Using AI to analyze data: The Lean In framework

Mary ·

Before You Start

Goal: Leave with a seven-step checklist for checking AI's data analysis and a plan to use it at work.

Read: "Running a Data Analysis with AI" (PDF attached below), by Charlotte Whatley, Head of Product at LeanIn.Org. Members are encouraged to read it beforehand, but a quick recap at the meeting works too.

Meeting Agenda · Approx. 60 mins

  • Member Updates · 10 mins
  • Icebreaker · 10 mins
  • Key Takeaways · 5 mins
  • Activities · 20 mins
  • One Action · 10 mins
  • Wrap-Up · 5 mins

👋 Member Updates · 10 mins

Go around your Circle and share personal updates. Keep them brief and focused on significant changes or important decisions since the last time you met. It's OK not to have one every month. If you have a One Action update from last time, share that too.

🧊 Icebreaker · 10 mins

Invite everyone in the group to share. Keep answers to one or two minutes per person.

What’s the most recent time you used numbers, a spreadsheet, or a metric at work? Did you use AI for any part of the analysis?

🔑 Key Takeaways · 5 mins

Take turns reading this checklist aloud around your Circle. Keep it on hand for both activities.

The seven-step checklist

  • Step 1: Start with the goal and the decision
    • Before using AI: Define your goal — what you want to learn from the data — and what decisions you'll make based on the data.
    • Ask AI: "Here's my data and my role. What questions can the data answer, and what decisions would each answer change?"
  • Step 2: Understand the data
    • Ask AI: "What is the title of each column and/or row corresponding to the data I'm asking about? What could make a title mean something other than what it says?"
    • Watch for: AI misunderstanding a column or row name.
  • Step 3: Explore
    • Ask AI: "What stands out for our question? Which measure are you using — average vs. median, total vs. percentage?"
    • Watch for: AI picking the wrong measure without telling you.
  • Step 4: Form possible explanations
    • Ask AI: "Give me 3 possible explanations for X finding, and what we'd see in the data if each were true."
    • Watch for: explanations the data can't prove or disprove, and explanations that sound good but are undercut by other data.
  • Step 5: Test
    • Ask AI: "How many cases (for example, rows, people, or responses) is this based on?"
    • Watch for: a striking result from a dozen cases.
  • Step 6: Play devil's advocate
    • Ask AI: "Argue the opposite. What's the strongest case against this?" Another tip: Open AI in a new window and ask it to double-check the analysis. It may find errors the first window missed.
    • Watch for: AI changing a right answer to a wrong one because you asked, "are you sure?"
  • Step 7: Create key takeaways
    • Ask AI: "Please summarize the findings and share your key takeaways and final recommendations."

🏋️ Activity 1 · 10 mins

Decide on your goal and decision

  1. Think of a dataset that you need to analyze at work — or choose an imaginary example from this list: sales by region, an employee or customer survey, or website or email results.
  2. Write one sentence for each:
    • Goal: what you want to understand.
    • Decision: what you will do differently depending on the answer.

Example: Goal: understand which notifications are useful to users. Decision: delete notifications so the ones we send are higher quality.

Discuss as a group:

  1. What goal and decision did you choose?
  2. How would you apply the seven-step checklist to your goal? Talk through how some of the steps might help.
  3. How else do you plan to apply the checklist in your role at work?

🏋️ Activity 2 · 10 mins

Decide what to check before you trust an answer

An AI tool gives you the three conclusions below from a data analysis. Each one sounds reasonable, but each has a gap or flaw.

  1. Pair up or work in groups of three.
  2. Take 5 minutes to go through the three claims below. For each, note down the questions you'd ask the AI to check before trusting the conclusion.

The claims:

  1. "Our email open rate rose 15% after we changed the subject lines, so the new style works."
  2. "Employees in the mentoring program get promoted more often, so mentoring leads to promotions."
  3. "Customer satisfaction is the highest it has ever been, up from 7.2 to 8.1."

Discuss as a group:

  1. Have each pair share a few of their questions with the group. Compare your questions with "Possible checks for Activity 2" at the bottom of the guide, but only after everyone has shared theirs.
  2. How do your questions differ from the possible checks at the bottom?
  3. Which claim sounded most convincing at first, and what made you want to check it?
  4. Are there any other checks you regularly use in your own data?

🎯 One Action · 10 mins

Commit to one concrete action — big or small — you'll take before your next meeting. Go around the Circle and share (1 minute or less per member). Vocalizing it helps hold each other accountable.

Run the seven-step checklist on one dataset from your work. Or paste a recent data analysis into a fresh chat and ask AI to find the flaws. Come back ready to share what happened.

📅 Wrap-Up · 5 mins

Before you break, plan your next meeting and close on an energetic note. Go around your Circle one by one and finish the sentence:

Now I am feeling… (choose one word).

  • Set the day, time, and location
  • Choose the next topic or resource
  • Confirm who's leading the discussion
  • Decide how you'll stay in touch in between

Possible checks for Activity 2

1. Email open rate: What else changed in the data collection period? How many emails were sent in each period?

2. Mentoring program: How do we know mentoring caused the change? Could something else have caused it?

3. Customer satisfaction: By which measure? How many people responded each time, and who answered?