Skip to the lesson
0%

Experiments and Testing

Expert 11 min read Lesson 14 of 15

Changing a bid strategy, a landing page or a match type on a live campaign and comparing before and after mixes the change with seasonality and noise. Experiments split traffic so you can see the real effect.

Campaign experiments

Create an experiment from an existing campaign, change one thing in the trial arm (bid strategy, broad match, landing page, assets), and split traffic or budget, often 50/50. Google reports the difference and whether it is statistically significant.

Design rules

  • Test one variable at a time.
  • Run long enough for enough conversions in both arms: usually four to six weeks for lead generation.
  • Decide the success metric (CPA, conversions, value) before starting.
  • Do not edit either arm during the test.

What to test first

Start with changes that affect cost per result most: landing page versions, bid strategy and target, broad match with Smart Bidding, and new offers in ads. Small headline wording tests rarely move business results on their own.

A clean experiment
1HypothesisBroad + tCPAlowers CPA2Split 50/50Control vs trial3Run 4 to 6weeksNo editsmid-test4ReadsignificanceOn CPA, notclicks5Apply ordiscardRoll out thewinnerOne change, split traffic, decide on a pre-set metric

Pro tips

  • Write the hypothesis and success metric in the experiment name.
  • Keep a log of every test and its result for future reference.
  • Apply the winning arm through the experiment tool so settings carry over cleanly.

Common mistakes

  • Ending a test after a few days because early numbers look good.
  • Changing budgets in one arm during the test.
  • Testing three things at once and not knowing which mattered.
Summary

Use campaign experiments to test one change at a time with split traffic, a pre-set metric and enough time, then apply only clear winners.

Quick quiz

1. Why use an experiment instead of before and after?
2. How many variables should a test change?
3. When should the success metric be chosen?

Stuck on something in this lesson? Ask our team on WhatsApp. We answer course questions for free.

Ask on WhatsApp