AI Use Case Scoring Framework: A Rubric for Comparable Decisions

Without a rubric, AI use cases are selected by whoever argues best: the loudest sponsor, the most impressive vendor demo, the executive's pet idea. A scoring framework replaces that with criteria applied the same way to every candidate, so the ranked list can be defended and revisited as evidence arrives. This guide provides the rubric and how to use it, drawing on FISTA Solutions' AI enablement practice. The portfolio process it feeds is in ai portfolio management and the prioritization method in how to prioritize ai use cases.

What are the six criteria?

Criterion	Question	Scale anchors
Value	How large is the measured baseline and how plausible is the mechanism?	1: no baseline; 3: baseline with plausible mechanism; 5: large baseline, clear mechanism, confident range
Feasibility	Is the data accessible and good, and can systems be integrated?	1: data missing or blocked; 3: accessible with work; 5: ready with integration paths
Risk	What is the tier and how controllable are the risks?	1: high consequence, weak controls; 3: medium with known controls; 5: low consequence or strong controls
Cost	What does it cost to build and run at scale?	1: high with uncertain run cost; 3: moderate and modeled; 5: low with predictable run cost
Time to evidence	How soon will there be measurable results?	1: over a year; 3: within two quarters; 5: within one quarter
Platform leverage	Does it reuse or build shared capability?	1: bespoke; 3: partial reuse; 5: reuses platform or builds it for others

Risk tiering is in ai model risk management and cost modeling in the AI total cost of ownership whitepaper.

How should value be scored?

On three things: the size of the measured baseline the use case addresses, in cost, time, error, or revenue; the plausibility of the mechanism by which AI changes that baseline; and confidence in the estimate. A use case without a measured baseline scores low by design, which is not a penalty but a signal that measurement should be the first funded step. Baseline practice is in ai business case template.

How should feasibility be scored?

Mostly on data and integration readiness: whether the data exists, is accessible with permissions, and is of usable quality; whether the systems the use case lives in have integration paths; whether domain experts are available to define correctness; and whether the task is within demonstrated model capability. Model capability is rarely the binding constraint; data and integration almost always are. Readiness assessment is in the ai data readiness checklist.

How should risk, cost, and time be scored?

Risk by tier and controllability, scored by the risk function rather than the proposing team. Cost by build estimate and run cost modeled at expected adoption, including model usage and human review. Time to evidence by how soon a measurable result can exist, which favors bounded workflows with existing baselines. Run cost drivers are in ai agent maintenance cost.

How should platform leverage be scored?

By whether the use case reuses shared capability such as the gateway, evaluation infrastructure, retrieval over already-indexed content, or existing integrations, or builds capability that later use cases will reuse. Leverage is why the third initiative should be cheaper than the first, and it is the criterion most frameworks omit. Platform design is in ai operating model.

How do weights and sessions work?

Weights reflect strategy: an organization prioritizing quick credibility weights time to evidence; one building foundations weights platform leverage; a regulated one weights risk. Publish weights before scoring and do not adjust them afterward. Score in short cross-functional sessions with business, technology, data, and risk present, record reasoning per criterion, resolve disagreements with evidence, and produce a ranked list. Rescore at each checkpoint. Session mechanics are in how to run an ai steering committee.

What does a scored example look like?

A support triage agent: value 4 with a measured ticket volume and handling-time baseline; feasibility 4 with ticket data accessible and a ticketing API; risk 3 as customer-facing with strong gating; cost 3 with modeled run cost; time to evidence 5 within a quarter; leverage 4 reusing the gateway and building retrieval others will use. A speculative forecasting model with no baseline: value 1, feasibility 2, risk 3, cost 2, time 2, leverage 2. The ranking is obvious and defensible. A worked build for the first is in how to build an ai ticket routing system.

What distorts scoring?

Sponsor seniority; vendor demonstrations standing in for feasibility; unmeasured baselines treated as facts; risk scored by the proposing team; weights changed after scores are known; and sessions that score dozens of use cases at once without evidence. Each is corrected by the rubric's design and by recording reasoning. Kill criteria that follow from low scores are in when to kill an ai project.

How FISTA Solutions applies use case scoring

FISTA Solutions runs scoring sessions with client leadership, measures baselines where they are missing, assesses data and integration readiness, models costs, and delivers the highest-scoring use cases with the evidence that lets the portfolio rescore truthfully. The AI enablement practice leads scoring and platform, forward deployed engineers deliver, and AI agents supplies the systems. The record behind the approach is 150+ projects for 50+ companies.

To choose AI use cases on a rubric rather than a pitch, message FISTA on WhatsApp, or read ai portfolio management for what happens after scoring.