Did the ads cause the sale, or just claim it? Choose the strongest test your data can support

An advertising report can correctly show that a customer clicked or viewed an ad before buying. That still does not prove the ad created the sale. Attribution assigns credit among observed touchpoints. Incrementality asks a harder question: how many customers would not have bought without the advertising?

Use the strongest evidence your business can support, then limit the claim to that evidence. A small account with 20 completed sales should not imitate the certainty of a powered experiment. A large account should not settle for platform credit when a controlled lift study is practical.

Start with the decision, not the test

Write one decision before choosing a method:

> If the campaign creates additional completed customers below our allowable acquisition cost, we will continue it. If the result is unresolved, we will preserve the current limit and improve the test rather than declare a winner.

Name the campaign, customer outcome, test window, allowable cost, and exact budget action. "Prove marketing works" is not testable. "Decide whether to keep $4,000 of monthly nonbrand Search spend" is.

Keep the outcome close to business value. A completed, nonrefunded order is stronger than a page view. A qualified opportunity is stronger than a form start, but it still is not collected revenue. Use the report-reconciliation guide to make the platform, analytics, CRM, and finance definitions explicit before the test.

Use this evidence ladder

Level 1 is useful for campaign operations. It helps explain customer paths and gives bidding systems a success signal. Do not rename it "incremental revenue." The ROAS guide also explains why attributed revenue is not contribution profit.

Level 2 can be an operational warning. If spend stops during a tracking outage and completed sales remain unchanged, the result deserves investigation. It is not a clean experiment. Price changes, weather, seasonality, competitor activity, sales staffing, promotions, inventory, and other campaigns can all move the outcome.

Level 3 is a screening design for businesses that have separable markets and stable data. It becomes credible only when the comparison units behaved similarly before the test, other marketing stayed consistent, movement between markets was limited, and the analysis accounts for uncertainty. Meta's open-source GeoLift documentation uses synthetic controls and power analysis for this reason. A simple spreadsheet comparison is not equivalent to GeoLift.

Level 4 provides the strongest evidence in this ladder. The platform or test system withholds advertising from a control group and compares it with a treatment group. Google and TikTok describe this basic design in their current Conversion Lift documentation. Access, data volume, conversion quality, cost, and study power still determine whether the result can answer the question.

Complete the seven-gate readiness card

If outcome, volume, or separation fails, do not force a lift claim. Stay at a lower evidence level, improve the underlying data, or defer the test. The conversion-tracking audit can establish whether the outcome is captured consistently before it becomes a study input.

Google's current user-based Conversion Lift setup lists at least 1,000 observed conversions and a $5,000 campaign budget, plus a compatible conversion action. It currently lists Display, Search, Video, Demand Gen, App, and Performance Max among supported campaign types, with exclusions. Google also says study power depends on the conversion actions, budget, duration, holdback, historical data, and expected lift. This means self-serve access is broader than it was, but it is not a fit for every small account. Verify the actual account rather than assuming eligibility from an announcement.

TikTok currently describes Conversion Lift as a managed service for eligible accounts, with minimum spending based on the objective and scope. Meta's GeoLift documentation says people-based experimentation is preferred where feasible and presents GeoLift as a research tool, not a Meta advertising product. None of these options turns sparse or inconsistent business data into a reliable answer.

Record contamination before launch

A control group is useful only if the treatment difference remains understandable. Keep a short register:

Google recommends caution with creative and audience changes during a running study because they make the learning harder to interpret. Meta's GeoLift best practices similarly call for matched markets, sufficient stable history, complete data, a test that covers the purchase cycle, and consistent local and national marketing.

Calculate a directional matched-market screen

The following example checks whether a deeper test is justified. It is not a powered causal study.

Suppose the test market had 200 completed orders during the baseline period and 230 during the advertising period. A comparison market had 250 baseline orders and 275 during the same advertising period. The comparison market grew 10%, so a simple counterfactual for the test market is 200 x 1.10, or 220 orders.

Control growth factor = control test-period outcome / control baseline outcome

Directional counterfactual = test baseline outcome x control growth factor

Directional incremental outcome = actual test outcome - directional counterfactual

The screen estimates 230 - 220, or 10 additional completed orders. If media cost $4,000 and setup and analysis cost $500, the full test cost was $4,500.

Directional incremental customer cost = full test cost / directional incremental completed customers

$4,500 / 10 = $450

If the ad platform credited itself with 34 orders, do not call all 34 incremental. The screen estimates 10 additional orders under its assumptions. It also does not prove that exactly 10 were caused by ads. A poor market match, a local event, small counts, or ordinary variation could change the estimate.

Compare the $450 directional cost with the business's allowable cost per completed customer. Use the profitability calculator for the contribution boundary, not a generic benchmark. If the uncertainty could move the decision across that boundary, the honest result is unresolved.

Read the result without overcorrecting

A lift result belongs to the tested campaign, population, outcome, dates, budget, and surrounding marketing. It does not automatically answer whether the entire channel should be stopped. A channel can influence later direct, organic, branded-search, or sales activity that a narrow study does not explain.

Record four numbers separately:

1. Platform-attributed customers.
2. Verified completed customers in the business system.
3. Estimated incremental customers with an uncertainty range.
4. Full cost, including media, creative, setup, analysis, and management.

Then choose one action:

Do not turn "unresolved" into "the ads failed." It means the test could not distinguish a useful effect from noise at the level the business needed. Keep the spending ceiling from the Google budget authority guide, repair the design, and test again only when the answer can change a real decision.

If you want help defining the outcome, auditing eligibility, and designing a bounded test without confusing platform credit with causal proof, review Ocean Media's paid advertising service.

For general education only. Platform details and examples can change. Verify important information before acting. Ocean Media Marketing is not responsible for decisions based solely on this resource. Suggest a correction