Seven questions to ask before building with AI

Before estimating an AI project, answer the questions that determine whether it should be built at all. The worksheet below is for a business owner or team lead preparing for discovery. Its purpose is to expose missing decisions, not produce a score that automatically approves the idea.

1. What task needs to change?

Name the person, input, output, and recurring difficulty. “Prepare an accurate handoff from meeting notes” is specific enough to examine. A general ambition to become more innovative is not yet a task brief.

2. How will we evaluate the result?

Record current effort and quality. Include checking and rework. Define unacceptable outcomes as well as desired improvements. For a handoff, an invented commitment could be unacceptable even if the first draft is fast.

3. What can the available tools actually do?

Test the essential behavior with approved access and representative inputs. Include incomplete and conflicting information. Distinguish a feature you have verified from one you expect a vendor to release.

4. Is custom software necessary?

Compare an existing feature, a better working method, an integration, and a custom build. Use the same task for each comparison. The build, buy, or wait framework helps make this choice explicit.

5. Who owns the result after launch?

Name a person who can review failures and coordinate changes. Identify technical support separately if needed. Shared participation does not remove the need for clear responsibility.

6. Where will information go?

List source systems, data categories, approved tools, access requirements, and retention or export needs. Resolve those questions with the appropriate administrator before using actual records in a trial.

This is project planning, not a substitute for legal or security advice where the data or use case requires it.

7. What is the smallest useful test?

Choose a complete task with a reviewable outcome. A prototype may test feasibility without being ready for production. Set the schedule around the work and approvals required; there is no universal two-week deadline for a sound evaluation.

Turn the answers into a decision

Write one paragraph describing the proposed test, who approves it, what evidence will be collected, and the conditions for continuing or stopping. If an answer is missing, name the next action needed to obtain it.

The consultant selection guide explains how to use this brief with a provider. AI Automation covers existing workflows and integrations; Application Development covers new software products and private tools.