AI at work

How to use AI as a product manager: start with one real workflow

A practical first workflow for PMs: choose a task, supply useful context, review the result, and measure whether AI actually helps.

You do not need to redesign your entire job to start using AI well. Choose one recurring task, define what a good result looks like, and test whether AI helps you reach that result with less effort or better evidence.

A useful first workflow is preparing a weekly product decision brief. It combines information you already have, produces something easy to inspect, and can expose whether the tool understands your context.

Choose work you can check

Start with a bounded task: organizing approved notes, comparing two proposed solutions, finding missing requirements, or drafting an update from a decision log. Avoid beginning with a broad instruction such as “Tell me our product strategy.” The model would have to invent too much missing context.

Write your current baseline. How long does the task take? Where do mistakes occur? Who uses the output? What does the recipient need to decide? A faster document is not useful if it creates another meeting to correct it.

Give the model a context packet

For a weekly brief, provide the product goal, intended audience, recent evidence, decisions already made, constraints, and unresolved questions. Mark dates and sources. Explain which material is authoritative if documents disagree.

Use only information approved for the tool and account you are using. An approved work account and a personal account may have different handling rules. Remove irrelevant personal details and avoid pasting credentials or customer secrets.

Keep three categories separate: observed facts, interpretations, and proposed actions. The distinction is more useful than asking the model to sound like a senior product leader.

Ask for a decision-shaped output

Try this prompt:

Using only the supplied context, draft a one-page product decision brief. Include the decision needed, supporting evidence with source IDs, contradictory evidence, options, tradeoffs, and a recommended next test. Label inference and missing information. Do not invent data, customer quotes, dates, owners, or commitments. End with the three questions most likely to change the recommendation.

Now inspect the result. Can you find every quoted source? Did a tentative idea become a commitment? Did the model silently combine different customer segments? Correct the brief and save those corrections as instructions for the next run.

Measure the whole workflow

Suppose preparing a brief normally takes 45 minutes. With AI, collecting inputs takes 12 minutes, drafting takes 3, checking takes 14, and correcting takes 6. Total time is 35 minutes: a hypothetical 10-minute saving, not a 42-minute saving based only on drafting time.

Also ask whether the brief helped the team make a clearer decision. Track major corrections, missing evidence, and the amount of follow-up required. If the AI version is fast but unreliable, narrow the task or improve the inputs before repeating it.

Turn a good experiment into a habit

Save the context structure, prompt, output example, and review checklist together. Assign someone to keep the context current. A repeatable workflow is more valuable than a prompt you cannot reconstruct next week.

Once this works, expand one step: compare stakeholder requests, prepare research questions, or identify gaps in a requirements draft. Your goal is to steadily improve the product work that matters, while preserving the judgment needed to decide what happens next.

Try it: Complete the AI Workflow Canvas for one task this week. Continue with the 30-day practice plan.