Finding and prioritising use cases
The best AI use cases come from understanding work. This module gives you a repeatable way to find, compare and choose them.
Learning objectives
- Find use cases by examining processes
- Describe use cases consistently
- Prioritise on value and feasibility
- Avoid common prioritisation traps
2.1 Where to look for use cases
Good use cases are found by looking at work, not at technology. Walk through processes with the people who do them and look for:
- Time spent reading, writing, searching or summarising
- Bottlenecks and queues — work waiting for a scarce expert
- High-volume decisions made from similar information each time
- Errors, rework and inconsistency between people
- Valuable data that nobody has time to analyse
2.2 Describe each use case the same way
A one-page use-case card makes ideas comparable:
| Problem | What's slow, costly or poor today — with numbers |
| Users & owner | Who uses it; which leader owns the outcome |
| AI's role | Draft, classify, extract, predict, answer, act — and where humans stay in the loop |
| Data | What's needed, where it lives, its quality and sensitivity |
| Value | Which lever, estimated size, how measured |
| Risks | Errors, bias, privacy, security, regulation, reputation |
2.3 Prioritise on value and feasibility
Score each use case on value (size of impact, strategic fit, urgency) and feasibility (data readiness, technical complexity, integration effort, risk, change required). Plot them:
- Quick wins (high value, high feasibility) — start here.
- Strategic bets (high value, lower feasibility) — invest in foundations first.
- Easy but small — fine for learning, don't over-invest.
- Avoid for now — low value, hard to do.
2.4 Risk is part of feasibility
Check early whether a use case touches high-risk territory — decisions about people's jobs, credit, health or access to services — where regulations such as the EU AI Act impose strict obligations. Higher risk doesn't mean "never", but it does mean more time, cost and governance, which belong in the feasibility score. See AI Governance.
2.5 Common traps
- The demo trap — a stunning prototype on clean examples that collapses on real data
- The pet-project trap — the most senior person's idea goes first, regardless of score
- The thousand-pilots trap — dozens of small experiments, none scaled
- The wrong-unit trap — automating a task when the real gain needs the whole process redesigned
List three AI use cases for your organisation. Score each 1–5 on value (impact, fit, urgency) and feasibility (data, complexity, integration, risk, change). Average each, place them on a value/feasibility grid, and decide which to start with.
See a worked example
Example: Support-reply drafting — value 4, feasibility 4 → quick win, start now. Demand forecasting — value 5, feasibility 2 (data scattered) → strategic bet, fix data first. AI-written product descriptions — value 2, feasibility 5 → easy but small, use for learning.
Key takeaways
- Look at work: reading, writing, queues, repetitive decisions.
- Use a one-page card for every idea.
- Score value and feasibility; start with quick wins.
- Regulatory and ethical risk belongs in feasibility.
- Beware demos, pet projects and endless pilots.