AI for product managers

Setup walkthrough

About 20 minutes

1. 1

   Create a shared project and give it the context

   You10 min

   In Claude, open Projects and create one called "Product", then share it with your team (how projects work; more in Claude projects). Download the file below: it's the context table from the top of this page plus the rules that go with it. Fill in the parts in brackets and paste it into the project's instructions. Then upload your strategy doc, your PRD template and last month's product review.

   Product project: context and rulesCopyDownload .md

   Show the full prompt (66 lines)Hide the prompt

   You work with [Company]'s product team. This file is the context and the
   rules for every answer in this project. When you apply a rule, say so.
   
   ## Context: how our analytics tools count
   Every tool decides who is active and who is one user. Use this to label
   figures, not to "correct" them.
   - Active users. Amplitude: any event makes a user active unless the event
     is marked inactive. Mixpanel and PostHog: users who did the chosen event.
     Our active events: [events]. "Weekly" means [calendar week from
     Monday / rolling 7 days].
   - Identity. Anonymous visitors who aren't merged with their account are
     counted twice in Amplitude. Mixpanel's merge can take up to 24 hours.
     Our counts include [identified users only / anonymous too].
   - Retention. Mixpanel defaults to "On or After", PostHog to "On",
     Amplitude to rolling 24-hour days. Ours: [On / N-day], [calendar /
     24-hour] periods, return event [event].
   - Time zone. Our project time zone is [zone]. Amplitude defaults to UTC.
   - Internal users. Excluded by [filter / data view / blocked at
     ingestion]. In PostHog the filter is per insight; check it's on.
   - Recent and sampled data. The last [2] days may still be receiving
     events. Amplitude and Mixpanel can sample 10% of users; say when a
     figure is sampled.
   - Experiments. Our tool is [tool], method [sequential / Bayesian /
     frequentist], confidence [95%]. Read both from the experiment's settings.
   - Feedback. Intercom conversations and Zendesk tickets are not customers.
     Merged conversations count unless filtered out.
   
   ## Definitions
   - Use the metric-glossary skill for every metric. If a metric isn't in
     it, ask; don't define it yourself.
   - Activation = [event] within [7] days of sign-up.
   - Sign-ups = [event]. Customers = [paying accounts / workspaces].
   - Rates are shown with their counts (813 of 2,140), never alone.
   
   ## Numbers
   - Use connected tools or attached files only. If a figure isn't
     available, write [NO DATA]. Never use a benchmark from memory.
   - Do calculations in code and show the code.
   - Show the chart, query or API call (with its URL) under every table.
   - State the date range and time zone at the top of every table.
   - A KPI tile always uses the same figure and definition as the table.
   
   ## Experiments
   - Report against the test plan: its primary metric, guardrails and length.
   - Run a sample ratio check before reporting a result.
   - Give differences in percentage points first, relative lift second.
   - Don't forecast annual impact unless I give you the volumes to use.
   
   ## Research and feedback
   - Every theme needs verbatim quotes with participant ID and timestamp,
     and a count of interviews it came up in.
   - Mentions in fewer than 3 interviews are "Not a theme yet".
   - Say who wasn't in the research. Don't infer why customers churn
     unless a churned customer said so.
   
   ## Tools
   - Read only. Don't create, edit or send anything in Amplitude, Mixpanel,
     PostHog, Linear, Jira, Confluence, Notion, Intercom or Slack unless my
     message asks for that exact change.
   - If a tool returns a capped or paged result, say how many rows you got
     and whether there were more.
   
   ## When you're unsure
   - Ask one clarifying question rather than choosing a definition yourself.
   - Count audits every answer automatically. Keep the queries and files you
     used visible in the chat so it can check each claim against them.

   
2. 2

   Connect the tools that sign in from Claude

   You2 min each

   These connect like any other app: open Customize > Connectors, pick the tool and sign in (connectors guide). On a Team plan, an Owner switches each one on for the company first. Jira and Confluence arrive together through Atlassian's Rovo MCP server. Linear has a read-only address; start with that one. If your Amplitude, Mixpanel or Intercom data is hosted in the EU, add it as a custom connector with the EU address from the vendor's guide. Tap a tool on the map above for its own guide.

   
3. 3

   Let Claude write your metric glossary from your analytics tool

   You15 min

   A skill is a short file Claude reads whenever it's relevant, and Claude can write this one for you. In a new chat with your analytics tool connected and code execution on, paste the prompt below with the names of the dashboards your team actually uses. Claude reads every chart on them and gives you back a glossary of what each metric really counts, with conflicts marked. Fix anything that's wrong, then upload it under Customize > Skills (how to add a skill). More on skills in Claude skills.

   Metric glossary from your analytics toolCopyDownload .md

   Show the full prompt (31 lines)Hide the prompt

   Build a metric glossary for our product team from our analytics tool,
   and package it as a skill I can upload to Claude.
   
   1. In [Amplitude / Mixpanel / PostHog], list the dashboards our team
      uses most: [dashboard names or links]. Open each one and list every
      chart on it, with the chart's URL.
   2. For each metric those charts show, write down exactly what it counts:
      - The events it uses, by their raw names, and any filters.
      - Whether it counts users or events, and which users: identified
        only, or anonymous too. Whether internal and test users are excluded.
      - The time window: daily, weekly, rolling 30 days or calendar month.
      - For retention: N-day (active in that period) or unbounded (active
        in that period or later), and the return event.
      - The project time zone.
   3. Where two charts call the same thing by different names, or the
      same name counts different things, list both and mark it [CONFLICT].
      Don't choose between them.
   4. List events that look renamed or duplicated (for example
      "Signed Up" and "sign_up"), with their volumes for the last 30 days.
   5. Write it as a skill: a file called SKILL.md that starts with
      ---
      name: metric-glossary
      description: [Company]'s product metric definitions. Use whenever
        answering a question about product metrics, retention, activation
        or experiments.
      ---
      then one entry per metric: name, plain-English definition, events,
      window, the chart it comes from (URL).
   6. Read only. Don't create or change any chart, dashboard or cohort.
      Package it as metric-glossary.zip (a metric-glossary folder
      containing SKILL.md) so I can upload it under Customize > Skills.

4. 4

   Connect Count

   You2 min

   Download the Count desktop app and follow its walkthrough; it adds Count to Claude for you. After that, Count audits every answer automatically, without leaving the chat. What Count is.

   Download for Claude Desktop

   
5. 5

   Ask your admins to approve Slack and Mixpanel

   Slack or Mixpanel admin5 min each

   Slack's connector only works once a workspace admin has approved it. Mixpanel's needs an organisation admin to switch MCP on under Settings > Org, unless your account is on a free or Growth plan and was created after 1 August 2026. Send each admin the vendor's guide from the map above; there's nothing for them to install. Until then, Zendesk and Mixpanel reports can come in as CSV exports.

   
6. 6

   Try it on a week you already know

   You10 min

   Run the first workflow on last week and compare it with the numbers in your last product review. If they differ, the difference is usually a definition to add to the project instructions or the glossary.