← All free workshops
Free workshop · Wednesday, July 22, 2026

Product Analytics 101: From Question to Decision with AI

Most analysis never leads to a decision. Learn the path from a question to one you can act on.

Recorded live on Maven.

What this session covers

Most analysis never changes a decision. This session teaches a seven-step path in three phases: before the data (decide, ask, hypothesize), the answer (measure), and after the answer (validate, share, act). It follows one request through Claude Code twice: should the Save for Later button be bigger because savers buy more? With only the data, Claude says yes. With the workbook, the answer turns out to be tenure, not the button.

  • Start from the decision, not the data. Name the call being made and who owns it before anyone writes a query.
  • Turn the request into an answerable question with a subject, a comparison and a timeframe, and write down what would prove you wrong.
  • Pick the one number that maps to the decision and set the threshold before you look.
  • A clean, confident answer proves nothing. Savers buy about 30% more, but they were longer-tenured, more active customers before they ever saved.
  • Claude owns the middle step. Deciding, asking, validating, sharing and following up are still yours.

Chapters

  1. 0:00 The analysis graveyard
  2. 1:34 Why good analyses change nothing
  3. 3:28 The seven steps
  4. 5:46 First run, nothing but the data
  5. 6:59 The friction that used to think for you
  6. 11:56 Second run, the workbook stops me
  7. 12:58 The overturn: tenure, not the button
  8. 15:31 Who owns each step
  9. 17:37 Where to go next

Questions this session answers

Why does so much data analysis never lead to a decision?

Because it starts from the data instead of a decision. Without a named call and owner, even a correct number ends up in a dashboard or deck and nobody acts on it.

What are the steps from a data question to a decision?

Decide what call is being made, ask an answerable question, state a hypothesis and what would disprove it, measure the one number that maps to the decision, validate it, share the version each person needs, and follow up on whether the number moved.

Why is a correlation like 'savers buy more' misleading?

The users who save items were already longer-tenured and more active before they saved anything, so the feature looks like it drives purchases when tenure and engagement do. In the randomized test in this dataset, the predicted 15% lift from a more prominent button never arrived.

What part of analysis can AI do for you?

AI can measure in seconds: write and run the query and return the number. The steps around it (framing the decision, writing the hypothesis, trying to break the answer, and following up) are where the judgment lives.

No transcript for this one yet.

The recording is the full session. Watch it above or on YouTube.

Free, every week

Join an upcoming free lesson.

Live on Maven. Learn AI analytics, prepare for interviews and grow your career. Bring a question from your own work.

WED NOV 18, 2026
10:00 AM PT
Check if your data is ready for AI
Register
WED DEC 2, 2026
10:00 AM PT
Stop Underselling Yourself in Data Interviews with AI
Register
THU DEC 3, 2026
10:00 AM PT
Think on the Spot in Analytics Case Interviews with AI
Register
WED DEC 9, 2026
10:00 AM PT
Master Growth Analytics with AI
Register
WED DEC 9, 2026
10:00 AM PT
Spreadsheets 101: Analyze Any Sheet with AI
Register
See all upcoming lessons →

Next cohorts start Jan 25 and Jan 11.

AI Analytics for Everyone
$1,800 · Jan 25 · ★ 4.9/5
Enroll on Maven
Agentic Analytics: Build an AI Analyst
$2,500 · Jan 11 · ★ 4.9/5
Enroll on Maven
Or join an upcoming free lesson. Register free