3.1.3. Prioritizing AI Initiatives: Scale, Pause, or Terminate
💡 First Principle: Resources are finite and opportunities are not, so every initiative competes with every other. Prioritization works when all initiatives are scored on the same criteria and re-scored as evidence arrives. A proof of concept answers "can it work?" Prioritization asks "is it worth it, can we sustain it, and does it matter strategically?" Different questions, different answers.
The exam guide names four prioritization criteria:
- Business value: size of the outcome (cost saved, revenue gained, risk reduced, experience improved) and how directly AI drives it.
- Feasibility: data availability and quality, technical complexity, integration effort, skills, regulatory constraints, and change-management difficulty.
- Sustainability: can it be run economically over time? Consider operating cost at scale (inference, monitoring, retraining), support model, dependency on scarce people, and environmental footprint of compute.
- Strategic alignment: does it advance stated corporate objectives, or is it interesting but peripheral?
A value-versus-feasibility matrix is the most common screening tool:
Use the quadrants as a starting point, then apply sustainability and alignment. A high-value, high-feasibility initiative that would cost more to run at scale than it saves fails the sustainability test. A quick win that advances no strategic objective may still be worth doing if cheap, because it builds skills and credibility (Phase 5).
Stage gates turn prioritization into decisions. At each review point, every initiative should exit with one of three verdicts:
| Verdict | When it applies | What happens next |
|---|---|---|
| Scale | Evidence of value against baseline, feasible at production scale, sustainable cost, strong alignment, ready owner | Fund production hardening, governance sign-off, rollout plan |
| Pause | Promising, but blocked by a fixable dependency (data access, regulatory clarity, missing sponsor, platform readiness) | Park with a named unblock condition and review date; redeploy the team |
| Terminate | Value not demonstrated, cost or risk exceeds benefit, strategic priority changed, or a better alternative exists | Stop, capture lessons and reusable assets, communicate openly |
Terminating a project that "works" technically is a sign of mature governance, not failure. A pilot that accurately extracts invoice data but saves only 20 hours a month across the company may be correct to stop. Celebrate what was learned and reuse components; do not let sunk cost drive continued funding.
⚠️ Exam Trap: A technically successful POC is not automatically scaled. When a scenario shows a working pilot with weak measured value, unsustainable operating cost, missing data at scale, or low strategic alignment, the right answer may be to pause or terminate. Look for options that apply consistent criteria and evidence rather than momentum.
Reflection Question: A generative AI pilot for drafting marketing copy is popular with users, but measured time savings are small and the monthly token cost at full rollout would exceed the savings. Which criteria fail, and is the right verdict scale, pause, or terminate? What evidence could change your answer?
