Use this template to turn an approved AI opportunity into an accountable delivery plan. Complete it before building deeply, then update it as evidence changes.
1. Outcome and baseline
Define the workflow, user, business problem, current performance, target outcome, and method of measurement. Include quality, cycle time, capacity, cost, risk, or experience as appropriate.
Decision gate: the problem is valuable, measurable, and owned.
2. Intended use and scope
State what the system will and will not do, the users and affected people, launch geography, expected volume, and degree of autonomy. Name the business and operational owners.
Decision gate: intended use and accountability are clear.
3. Workflow design
Map current and future steps. Identify AI inputs and outputs, human judgments, approvals, exceptions, fallbacks, and downstream actions. Remove unnecessary automation from steps where review is difficult or consequence is high.
Decision gate: the future workflow is workable, not only the model demonstration.
4. Data and architecture
List data sources, classifications, permissions, retention, model and vendor choices, integrations, environments, identity, and logging. Record open dependencies and owner for each.
Decision gate: the proposed system can access the right information without exceeding authority.
5. Risk and controls
Assign a risk tier. Document failure modes, affected people, security and privacy issues, human oversight, action limits, and incident response. Link each material control to an owner and test.
Decision gate: residual risk is accepted for the pilot scope.
6. Evaluation plan
Define representative test cases, difficult cases, metrics, thresholds, reviewer qualifications, negative results, and limitations. Include end-to-end workflow testing, not only model accuracy.
Decision gate: results meet the pre-agreed bar and failures are understood.
7. Adoption plan
Identify user groups, manager responsibilities, role-based learning, communications, support, office hours, feedback, and capacity for practice. Define how the new process changes expectations.
Decision gate: users and managers are prepared for the controlled launch.
8. Launch and monitoring
Set scope, date, rollback plan, quality sampling, alerts, adoption measures, support owner, and review cadence. Define thresholds that trigger pause, remediation, or reassessment.
Decision gate: operational owners can observe, support, and stop the workflow.
9. Scale decision
Compare pilot evidence with the baseline and total cost. Choose scale, refine, hold, or stop. Record conditions, approvers, and next review date.
The plan should remain concise enough to use. Link detailed evidence rather than copying it into a static document. Every section should answer three questions: Who owns this? What proves it? What decision comes next?