Google Ads campaign experiments let you test a meaningful change against a controlled version before applying it broadly. This guide shows how to compare experiment methods, define a useful hypothesis, estimate the traffic and conversion requirements, monitor interference, and turn the result into a documented decision about bids, budgets, or targeting.
Overview
A campaign experiment is valuable when a proposed change could materially affect performance but the likely outcome is uncertain. Examples include testing a target CPA or target ROAS adjustment, changing match-type coverage, reallocating budget, modifying location targeting, or introducing a new audience signal.
The core idea is simple: preserve a control and expose a comparable portion of traffic to a treatment. The control represents the existing setup. The treatment contains one planned change. After a defined observation period, compare the results using metrics that match the original business decision.
Not every change requires a formal experiment. A broken conversion tag, disapproved ad, tracking error, or obvious negative keyword should be fixed rather than tested. Experiments are most useful for decisions where both options are plausible and the cost of being wrong is meaningful.
Campaign experiments, drafts, split tests, and before-and-after analysis
| Method | Best use | Main strength | Main limitation |
|---|---|---|---|
| Campaign experiment | Testing a defined change against an existing campaign | Creates a planned control-and-treatment comparison | Setup, eligibility, and traffic controls depend on the campaign and platform configuration |
| Draft | Preparing or reviewing changes before launch | Allows collaboration and quality checks | A draft alone does not produce performance evidence |
| Split test | Dividing comparable traffic between alternatives | Reduces some timing and seasonality problems | Requires enough volume and clean separation between variants |
| Before-and-after analysis | Reviewing a change that has already been rolled out | Fast and practical when a controlled test was not possible | Results may be influenced by seasonality, demand shifts, competitors, or other account changes |
Google Ads features and experiment controls can change over time. Check the current interface and eligibility requirements before launch, and treat this framework as the decision process around the platform feature rather than as a guarantee that every campaign can use every experiment option.
How to estimate
Before creating the test, estimate whether the campaign can generate enough observations to support a useful decision. You do not need false precision. You do need to understand the relationship between traffic, conversion volume, expected effect, and test duration.
1. Define the decision
Write the decision as an action: “If the treatment improves qualified lead cost without reducing lead quality, adopt it.” This is stronger than “Test a new bid strategy,” because it states what evidence will change the plan.
2. Write one primary hypothesis
Use this format:
Because [reason], changing [one variable] from [control state] to [treatment state] will change [primary metric] by approximately [decision threshold] without worsening [guardrail metric].
For example: “Because the campaign is limited by budget during high-intent hours, increasing the daily budget will increase qualified conversions while keeping cost per qualified conversion within the approved range.” The threshold is a planning assumption, not a forecast.
3. Estimate conversion volume
A basic planning estimate is:
Expected conversions per variant = eligible clicks per day × conversion rate × variant traffic share × test days.
If a campaign receives 100 eligible clicks per day, converts at 4%, sends 50% of traffic to the treatment, and runs for 28 days, the treatment estimate is 100 × 0.04 × 0.50 × 28 = 56 conversions. The control has a similar planned volume if allocation and delivery remain comparable.
This calculation is an estimate, not a promise. Delivery can vary, conversion reporting can be delayed, and a bid or budget change can alter click volume and conversion rate. Use a range when possible: calculate a conservative, expected, and optimistic scenario.
4. Estimate spend and exposure
For a simple planning model:
Expected spend per variant = eligible clicks per day × average cost per click × variant traffic share × test days.
Using 100 clicks per day, an assumed average cost per click of 2 currency units, a 50% allocation, and 28 days gives 2,800 currency units per variant. Replace the assumed cost with the campaign’s recent comparable-period value and label it clearly. A bid test may change that cost, so include a contingency in the approved budget rather than treating the estimate as fixed.
5. Set a stopping rule
Choose the minimum observation period, minimum conversion volume, and maximum acceptable loss before the test begins. Avoid stopping immediately after a short-term spike or dip. At the same time, do not let a test run indefinitely when the treatment is clearly damaging an important guardrail.
Inputs and assumptions
Record the following inputs in an experiment brief or campaign tracking document:
- Campaign and scope: campaign name, markets, devices, networks, conversion actions, and the exact entities included.
- Control: the current bid, budget, targeting, keyword, ad, or landing-page configuration that remains unchanged.
