◈ Delivery Hub for AI programmes
Launch readiness

Answer honestly.
That is the whole method.

Twenty-four questions across six weighted dimensions. Run it with your engineering, data, risk and business leads together — the disagreement between their answers is more valuable than the score.

Used as the heading on your result. Leave blank and it scores anyway.
◎
Value & Problem Fit
weight ×1.3
The business outcome is quantified with a baseline we can measure today.
We have evidence that AI (not rules or search) is the right tool for this problem.
A named business owner accepts accountability for adoption, not just delivery.
Unit economics work at projected volume, including inference cost.
⛁
Data Foundations
weight ×1.2
Training/grounding data has documented provenance and licence to use.
Sensitive data is classified, minimised and access-controlled end to end.
We monitor drift in input distributions, not just model outputs.
A golden evaluation set exists and is version controlled.
⟡
Evaluation & Quality
weight ×1.4
Offline eval runs automatically on every prompt or model change.
We have agreed quality thresholds that block release when breached.
Online quality is measured with real user signals, not vibes.
Regression risk from model/provider upgrades is tested and rehearsed.
⚖
Responsible AI & Risk
weight ×1.35
A documented impact assessment covers fairness, safety and misuse.
Red-teaming covered prompt injection, data exfiltration and jailbreaks.
Users are told they are interacting with AI and how to challenge output.
Human oversight is designed for the consequence level of the decision.
⚙
Engineering & Operations
weight ×1.1
The whole system (prompts, config, eval sets) is version controlled.
We can roll back a model or prompt change in under 15 minutes.
Latency, cost and error budgets are defined with alerting.
On-call ownership covers AI-specific failure modes.
⇗
Adoption & Change
weight ×1.25
Target users were involved in shaping the workflow, not just testing it.
Enablement, training and comms are planned and resourced.
Support teams can diagnose and explain AI behaviour to users.
We have an adoption target and a plan if we miss it.
Unanswered questions score as “No”.

Delivery Hub for AI Programmes — © 2026 Tariq Khan. All rights reserved. Developed by Tariq Khan.