Best of the inbox: how to actually run a subject-line test

Best of the inbox: how to actually run a subject-line test
Most 'A/B tests' prove nothing. Here's the discipline that gives you a result you can trust.

🧪 Design
— One variable only — test length OR personalization OR emoji, never two at once. Otherwise you can't attribute the lift.
— Size the sample before sending — under ~1,000 per arm and your open-rate gap is usually noise. Use a calculator, don't eyeball it.

📊 Read it right
— Measure replies/clicks, not just opens — Apple MPP inflates opens; downstream action is the honest metric now.
— Hold-out the winner for 24h — re-send the champion to a fresh slice to confirm it wasn't a time-of-day fluke.

📥 Bank it
— Log every test in one sheet — variable, lift, significance. Your swipe of proven angles compounds over a year.

Why these made the cut: they kill the false positives that make teams chase phantom winners.

Saved you ~2 hours and a few bad conclusions.