope or tier.Kill
Value for money 5 or lessEnd this proposal. Name who closes it and keep what was learned.
Goes to Retire.Overrides
Pause
A named blocker must clear firstHold use until one blocker clears: a legal or safeguarding gate, an unsigned DPIA, missing evidence, or no capacity. Name an owner and a restart date, then re-score.
Holds at its stage. Pause is a hold, so it has no zone.Red card
Quality or Risk at 0 or 1No Scale, whatever the total.
Rule and judgement. The rule-based call is a formula. The proposed call is a judgement; when the two differ, the pilot says why.
3
Where each pilot stands
Each card shows the four scores, the two totals, the call and when it is next due. The scope, next action and stop rule for each pilot are kept in the school's internal decision sheet.P01Production, six staff
AI Marking app
Shrinkmatches the rule
V3
C4
Q2
R3
Value for money 7Readiness 5
- Next due
- 27 Nov 2026
- Review
- 12 Feb 2027 · Academic Board
- Sponsor
- Deputy Headteacher (Academic)
P02Live
Staff dashboard, LEA assistant and Governors' report
Shrinkmatches the rule
V3
C3
Q2
R1
Value for money 6Readiness 3Red card
- Next due
- 11 Dec 2026
- Review
- 29 Jan 2027 · AI Steering Group
- Sponsor
- Deputy Headteacher (Pastoral)
P03Pilot, live since September 2026
UCAS reference app
Rule says KillPause
V2
C0
Q1
R1
Value for money 2Readiness 2Red card
Why not the rule: The rule reads Kill because evidence has not been collected yet. The pilot already has a defined test (its first 30 references), so it holds until a named blocker clears, then is re-scored.
- Next due
- 23 Oct 2026
- Review
- 27 Nov 2026 · AI Steering Group
- Sponsor
- Assistant Headteacher (Pastoral)
P04Idea
Trips optimisation
Rule says KillPause
V0
C0
Q0
R0
Value for money 0Readiness 0Red card
Why not the rule: No pilot has run, so the rule reads Kill. Pause until a pilot card exists with a scope, a measure and a stop rule.
- Next due
- 27 Nov 2026
- Review
- 29 Jan 2027 · AI Steering Group
- Sponsor
- Business Director
P05Built, not yet in use
Instructional coaching: EdTech and AI (TPACK and SAMR)
Rule says KillShrink
V1
C4
Q2
R4
Value for money 5Readiness 6
Why not the rule: The rule reads Kill because there is no usage evidence yet. Cost and risk are low, so a bounded test with two departments is cheaper than stopping.
- Next due
- 11 Dec 2026
- Review
- 12 Feb 2027 · Academic Board
- Sponsor
- Deputy Headteacher (Academic)
P06Baseline done
4D AI framework for staff
Rule says KillShrink
V2
C3
Q2
R4
Value for money 5Readiness 6
Why not the rule: The rule reads Kill on value for money because delivery has not started. Run the shared core first and measure it before wider rollout.
- Next due
- 11 Dec 2026
- Review
- 26 Mar 2027 · Academic Board
- Sponsor
- Headteacher
P07Sandbox
Staff-built (vibe-coded) apps
Rule says KillShrink
V2
C3
Q1
R1
Value for money 5Readiness 2Red card
Why not the rule: The rule reads Kill, but the value of staff building is part of the strategy. Keep building on made-up data inside the school's sandbox while evidence is gathered.
- Next due
- 27 Nov 2026
- Review
- 29 Jan 2027 · AI Steering Group
- Sponsor
- Business Director
P08Running, Autumn 1
Student AI tools and 4D framework
Shrinkmatches the rule
V2
C4
Q2
R3
Value for money 6Readiness 5
- Next due
- 27 Nov 2026
- Review
- 12 Feb 2027 · Academic Board
- Sponsor
- Assistant Headteacher (Student Engagement)
P09Built, not yet released to staff
A* Toolkit: EdTech tools for staff, mapped to SAMR, TPACK and Rosenshine
Rule says KillShrink
V1
C4
Q2
R3
Value for money 5Readiness 5
Why not the rule: The rule reads Kill because staff have not used it yet. It costs nothing to run and keeps pupil data out, so a bounded trial in two departments is the cheaper way to find out.
- Next due
- 11 Dec 2026
- Review
- 12 Feb 2027 · Academic Board
- Sponsor
- Deputy Headteacher (Academic)
No pilots have this call.
Not yet a pilot: F1 Read-only data agent. A future deliverable that needs the AI-safe data layer first.
4
What is due, and when
Every next-action deadline and review date in date order. Dates are proposals until they are matched to real meeting dates.5
When to re-score
Pilots move through the eight Idea-to-Production stages. They are re-scored at four gates (gold), where new evidence arrives, and the call at each gate decides the move.2SandboxExplore
Re-score: Quick scoreBefore filing the Build request
Scale or Shrink: file the Build form at the level and tier the scores support. Pause: stay in Sandbox. Kill: stop; nothing to close.
Digital Lead with the SLT sponsor
3BuildBuild and rehearse
No re-score
Level and tier assigned; the scores carry forward.
Digital Lead
4StagingBuild and rehearse
Re-score: Quality and RiskBefore leaving Staging
A red card blocks the Pilot. Otherwise the pilot goes ahead.
DPO and IT/Network Manager
No pilots here
5PilotProduction
Re-score: Full scoreAt the end, against a rule written in advance
Scale: go to Production at the tier the evidence supports. Shrink: re-run narrower. Pause: hold until the blocker clears. Kill: Retire.
Digital Lead and the sponsoring head of department; SLT sponsor signs
6ProductionProduction
No re-scoreUses the Pilot score
Signatories sign by level and tier (Governance tab).
Signatories by level and tier
7LiveProduction
Re-score: TermlyEach termly review, and after any incident or model change
Scale: move up a tier, back through Staging. Shrink: fewer users or less scope. Pause: suspend use. Kill: Retire.
SLT sponsor at the termly governance meeting
8RetireProduction
No re-score
Close it down and keep what was learned.
Digital Lead; DPO confirms deletion
No pilots here
Re-score outside the schedule when a user group, dataset or integration goes beyond the pilot's level or tier; the AI model behind it changes; an incident, complaint or safeguarding concern arises; a termly review fails; or the SLT sponsor asks for it.
6
Who owns it and who decides
Each pilot has one SLT sponsor who owns it and signs it off. Its forum takes the call at each review.SLT
Money, priorities and college-wide sign-off. The Principal signs anything used college-wide.
Academic Board
Pedagogy and assessment: AI marking policy, coaching, the 4D frameworks and student AI tools. Led by the Deputy Headteacher Academic.
AI Steering Group
Governance gates: DPIAs, risk tiers, security and the app register. Digital Lead, DPO, IT/Network Manager and DSL.
- One sponsor per pilot owns it and signs it off.
- Re-score at four gates. Scale goes on, Shrink narrows, Pause holds, Kill retires.
- Other sign-offs follow level and tier: DPO and IT/Network Manager from Level 2 or T2, the DHT Academic from Level 3 or T2, the Headteacher for Level 4 and all-staff or all-student use.
- Stakeholders hear first. Nothing widens before the people it affects are told.
- Termly report. Each sponsor reports KPIs, the latest scores and the call. Decisions go in the Decision log.