A/B Testing for Product Managers

The Four Deaths

What a Test Can and Cannot Tell You

Why "Let's Just Test It" Is Often Wrong

The Shape of a Good Test

What This Course Covers

The Three Questions

When to Just Ship It

Test, or Go and Talk to Someone?

Exercise 1: Triage Your Test Ideas

Why "We Think This Will Improve Conversion" Isn't One

The Four-Part Hypothesis

Saying in Advance What Would Make You Wrong

One Change or Several?

Exercise 2: Write a Real Hypothesis

The Four Numbers

The Calculation, Worked Through

Why Most Tests Are Too Small

The Weekly Cycle Rule

Exercise 4: Do the Arithmetic Before You Build

Before You Turn It On

What to Check on Day Two

The Six Things That Invalidate a Running Test

Exercise 5: Set Up Properly

What "Significant" Actually Means

Effect Size and the Range

The Flat Result

When the Result Contradicts What You Believed

Exercise 6: Read a Result Properly

Winning Isn't the Same as Worth Shipping

Exercise 7: Decide and Record

Slicing Until Something Wins

Novelty and Primacy

Contamination, Mismatch and Survivorship

Exercise 8: Audit Your Own Results

Five Situations Where Testing Won't Help

What to Do With Too Little Traffic

Knowing When to Stop

Exercise 9: Diagnose Your Testing Programme

Final Project: Design, Run and Report One Real Test

Before You Build: Deciding Whether to Test

Before You Build: Writing and Stress-Testing a Hypothesis

Before You Build: Choosing Your Metrics

Before You Build: Sample Size and Duration

Running It: Pre-Launch Review

Reading It: Reading a Result Honestly

Reading It: Auditing for the Six Lies

Afterwards: The Test Readout

Checking Yourself: Is Testing Right for Your Situation?

Glossary: Words Explained

Why "We Think This Will Improve Conversion" Isn't One | A/B Testing for Product Managers

Lesson 1

Why "We Think This Will Improve Conversion" Isn't One

That sentence contains no information. It doesn't say what you're changing, for whom, why you believe it, or how much you expect. Every part of it is unfalsifiable in advance and reinterpretable afterwards, which is exactly what you don't want.

The practical consequence shows up at readout time. The result comes in at +0.4% and the debate begins: is that an improvement? Was that what we expected? Should we ship? None of these are answerable, because nothing was specified. A proper hypothesis answers all three before the test starts.