Decision guide
How to choose a responsible first AI workflow
The best first workflow is rarely the most impressive demonstration. It is the work that can produce useful evidence while remaining governable.
Use six screens
Score candidate workflows against the same criteria. A workflow that fails one screen may need redesign before it is ready. A workflow that fails several screens is probably not an appropriate first implementation.
Useful
The workflow addresses a real operating problem with a visible cost, delay, burden, or service consequence.
Bounded
The beginning, end, owner, inputs, outputs, and exceptions can be clearly described.
Governable
Information use, human review, authority, escalation, and documentation can be controlled.
Measurable
The institution can establish a baseline and evaluate time, quality, adoption, risk, or service improvement.
Recoverable
Errors can be detected and corrected without creating unacceptable harm or irreversible consequences.
Adoptable
The people doing the work are available to shape, test, learn, and take ownership of the process.
Avoid these first-workflow traps
- A high-stakes decision where an error could materially affect a person’s rights, access, employment, finances, or safety.
- A process that depends on information the institution is not permitted or prepared to place in the selected system.
- A workflow with no accountable owner, no baseline, or no staff capacity to participate.
- A project selected because a vendor demonstrated a tool rather than because leadership defined a problem.
- An implementation that cannot be paused, reversed, or handed back to a reliable human process.