Sprint 2: Stakeholders & human-in-the-loop
Your initiative: _______________________ Group: __________
A working AI system isn’t a tool. It’s a division of labour between human judgement and machine output. This sprint you design that division, and the stakeholder plan around it. Interview Emma (Managing Director), Tom (Customer Service) and David (CFO).
1. Stakeholder plan
| Stakeholder | What they need / fear | How you’ll manage them | Power × Interest |
|---|---|---|---|
| Emma (Managing Director / sponsor) | High / High | ||
| David (CFO) | |||
| Tom (CS / frontline) | |||
| add others you met |
2. Human-in-the-loop checkpoints
For every point where AI output could be wrong and it matters, design a checkpoint.
| Checkpoint (where a human checks AI output) | Who has the expertise to actually evaluate it? | Does this checkpoint build that expertise, or just consume it? |
|---|---|---|
The evaluator’s advantage: experienced people spot the confident-but-wrong answer because they know what right looks like. Judgement is unevenly distributed, so staff your checkpoints with the right humans, not just any human.
The apprenticeship trap: if you automate all the grunt work, you stop producing the next generation of evaluators. Mark at least one checkpoint that you keep human deliberately, to build capability, not only for quality.
3. The one you’ll defend
Which single human-in-the-loop decision are you most confident is non-negotiable, and why? ______________________________________________
Leave Sprint 2 with: a stakeholder plan and a human-in-the-loop checkpoint design where each checkpoint names its evaluator and is marked builds-vs-consumes.