Answer first
Match the support type to the missing evidence. Advisory resolves what to prioritize and whether to invest. Training develops a defined capability in a defined audience. Development turns a sufficiently specified workflow into an operating system or agent.
NIST's Generative AI Profile emphasizes governance, context, measurement, and management across the lifecycle. Moving to development before those responsibilities are clear usually creates rework rather than faster adoption.
Choose by the decision you need to make
The service label matters less than the concrete decision and artifact expected at the end.
| Need | Best starting point | Expected evidence |
|---|---|---|
| Select and rank AI themes | Advisory | Prioritized backlog, assumptions, risks, and next test |
| Decide whether to fund a PoC | Advisory | PoC scope, baseline, evaluation, and go/no-go rule |
| Build repeatable staff capability | Training | Role-specific exercises, guidance, and observed competency |
| Implement a defined workflow | Development | Requirements, integration, controls, tests, and operating owner |
| Improve a live AI workflow | Development plus advisory | Logs, error analysis, KPI movement, and change decisions |
Check prerequisites before buying development
Development is premature when the business owner cannot name the workflow, representative cases are unavailable, risk approvals are unknown, or no outcome can be measured. Resolve those gaps through discovery or a short advisory phase.
Training is also not a substitute for ownership. It works best when participants have a real workflow, approved tools, and a manager who will reinforce the new practice.
Identify the right starting point
The AI theme diagnostic classifies the current stage and provides a practical next action.