Google Ads Experiments: How to Test 'Crazy Ideas' Without Killing Your ROI (2024)

"I think we should change the bid strategy to Target CPA." "I feel like Broad Match might work now."

In digital marketing, "I think" is a dangerous phrase. It gets people fired. You should replace it with "Let's test."

Google Ads has a built-in feature called Experiments (formerly Drafts & Experiments) that allows you to act like a scientist. You can take a running campaign, slice off 50% of the traffic, and effect change ONLY on that slice.

If the idea fails? You only hurt 50% of the traffic, and you can shut it down instantly. If the idea wins? You just unlocked 20% more growth.

This guide covers the Protocol for running valid experiments.

---

Part 1: The "Cookie-Based" Split

This is the most important technical concept.

When you launch an experiment, Google uses a Cookie-Based Split.

• User A is assigned to the Control group (Original Campaign).
• User B is assigned to the Experiment group (Test Campaign).

If User A searches for your keyword 10 times over 2 weeks, they will only see the Control version. This is critical because it prevents Data Contamination. You are testing complete user journeys, not just random impressions.

---

Part 2: The "Big Three" Growth Tests

We run experiments constantly. These are the three most common protocols we use to scale accounts.

Test 1: The Bidding Strategy (The "Brain" Transplant)

Goal: Determine if Smart Bidding beats Manual, or if tCPA beats Max Conversions.

• Control: Manual CPC (Current).
• Experiment: Target CPA (set to your historical average).
• Hypothesis: tCPA will lower CPCs and maintain conversion volume, increasing ROAS.
• Duration: 4 Weeks (Smart Bidding needs 2 weeks just to learn).

Test 2: The Broad Match Expansion (The "Net" Cast)

Goal: Find more volume without tanking efficiency.

• Control: Phrase Match Only.
• Experiment: Add Broad Match keywords (using the exact same root words).
• Hypothesis: Broad Match + Smart Bidding will find cheaper conversions in queries we didn't think of.

Test 3: The Landing Page Duel

Goal: Test a new offer or page layout.

• Control: Send traffic to domain.com/landing-page-a
• Experiment: Use "Find and Replace" in the experiment settings to change Final URL to domain.com/landing-page-b.
• Hypothesis: A shorter form on Page B will increase Conversion Rate by 15%.

---

Part 3: The Protocol (Step-by-Step)

Here is how to set it up without breaking anything.

1. Select Campaign: Go to the campaign you want to test.
2. Create Custom Experiment: Click the "Experiments" tab on the left → "All Experiments" → Blue "+" Button → "Custom Experiment".
3. Choose Base: Select "Search" (or Display) and pick your base campaign.
4. Name It: EXP - [Date] - [Hypothesis]. E.g., EXP - Jan 2026 - Test tCPA.
5. Configure Experiment:
   • Suffix: Google adds _Experiment to the campaign name.
   • Description: Write down exactly what you changed so you don't forget in 3 weeks.
6. Schedule:
   • Start Date: Tomorrow (Always start experiments at midnight).
   • Split: 50% (Standard). If you are very nervous, you can do 30% Experiment / 70% Control, but it will take longer to get significance.

CRITICAL STEP: Once the experiment is created, go into the Experiment Campaign settings and make the change.

• Change the Bidding Strategy.
• Change the Keywords.
• Change the URLs.
• Do not touch the Original Campaign.

---

Part 4: Analyzing the Results (Statistical Significance)

After 30 days, opening the Experiment tab is like opening a Christmas present.

Google gives you a scorecard comparing Control vs. Experiment. Look at the stars (Confidence Intervals).

• Metric: Conversion Value / Cost (ROAS).
• Result: +24% (Blue Star).
• Meaning: The Experiment won with 95% statistical confidence.

Decision Time:

1. Apply: Click "Apply to Base Campaign". This overwrites your original campaign with the experimental settings. The experiment ends.
2. Convert: Click "Convert to New Campaign". This keeps both running (rarely recommended unless you want to target different audiences).
3. End: If it lost, just click "End Experiment". The traffic flows back to the original campaign as if nothing happened.

---

Warnings & Pitfalls

1. The "Learning Phase" Shock: Even an experiment has a learning phase. The first 3-5 days of the Experiment arm will usually be bad. Do then panic. Evaluate data from Day 7 to Day 30.
2. Budget Constraints: If your original campaign was "Limited by Budget", splitting it in half might starve both sides. Ensure you have enough data (at least 30 conversions a month total) to split.
3. Testing Too Much: Change ONE variable. If you change the Bidding Strategy AND the Headline, and results improve, you won't know which one did it.

Summary

Experiments act as your R&D Department. They allow you to be aggressive with innovation while being conservative with risk. If you aren't running an experiment right now, you aren't learning.