Digital asset

Analyze Anything: The Root-Cause and Experiment Playbook + 3 Guides

AI Analyst Lab

AI Analyst Lab

AI Analytics by Sravya, Shane, and Hai

See all products from AI
Almost every important conversation on a product team is one of two questions. A number moved, and why. Or we changed something, and did it work. AI made the mechanical half of both easy. It will tell you conversion is down 4 points in a minute, and that the number went up after your launch in seconds. What it will not tell you is that the drop lives entirely on one phone model, that the product never got worse and the mix of visitors just changed, or that a holiday sale is the real reason the line climbed. That judgment is still yours. These four guides are about that judgment.

What's inside

  • The Root-Cause and Experiment Playbook (37 pages). The flagship. All three parts end to end, from a red arrow in Slack to a causal claim someone can bet a roadmap on.

  • Why Did the Number Move? (14 pages). Root cause decomposition, the funnel debugging map, segment-first thinking, and the mix-shift trap that fools almost everyone.

  • Run the Experiment (14 pages). Testable hypotheses, power intuition without formulas, the result interpretation tree, mixed results, and the one-page experiment brief.

  • Prove Impact Without an A/B Test (13 pages). The correlation trap, input-to-outcome maps, and three practical methods: before-versus-after, compare-the-changes, and find-the-twins.

What you'll be able to do

  • Split a moved number into its parts until one cause is left standing, instead of running a meeting on competing hunches.

  • Catch the mix-shift trap, where an average moves even though nothing about the product actually changed.

  • Write a hypothesis specific enough to be proven wrong, and tell whether a test is even big enough to detect what you are hoping for.

  • Read a result honestly, including the ugly case where the primary metric wins and a guardrail quietly gets worse.

  • Make a credible causal claim when you never got to randomize, and say out loud how much to trust it.

Who it's for

Product managers, designers, engineers, and founders who get handed a number and have to explain it. No SQL, no Python, no tools to install. Every example runs on one made-up online store called NovaMart, explained from scratch, so you can follow it with no background at all.

Free

Four free guides on why a number moved and whether your change actually worked. 78 pages. No SQL, no Python, no tools.