How to choose your first business AI project

Start with a workflow you can describe

Write down the people involved, the inputs they receive, the decisions they make and the result they need. For example, an operations team might spend time reading incoming requests and routing them to the right colleague. That is more specific than a general ambition to become an AI-powered business.

Ask the people doing the work where time is lost and where mistakes cause trouble. A workflow map can reveal missing information, unclear responsibilities or unnecessary handoffs that should be addressed alongside any technology change.

• Name the person who owns the workflow.
• Describe a normal case and an exception.
• Record how the work is completed today.

Compare AI with the alternatives

Some problems are best handled by a clearer form, a consistent rule or an integration between existing systems. Consider AI when the work involves information that is difficult to express as fixed rules, such as interpreting varied documents or helping people find relevant material.

Include the cost of reviewing outputs in the comparison. If every answer needs extensive correction, the proposed workflow may not yet offer an advantage. A smaller task or a different approach could be a better starting point.

Check data and operating readiness

Identify what information the system would need and whether your team can access and use it appropriately. Review a representative sample rather than assuming that a large document collection is consistent or current.

The NIST AI Risk Management Framework provides a useful reference for considering AI risks across design and use. For your project, turn broad considerations into practical questions about access, review, accountability and the consequences of an incorrect result.

• Can the relevant information be accessed with the right permissions?
• Who checks uncertain or consequential outputs?
• What happens when an input is missing or the system cannot answer?

Define a pilot with a decision at the end

Select a limited group of users and a bounded task. Record the current effort and quality, then agree how the team will compare the proposed workflow with that baseline. Use realistic examples, including difficult and incomplete inputs.

A pilot should produce evidence for a decision: expand, revise or stop. Set those criteria before the team becomes invested in a particular implementation. Include the people who will support the workflow after the trial.

Your AI project selection checklist

Use these questions in the first discussion. A gap is a planning task to resolve; it should not be hidden by a polished demonstration.

• We can explain the business problem and its current impact.
• We have considered a simpler alternative.
• We know what data is required and who owns it.
• We have defined how people review and override outputs.
• We can measure the pilot against a baseline.
• We have agreed integration, support and stopping criteria.

Frequently asked questions