A practical screening framework
Strong first candidates usually have clear volume, visible friction and an outcome that can be measured. The goal is not to find the most impressive AI use case. It is to find a workflow where a focused proof can answer an economically useful question.
1. Pain and frequency
How often does the work occur, how much skilled time does it consume, and where does it create delays, rework or missed opportunities?
2. Bounded inputs and outputs
Can you describe what arrives, what a successful result looks like, which systems are touched and when the work should escalate?
3. Data and system access
Does the workflow depend on documents, databases or APIs that can be accessed appropriately? A use case may be attractive but impractical if its required information is unavailable or unreliable.
4. Measurable acceptance
Define success before building: task quality, cycle time, exception rate, adoption, cost or another outcome that matters to the workflow.
5. Risk and reversibility
Prefer early workflows where mistakes can be detected, reviewed and corrected. High-impact actions may require stronger controls or a different starting architecture.
Score the opportunity, not the hype
| Question | Strong first-use signal |
|---|---|
| Is the work repetitive? | High enough volume to justify engineering and measurement. |
| Is the result observable? | Quality or time can be compared against a baseline. |
| Can systems be accessed? | APIs, data or controlled interfaces are available. |
| Can exceptions be handled? | Human review or safe fallback is possible. |
| Is there a business owner? | Someone is accountable for the workflow and the result. |
If several workflows appear promising, an AI Opportunity & Workflow Workshop can compare them before a build decision.
References and further reading
This guide applies PlanckCyber’s workflow-screening method alongside risk and measurement principles in the NIST AI Risk Management Framework and the NIST Generative AI Profile.