The Framework: Two Dimensions, Ten Sub-Scores
Every use case gets scored 1-5 on ten sub-dimensions, five for business value and five for technical feasibility. The averages place it in one of four quadrants. The discipline is in the sub-scores: they force the conversation the roadmap usually skips.
Y AxisBusiness Value Score (BV)
- Strategic alignment (1-5): Does it advance a board-level priority, or a department preference?
- Revenue or cost impact (1-5): Is the financial case quantifiable, with a number finance would sign?
- Competitive differentiation (1-5): Does it create defensible advantage, or table stakes everyone will have?
- Speed to value (1-5): Can it generate measurable ROI within 12 months?
- Executive sponsorship (1-5): Is there a C-suite owner with budget authority, not just enthusiasm?
X AxisTechnical Feasibility Score (TF)
- Data readiness (1-5): Is the required data available, clean, and accessible today?
- Model maturity (1-5): Does a proven model architecture exist for this problem class?
- Integration complexity (1-5, inverted): Score 5 if it slots into one system, 1 if it touches your core ledger. Lower = harder.
- Team capability (1-5): Does the internal team have the skills, or is every step outsourced?
- Regulatory risk (1-5, inverted): Score 5 for unregulated internal tooling, 1 for regulated customer-facing decisions. Lower = more regulated and risky.
Score Your Pipeline
Name a use case, score the ten sub-dimensions plus four risk factors, and add it to the matrix. Plot up to 10 at once. Hover any dot for the full breakdown. Nothing leaves your browser.
New Use Case
Business Value
Technical Feasibility
Risk Score (flags, does not move the dot)
0 of 10 use cases plotted. Risk score above 3.5 gets a warning flag.
Kill Criteria: The Gates Before Development
A high matrix score is necessary but not sufficient. Every use case must pass all six gates before a dollar of development spend. These are pass/fail, not weighted: a single failure holds the use case at the gate until fixed.
Is the required data legally accessible and ethically usable?
Contract terms, consent scope, and cross-border transfer rules kill more use cases than model performance does. Confirm usage rights in writing before scoping, not during legal review at launch.
Is there a measurable success metric the business will accept?
If the sponsoring business unit will not commit in advance to the metric and the target, the pilot result will be relitigated no matter what it shows.
Is there an executive sponsor with budget authority?
Enthusiasm without budget authority is a workshop, not a sponsor. The sponsor must be able to fund production scaling without a new approval cycle.
Has legal and compliance reviewed the use case?
A 30-minute pre-review before development is cheap. A compliance objection at launch, after six months of build, is how pilots die at 90% complete.
Is there a plan for human oversight during the first 6 months?
Name the people who will review outputs, define the escalation path, and budget their time. "The model is accurate" is not an oversight plan.
Has the team defined what "good enough" model performance looks like?
Set the acceptance threshold before building. Teams that define it afterward always discover the threshold is exactly what the model achieved.
Sequencing: Order Matters as Much as Selection
Two roadmaps with identical use cases can produce opposite outcomes depending on sequence. Four rules govern the order.
Infrastructure builders go first
Use cases that force the creation of shared data pipelines, feature stores, or governance processes pay a dividend on every use case that follows. Sequence them early even if their standalone ROI is middling.
Early wins must be visible to skeptics
The first production win should land in a function whose leaders are undecided about AI. A win inside the team that already believed changes no one's mind and buys no political capital.
High risk follows low risk
Run low-risk use cases first to build organizational confidence, oversight muscle, and governance track record. A high-risk failure in month three can freeze the entire program; the same failure in month eighteen is a managed incident.
Platform use cases unlock portfolios
Some use cases are doors, not rooms: a document intelligence capability built for claims can be redeployed to contracts, onboarding, and audit. Score the downstream unlock, not just the first application.
The 12 Most Common AI Use Cases, Pre-Scored by Industry
Typical BV and TF scores for the use cases that dominate enterprise pipelines, with the rationale. Use these as calibration anchors when scoring your own, then adjust for your data reality and regulatory posture.
| Use case | BV | TF | Quadrant | Rationale |
|---|
Run this as a workshop, not a spreadsheet
The scores matter less than the argument they force between business and technology leaders. Arjun Jaggi facilitates AI use case prioritization workshops for executive teams.
Book a prioritization workshop