|
You&AI
Operating Playbook · Highest Priority
The AI Prototype FrameworkOne repeatable path from idea to production — how we run prototypes, decide what is worth building, evaluate the outcome honestly, measure success, and safely move the winners into live service. Read Part One to understand the framework. Use Part Two — the workbook — for every new AI project: copy it, fill it out at each stage, and keep it as the project's record. OWNER AI & Data Platform Team APPLIES TO All new AI initiatives VERSION 1.0 · July 2026 STATUS Adopt & follow Why this exists
Most AI dies at the demo.Excitement gets an idea to a working prototype. Then nobody knows what happens next — no evaluation, no route to production, no owner. The framework replaces that dead-end with a lifecycle. Every project moves through the same stages, and every stage has a decision attached to it. Where most teams are
Idea
↓ Excitement ↓ Prototype ↓ Demo ↓ Nobody knows what happens next Where this takes us
Idea
↓ Assessment ↓ Prototype ↓ Evaluation ↓ Governance ↓ Production ↓ Monitoring ↓ Retirement 01The AI Project LifecycleSeven stages, each with a clear question, a clear output, and a clear gate. A project cannot skip a gate. If it fails a gate it is either reworked or stopped — stopping early is a success, not a failure. Stage
{{ s.n }}
{{ s.title }}
{{ s.gate }}
{{ s.desc }} 02Risk Categories & Approval PathsBefore a prototype starts, classify it. The category sets the approval path, the security bar, and how closely it is monitored. This is practical governance, not bureaucracy — a low-risk note-taker should not carry the same weight as an automated eligibility decision. {{ r.name }}
{{ r.approval }}
Examples
{{ r.examples }}
Each category defines three things: an approval path (who signs off), security requirements (controls the build must include), and monitoring requirements (what is watched once live). Set these once, then reuse them for every project in that band.
03How We Measure SuccessMost AI projects have no measurable outcome — so nobody can say whether they worked. Every prototype must define its KPIs before it starts, drawn from these four families. Pick the two or three that matter for your use case and set a target for each. {{ k.name }}
{{ k.metrics }}
04Taking Prototypes to ProductionA good demo is not a production system. Before anything goes live it is scored against six weighted areas. The score is objective, comparable across projects, and sets a clear bar: no project ships below the minimum.
Minimum to ship: 70 / 100, with no single area scoring zero. Score 55–69 → conditional, fix named gaps and re-score. Below 55 → not ready; return to prototype or stop.
Part Two
The Project WorkbookComplete one workbook per AI project. Fill each stage as you reach it and record the gate decision before moving on. On paper: write in the boxes. On screen: type into the fields, then Print / Save PDF. Project record
Project name Business owner Technical lead Registry ref Date started Risk category (circle one)
Low
Medium
High
Restricted
STAGE 0
Idea IntakeCapture the idea in plain language. Answer every question — vague intake is where projects go wrong. {{ f }} {{ gate0 }}
STAGE 1
Feasibility AssessmentRate each dimension and add a note. If any is red, resolve it before prototyping — don't build on top of an unknown. {{ gate1 }}
STAGE 2
Prototype CharterA prototype is time-boxed and cheap by design. Agree the limits and success criteria up front — then build. Time box Max 4 weeks Budget Max £500 Criteria set Before build Start & end date Estimated cost Success criteria — what "good" looks like at the end (be specific & measurable) KPIs to be captured (from Section 03 — set a target for each) {{ gate2 }}
STAGE 3
Evaluation ScorecardScore the prototype 1–5 against each area and record the evidence. This decides whether it advances, is reworked, or stops. {{ gate3 }}
STAGE 4
Production ReadinessWork the checklist, then score the weighted assessment. Both must pass before go-live. Readiness checklist
{{ c }}
Weighted readiness score
{{ gate4 }}
STAGE 5
Production Sign-offThe go-live decision, on the record. Named people, named date. Business owner sign-off Security / governance sign-off Go-live date Named production owner
STAGE 6
Continuous Review & MonitoringLive is not done. Review quarterly against the KPIs; decide to keep, improve, or retire. Log each review. |