Business case and benefits
- Business owner
- A · Accountable
- Programme manager
- R · Responsible
- Product owner
- R · Responsible
- Solutions architect
- C · Consulted
- Technical PM
- C · Consulted
- Data protection officer
- C · Consulted
- Service owner
- C · Consulted
The course gave you the frameworks. This is what they look like on a real programme, with real systems, real data and a board waiting for results.
Phases and gates
Each gate is a short list of things that must be true. If they aren't, you don't pass, however good the demo looked.
Frame the problem, measure today, find the data, name the owner, and test the riskiest assumption on real data.
Gate · must be true to passBaseline agreed, data access confirmed, DPIA started, compliance involved.
One process, in shadow mode: the AI recommends, people still decide, and the two are compared.
Gate · must be true to passAccuracy on your own cases within agreed limits; no customer group worse off; compliance sign-off.
Monitoring, support, rollback and training in place before the switch, not after.
Gate · must be true to passRun cost and quality stable for a full month.
More cases and more teams, reusing the same pattern and the same gates.
The business case
Most of this is the same as any business case. The last three items are where AI cases usually fall short.
Roles
Adapt the roles to your organisation. The tagged columns are the roles I can fill.
| Activity | Business owner | Programme manager | Product owner | Solutions architectI can fill this | Technical PMI can fill this | EngineersI can fill this | Data protection officer | SecurityI can fill this | Service owner |
|---|---|---|---|---|---|---|---|---|---|
| Business case and benefits | Accountable | Responsible | Responsible | Consulted | Consulted | Not involved | Consulted | Not involved | Consulted |
| Target architecture | Informed | Consulted | Consulted | Accountable and Responsible | Consulted | Consulted | Consulted | Consulted | Consulted |
| DPIA and regulatory review | Accountable | Consulted | Consulted | Consulted | Informed | Not involved | Responsible | Consulted | Informed |
| Build and integration | Informed | Consulted | Consulted | Accountable | Responsible | Responsible | Informed | Consulted | Consulted |
| Evaluation on your own cases | Informed | Informed | Accountable | Responsible | Consulted | Responsible | Consulted | Not involved | Consulted |
| Go-live decision | Accountable | Responsible | Consulted | Consulted | Responsible | Informed | Consulted | Consulted | Consulted |
| Run, monitor and improve | Informed | Informed | Consulted | Consulted | Consulted | Responsible | Consulted | Consulted | Accountable |
R ResponsibleA AccountableC ConsultedI Informed· not involved
Sound familiar?
None of these are in the textbook. All of them are in most programmes.
The data is in six systems, and two of them are spreadsheets.
Courses assume clean, available data. Real programmes spend their first month finding it, and their second getting permission to use it.
The pilot worked. Production is a different problem.
Demos skip the API layer, the queue, the monitoring and the support rota. That's where most of the effort actually goes.
Compliance joined in month five.
The DPIA and your regulator's rules shape the design. Bring them in at the first gate, not the last.
Nobody owns it after go-live.
Prompts, evaluation sets and models drift. Without a named owner, quality falls quietly until someone complains.
For coursework
Every playbook uses the same sections, so you can compare them. This is where each part of a typical course assignment lives.
Every playbook's The business problem section.
Every playbook's How success is measured section.
Every playbook's UK regulation and controls section.
Every playbook's The solution, and components side by side section.
Every playbook's Process and people section.
Every playbook's Delivery plan section.
Every playbook's Roles and RACI section.
Every playbook's Where this goes wrong section.
For students
Use these playbooks as worked examples. Each one has the business case, measures of success, risks and RACI your coursework asks for. I hold a Leadership with AI certificate myself, so I know what the courses cover, and where real projects differ.
Delivering an AI programme and want someone to test the phases, gates and roles before you commit? Let's talk it through.