Before You Add AI

ANDRÁE ALEXANDER, PH.D.

Before You Add AI

Five questions for music, business, and creative work

A useful starting point

Before choosing an AI tool, decide what you need it to do. A fast answer is not automatically a useful answer, and an impressive demonstration does not tell you how a system will behave when the information is incomplete.

Use these five questions with one real task. Keep the first test small enough to review. The purpose is not to prove that you should use AI. It is to decide whether a particular use improves the work without creating a larger problem.

1. What problem are you actually solving?

Describe the task without naming a tool. "Use AI for my music business" is too broad. "Draft replies to routine booking inquiries using my approved performance information" gives you something to investigate.

Write down how the task works now, who does it, how often it happens, and what makes it difficult. Then name the improvement you want. Less review time, fewer missing details, or more consistent answers are different goals. Choose the one that matters.

Your note: The task is ____. The current difficulty is ____. A useful improvement would be ____.

2. What information is the system allowed to use?

List the material needed for the task and identify its source. For a booking assistant, that may be an approved biography, performance options, technical requirements, and a current calendar. It does not automatically include every contract, contact, or unreleased recording you possess.

Check permission before uploading information. Review the service's current settings and terms for storage, reuse, access, and deletion. Separate material that can be shared from material that must stay private. When a source is missing or uncertain, the system should identify the gap rather than invent an answer.

Your note: Approved sources are ____. Material excluded from this test is ____. The person responsible for keeping the sources current is ____.

3. Which decisions stay with a person?

Be specific about the boundary between assistance and authority. A system might draft a reply without being allowed to quote a fee, promise a date, accept a contract, or send the message.

Give a named person responsibility for reviewing the output. Decide what should stop the workflow: conflicting information, a question outside the approved scope, a request for confidential material, or an action that would commit money or make a promise. "A human is involved" is not enough unless you can explain what that person checks and can prevent.

Your note: The system may ____. It may not ____. Review and approval belong to ____.

4. How will you test whether it is useful?

Prepare ordinary examples, difficult examples, and examples the system should decline or pass to a person. Use material you are permitted to test. Compare the results with the current process, including the time required to correct mistakes.

Decide what counts as an unacceptable error before running the test. A polished response containing the wrong fee is not a success. Neither is a draft that saves a minute to generate but takes ten minutes to verify. Record the examples, the output, the correction, and the final decision so the comparison can be repeated.

Your note: The test examples are ____. Unacceptable errors include ____. We will compare ____ with the current process.

5. Who will maintain it, and when will you stop?

A workflow needs attention after the demonstration. Someone must update information, review failures, manage access, and decide whether changes to a tool require another test.

Include subscription costs, review time, and maintenance in the decision. Keep a way to return to the previous process. Decide when the system should be paused, revised, or retired. Continuing to use it should remain a choice supported by the work, not a commitment made because the first demonstration looked good.

Your note: The owner is ____. The review date is ____. We will pause or stop when ____.

Worked example: routine performance inquiries

This is an illustrative test, not a report of a completed client project.

Task: prepare a first draft replying to ordinary performance inquiries. Approved material: a current artist biography and performance-information sheet. Human decisions: availability, fees, contract terms, and all outbound messages.

A useful test includes a normal inquiry, an impossible date, a request for an unlisted fee, a contradictory instruction, and a request for confidential information. The reviewer checks the draft against the approved information and records both correction time and errors.

Keep the workflow only if the result is useful after review. When the source material does not answer the question, the appropriate draft is a request for clarification or a handoff to the responsible person.

One-page decision record

Decision

Your answer

Task and intended user

Current process and baseline

Approved information and exclusions

Actions that require a person

Test examples and unacceptable errors

Review time, running cost, and owner

Decision: proceed, revise, or stop

Next review date

A clear "not yet" can be a good decision. It tells you what must change before the tool belongs in the work.

Continue the conversation

I work with artists and organizations on music-business strategy, research questions, and practical uses of AI. Describe the task and the decision you are facing when you get in touch.

Explore consulting | Join my email updates

Further reading: NIST's AI Risk Management Framework provides a broader voluntary reference for thinking about AI-related risks. This worksheet is not a NIST certification or endorsement.

NIST AI Risk Management Framework

AI disclosure: AI tools assisted with research, drafting, and editing of this guide. It is an educational decision worksheet, not a security certification or individualized professional opinion.