Build a personal AI chief of staff for one small business experiment

Give the assistant a bounded job

Choose one active experiment, such as testing demand for a specialized service package. State the decision you are trying to make, the current stage, the available time, and the maximum budget you have already approved. List the people involved and the commitments you have actually made.

Use a short role description: “Help me prepare decisions and track progress on this experiment.” Specify the outputs you want: a prioritized work list, a list of unanswered questions, and a concise review of new evidence. Retain control over spending, publishing, customer contact, and changes to your accounts.

Create a small source of truth

Use one document or folder containing the current brief, task list, evidence, and decision log. Give important notes a date and a source. Distinguish a customer statement from your interpretation and an AI suggestion. Remove outdated instructions or mark them superseded so yesterday’s plan does not silently compete with today’s.

For a hypothetical service experiment, the brief might state: three hours available this week, an existing landing page, two customer conversations completed, and no paid advertising approved. Those constraints should shape the next action. The assistant should not turn a small learning goal into a complicated launch program.

Use a daily brief to expose the next decision

At the start of a work block, provide the current task status and any new evidence. Ask for the smallest useful set of priorities. Make dependencies explicit: interviewing a customer may matter more than polishing a page if the offer itself remains unclear.

Limit the output to something you can act on. A long plan can become another task to manage. Ask the assistant to identify what can wait, what is blocked, and what decision only you can make. Have it say when the available information is too old to support a recommendation.

Prompt to try

Act as a planning assistant for this single business experiment: [current brief, dated evidence, task status]. I have [time] available today. Recommend at most three actions, with the decision or uncertainty each action addresses. Identify blockers, outdated information, and one task to defer. Separate facts from suggestions. Do not contact anyone, spend money, publish, or change connected systems.

Keep an evidence-based decision log

After a meaningful choice, record the decision, date, evidence, expected result, and the condition that would make you revisit it. This prevents the assistant from repeatedly reopening settled questions or treating tentative ideas as commitments. Update the source document yourself or review the assistant’s proposed edits.

Ask for disagreement when it will improve the decision. The useful question is not whether the assistant likes the idea. Ask which assumption carries the most risk and what observation would contradict it. A confident critique is still a hypothesis until you check it against reality.

Run a weekly review that ends with a choice

Compare completed work with the purpose of the experiment. Did you reduce uncertainty, obtain useful customer evidence, or complete a necessary operational step? Count activity only when it helps explain progress. Ten generated social posts do not establish that anyone wants the offer.

Choose continue, change, pause, or stop, and state why. If the evidence is weak, the next action may be one targeted conversation or a closer look at an existing result. Keep estimated outcomes and actual outcomes in separate columns so enthusiasm cannot quietly become a claimed result.

Prompt to try

Review this experiment using the dated brief, task history, and evidence: [materials]. Summarize what changed, what we actually learned, and what remains assumed. Identify the strongest evidence against our current direction. Recommend continue, change, pause, or stop, with reasons and confidence limits. Propose the smallest next action that could change the decision. Do not infer demand or revenue from activity alone.

Add scheduling only when the routine is useful

A saved prompt is a reusable instruction, not a scheduled assistant. If you want a recurring briefing, configure an actual scheduling feature or automation with a trigger, timezone, source access, output destination, and failure handling. Verify the next run and test the result. Features and account access vary, so check the tool you are using.

Keep the first scheduled output informational and review it yourself. An assistant reading a stale document will produce a stale briefing more efficiently. Decide who updates the source and what the output should say when information is missing. Microsoft’s agent measurement guidance stresses that value requires outcomes and monitoring, not just activity.

Judge this assistant by whether it helps you finish useful work and make clearer decisions. If maintaining it costs more effort than it returns, simplify the brief, reduce the frequency, or use it only at the moments where it helps.

Try it on one real task

1. Create a one-page brief for one active experiment with explicit time and spending limits.
2. Run the daily brief manually and complete one recommended action.
3. Record a decision and use the weekly-review prompt before adding any scheduler.

Check before you use the result

• The assistant serves one defined experiment.
• Facts, interpretations, and suggestions are distinguishable.
• Publishing, spending, and customer contact require your explicit decision.
• A recurring prompt is not described as active until scheduling is configured and verified.

Take the next step

Download the Wealth Experiment AI practice pack for these prompts, exercises, and review checklists.

New to AI? Take the free first-workflow course at AI Teachers. Already using it? Practice building a repeatable workflow.

Sources and further reading

• Microsoft Learn: Monitor, measure, and report value
• NIST: AI Risk Management Framework

Sources support the principles discussed; the examples and suggested workflows are original teaching exercises.