- Treatment: one material change, with the old and new settings recorded precisely.
- Traffic allocation: the planned control and treatment shares. Confirm how the platform implements the split and whether allocation is available for this campaign type.
- Primary metric: the metric tied directly to the hypothesis, such as qualified conversion rate, cost per qualified conversion, conversion value, or return on ad spend.
- Guardrails: metrics that must not deteriorate beyond an agreed limit, such as conversion volume, lead quality, impression share, or spend.
- Conversion definition: identify which conversions count and whether reported values may arrive after the click or after the test ends.
- Test dates: planned start, minimum run time, review date, and any dates that should be excluded because of known promotions or tracking changes.
- Interference controls: related campaigns, shared budgets, automated rules, scripts, keyword changes, creative updates, or landing-page releases that could affect both variants.
Keep the experiment narrow. If you change the bid strategy, budget, keyword set, ad copy, and landing page simultaneously, a result may be useful operationally but cannot explain which change caused it. For keyword or targeting tests, review search terms and maintain a documented negative keyword list so irrelevant demand does not distort the comparison.
Use consistent attribution and naming. A clear UTM naming convention helps connect platform results with analytics and CRM outcomes. Before launch, check imports, conversion actions, and settings with an advertising import checklist when changes involve another platform.
Worked examples
Example 1: Testing a budget change
A search campaign receives 120 eligible clicks per day, has an assumed 3% conversion rate, and averages 1.80 currency units per click. The proposed treatment raises the daily budget; the control keeps the existing budget. With a 50/50 allocation and a 28-day test:
- Estimated conversions per variant: 120 × 0.03 × 0.50 × 28 = 50.4.
- Estimated spend per variant: 120 × 1.80 × 0.50 × 28 = 3,024 currency units.
- Primary metric: qualified conversions per unit of spend.
- Guardrail: cost per qualified conversion must not exceed the approved limit.
The test should not be judged only on total conversions. A larger budget can produce more conversions simply by buying more traffic. Compare efficiency and quality as well as volume, and check whether the treatment actually received comparable delivery.
Example 2: Testing a bid strategy change
Suppose a campaign has stable conversion tracking and enough recent data for a planned comparison. The control retains its current bidding approach; the treatment uses the proposed approach. The hypothesis is that the treatment will improve conversion value per spend by at least the team’s decision threshold without reducing qualified conversion volume.
During the test, avoid making unrelated keyword, ad, landing-page, or budget changes. Review spend pacing, conversion lag, search terms, and delivery each day for anomalies, but do not optimize the treatment toward the result while the control remains unchanged. At the end, compare the same attribution window and document whether the evidence is strong enough to adopt, reject, or retest.
Reusable experiment brief
Decision: What will we do if the treatment wins, loses, or remains inconclusive?
Hypothesis: What change is expected to affect which metric, and by how much?
Control: What remains unchanged?
Treatment: What is the single planned change?
Allocation and dates: How will traffic be divided, and when will the test run?
Primary metric: What decides the outcome?
Guardrails: What loss is unacceptable?
Interference plan: Which other changes are paused or logged?
Result: What happened, with the comparison period and attribution rules stated?
Decision owner: Who approves the next action?
For related ad-copy tests, use the ad copy testing playbook. For bid-specific assumptions, review the bid strategy optimization guide.
When to recalculate
Recalculate the estimate whenever an input that affects exposure, cost, or measurement changes. That includes a material shift in click volume, average cost per click, conversion rate, conversion value, budget, traffic allocation, targeting, or attribution settings. Recalculate before extending a test into a new season, promotion, market, or device mix.
Revisit the framework when Google Ads changes experiment creation, traffic-split controls, reporting, conversion requirements, or the campaign types eligible for testing. Platform screenshots and step-by-step instructions should be checked periodically because interface details age faster than the underlying decision method.
At the end of each test, take one of three actions: adopt the treatment when it meets the primary target and guardrails; reject it when the downside is clear; or retest when the result is inconclusive, interference occurred, or the test did not collect enough comparable evidence. Record the conclusion in the campaign’s experiment log, then add the next question to a prioritized testing backlog. This turns isolated Google Ads split tests into a repeatable PPC experiment framework for bid strategy optimization, budget pacing, and targeting decisions.