Choosing an AI Initiative Before Choosing a Platform

Describe the work that needs to change

'We need AI' does not identify a decision, a user, or an operating constraint. A better brief explains where work stalls, who is affected, what evidence demonstrates the problem, and which part of the process the organization can actually change.

The MAP function in NIST's AI RMF Playbook emphasizes understanding purpose, context, and business value. It provides a useful starting point for framing an initiative before procurement. Our selection questions below are editorial guidance, not a claim that NIST endorses a particular investment or platform. NIST AI RMF Playbook: MAP 1, purpose, context and business value.

Use a one-page initiative brief

• Problem: describe one recurring task or decision, including the current baseline and its limitations.
• People: identify the process owner, users, and people affected by errors or delays.
• Alternative: explain what a better form, search function, rule, or process change could achieve without AI.
• Evidence: identify suitable data and how its quality, permissions, and gaps will be assessed.
• Evaluation: define a useful result, unacceptable outcomes, and a practical way to compare options.
• Decision: name who can proceed, revise, or stop the initiative after the evaluation.

Compare options using the same problem

Hypothetical example: staff spend time finding current internal procedures. The options might include fixing document ownership and search, introducing a guided knowledge interface, or combining the two. Test each against the same representative questions, including questions whose answer is absent or outdated.

Do not count a fluent response as a solved process problem. Ask whether the result is supported by current material, whether the user can check it, and what happens when no reliable answer exists. Include the effort of maintaining the source information, not only the demonstration experience.

Make a no-go decision useful

A short evaluation can legitimately conclude that the organization needs clearer data ownership or a simpler workflow first. Record that finding and its next action. This prevents a postponed platform purchase from being treated as a failed project when it has exposed the real prerequisite.

Only then compare platforms against requirements that survived the evaluation: integration, controls, maintainability, user needs, and an accountable operating model. Keep the original brief visible so a persuasive demonstration does not quietly redefine the problem.

References & further reading

• NIST AI RMF Playbook: MAP 1, purpose, context and business value