Gains:
- Ability to collect artificial intelligence usage areas from a wide pool and clarify them with the 'problem/who/data/result/measurement' template
- Ability to score usage areas on the value and applicability axes and place them in a four-zone matrix
- Ability to focus on the wave of the first 90 days by creating momentum with quick gains and planning big bets
The success or failure of an AI strategy often boils down to a single question: “What are we saying yes to and what are we saying no to?” Dozens of ideas circulate in most institutions; but the resource is limited. In this unit, you will learn how to systematically identify and prioritize AI use cases (a specific scenario in which AI is applied to solve a concrete business problem). The aim is to choose "the one that will produce the most value" rather than the "most talked about" one.
How to collect usage areas?
First create a pool. Good use cases come from three sources: pain points (employee time-consuming, tedious, error-prone work), opportunities (where it can drive revenue or customer experience), and benchmarking (what works in a similar industry). Keep the pool wide; elimination comes later.
A good use case description includes the following elements: what business problem, for whom, what data, expected outcome, and how to measure it. “AI in customer service” is not a use case; "Shortening the forwarding time by automatically classifying incoming e-mails according to their subject and directing them to the right team" is a usage area.
Tip: Good use cases tend to be "high volume + repetitive + text/data heavy + reasonable fault tolerance" jobs. The more of these four attributes a job has, the higher its AI value.
Prioritization: value/applicability matrix
The most common and practical tool is a two-axis matrix:
- Value axis: How much business impact does this create? (Revenue, cost, risk reduction, customer experience)
- Feasibility axis: How easily can we do this? (Is the data ready, technical difficulty, risk, burden of change)
Region
Value
Applicability
What to do?
quick gain
high
high
Start now
big bet
high
low
Plan, set up prerequisites
filler
low
high
Do it if you have time.
trap
low
low
Don't/postpone
Strategy: Build momentum and confidence with quick wins, nurture by planning big bets, avoid pitfalls.
Step by step: from idea to priority
1. Tidy up the pool. A broad list of use cases emerge from the departments (30-50 ideas are normal).
2. Describe using standard template. Put each idea in the “problem / who / data / outcome / measurement” format. The blurry ones become clear or are eliminated.
3. Score on two axes. Score each idea for value (1-5) and feasibility (1-5). Be evidence-based.
4. Insert into matrix. Distribute to four regions; highlights quick gains.
5. Select the first wave. Start with 2-4 quick wins + 1 big bet discovery. Not much more.
three mini cases
Case 1 — Tempting but impractical. An e-commerce company fell in love with the idea of a “fully personalized storefront with AI” (high value). But the data was fragmented, integration would take 9 months (low feasibility). The matrix showed this as "big bet". The company first started the rapid acquisition of "generating product description texts with artificial intelligence"; Content production time dropped by 70% in 6 weeks, savings financed the big bet.
Case 2 — Elimination of the template. An insurance company collected 41 ideas. When fitted into the standard template, it was seen that 14 of them actually described the same problem, while 9 of them did not have a measurable result. The pool has shrunk from 41 to 12 actual use areas; Prioritization became much clearer.
Case 3 — Escape from the trap. One manufacturer enthusiastically proposed the idea of “summarizing executive meeting notes with AI.” Low value in the matrix (few people, little time saved) and medium feasibility output: backfill/trap limit. The resource was instead directed to the high-value acquisition of “analyzing production line fault records and finding recurring causes”; Annual downtime decreased by 11%.
Four copyable templates
1) Defining the area of use:
Your role: AI business analyst. Describe the following business problem as a use case: [problem]. Template: (1) business problem to be solved, (2) for whom, (3) data required, (4) expected measurable outcome, (5) key risk. Ask me back for any vague points as questions.
2) Pool production:
Suggest 10 AI use cases for [department] in [industry]. Provide a one-sentence problem description and estimated value type (revenue, cost, risk, experience) for each. Focus on high-volume, repetitive work.
3) Value/applicability scoring:
Score the following use cases on a score of 1-5 on two axes: value (business impact) and feasibility (data preparation, technical difficulty, risk). Write a one-sentence justification for each point and place it in four areas (quick win, big bet, filler, trap). Fields: [list]
4) First wave selection:
Select the wave to start in the first 90 days from this prioritized list: up to 3 quick wins and 1 big bet discovery. For each choice, write down why now, what metric it will be measured by, and the first milestone. List: [text]
Weak prompt / Strong prompt
Weak: “Give AI ideas for my company.”
Result: A generic list without context; its value and applicability are uncertain, no choice can be made.
Güçlü: "I am an accounting services firm with 150 people. Our most time-consuming jobs are invoice entry, reconciliation, and responding to customer emails. Suggest AI use cases for these three areas, describe each with the 'problem/who/data/result/risk' template and place it in the value-applicability matrix. Write with the rationale what you recommend I start with in the first 90 days."
The result: a contextualised, templated, prioritized and actionable choice.
Common mistakes
- Choosing the most talked about one. The most popular idea is not the most valuable idea; The matrix fixes the emotion.
- Blurred usage area. Descriptions like “X with AI” cannot be measured; standard template is required.
- Ignoring practicality. Blindly diving into high-value but data/integration-impossible ideas means months of worthless labor.
- Too many starting at once. Limiting the first wave to 3-4 provides focus and learning.
- Bypassing fault tolerance. It is dangerous to mistake areas with low error tolerance (e.g. direct medical/legal decision making) for quick wins.
Caution: AI may suggest a use case list and scoring; But it is the manager who decides whether the value and applicability scores comply with the reality of the institution and which wave to choose. Additionally, although a use case may be technically "doable", it may be "not done" from an ethical or compliance perspective; We will cover this in Units 8 and 9.
In summary
Usage prioritization is the art of directing limited resources to the highest value. First collect a large pool, clarify each idea with the “problem/who/data/result/measurement” template, then score it on the value and feasibility axes and place it in the matrix. Build momentum with quick wins, nurture it by planning big bets, avoid low value traps. Limit the first wave to 3-4 areas. In the next unit, we will see how to prove with ROI and metrics whether the use cases you have chosen are actually producing value.
Application task
Collect 10-15 AI use case ideas from your organization. Fit each into the standard template; clarify or eliminate blurry ones. Score the remainder 1-5 on the value and feasibility axes and place them in a matrix. Choose the wave of the first 90 days (up to 3 quick wins + 1 big bet) and write a one-sentence justification for each choice.
checklist
- [ ] I have collected a large pool of usage areas.
- [ ] I fitted each idea into the standard template.
- [ ] I eliminated fuzzy and repetitive ideas.
- [ ] I scored value and feasibility based on evidence.
- [ ] I placed the ideas in the four-zone matrix.
- [ ] I limited the first wave to 3-4 areas.
- [ ] I have specifically marked areas with low error tolerance.