GeoLift Use Cases and Best Practices

Attribution shows which marketing touchpoints received credit for a conversion. GeoLift answers a different question:

What changed because the advertising ran?

GeoLift compares geographic regions that receive the advertising intervention with similar regions that do not. For an active campaign, advertising continues in the test regions and is paused or reduced in the holdout regions. For a new or paused campaign, advertising launches in the test regions while the comparison regions remain unexposed.

Triple Whale then estimates what would likely have happened without the advertising. The difference between the actual result and that estimate is the campaign’s incremental impact.

Use GeoLift when you need evidence to support a specific marketing decision, such as:

A useful GeoLift experiment begins with a decision you need to make.

Examples include:

A focused question makes it easier to choose the campaigns, primary metric, budget, duration, and geographic design.

Avoid combining unrelated questions in one experiment. If you test several different tactics together, the result will reflect their combined impact and may not tell you which tactic caused the change.

The question

Is an active campaign or channel creating additional revenue or acquisitions?

A campaign can report strong attributed performance while receiving credit for purchases that would have happened anyway. GeoLift helps you determine how much additional value the campaign actually created.

During Test Design, select Active Campaigns.

Choose the campaigns you want to measure and select the primary metric that matches your decision:

The selected campaigns continue running in the test regions and are paused or reduced in the holdout regions.

A reliable positive result supports continued investment in the tested activity. An inconclusive result means the test did not produce enough evidence to separate the campaign’s effect from normal variation.

An inconclusive result does not prove that the campaign had no impact. Review the test’s spend delivery, duration, geographic design, and result range before deciding what to do next.

The question

Will a campaign or channel that is not currently active create additional sales when it launches?

This can help you evaluate a new channel, relaunch a paused campaign, or test activity for which you do not yet have enough historical performance data.

During Test Design, select New or Paused Campaigns.

The campaign launches in the selected test regions while the comparison regions remain unexposed to the campaign.

Choose the primary metric that represents the result you want the new activity to create.

Use the final lift and incremental-efficiency metrics to decide whether the activity deserves continued investment.

Apply the result to the campaign, budget, markets, and dates you actually tested. A successful test at one spending level does not guarantee that performance will remain the same as the campaign scales.

The question

Is branded search creating additional sales, or capturing customers who were already planning to purchase?

Branded search often appears highly efficient in attribution reports because it is close to the final purchase. However, customers may search for the brand after encountering another marketing channel or after already deciding to buy.

GeoLift can test whether branded search is producing additional revenue or primarily capturing existing demand.

Select the eligible branded search campaigns you want to test.

The campaigns continue running in the test regions and are paused or reduced in the holdout regions. GeoLift then compares revenue or acquisitions between the geographic groups.

Keep non-tested marketing activity as consistent as possible during the experiment.

If branded search produces reliable incremental lift, the campaign is contributing value beyond capturing existing demand.

If incrementality is lower than attributed performance, the campaign may be receiving credit for demand created elsewhere. Use the GeoLift result alongside attribution to make a more informed budget decision.

This experiment measures the effect on your selected revenue or acquisition metric. It does not directly prove whether organic search absorbed every click or impression that branded search would otherwise have received.

The question

Do upper-funnel campaigns contribute to downstream revenue or acquisitions?

Awareness and consideration campaigns can influence customers before they are ready to purchase. Their impact may not appear clearly in click-based or last-click attribution.

Select the eligible upper-funnel campaigns you want to evaluate.

Keep those campaigns running in the test regions while withholding them from the holdout regions. Continue the rest of your marketing activity consistently across both groups.

Use Revenue or Acquisitions as the primary metric so the experiment measures whether the upper-funnel activity ultimately affected business outcomes.

A reliable positive result indicates that regions exposed to the upper-funnel activity generated more revenue or acquisitions than they would have without it.

An inconclusive result may mean that the effect was too small to detect with the selected budget, duration, or geographic design. It does not automatically mean the upper-funnel activity had no value.

The question

Does controlled experimental evidence support what attribution or marketing mix modeling says about a channel?

Attribution, MMM, and GeoLift answer different questions:

Choose a channel, campaign group, or tactic for which attribution or MMM has produced a decision-relevant finding.

For example, you may want to test:

Test one clearly defined activity whenever possible. If multiple channels are included together, the result reflects their combined impact.

Compare the GeoLift result with the attributed or modeled result for the same activity and period.

If attribution reports a much higher return than GeoLift, attribution may be over-crediting the activity. If GeoLift finds stronger impact than attribution reports, the attribution model may be missing part of the activity’s contribution.

Use the experiment to calibrate future decisions, but keep its scope in mind. One test does not permanently define a channel’s value under every budget, market, or season.

The question

Does advertising on one channel contribute to sales somewhere else?

A customer might see an ad on Meta or Google and later purchase through another storefront or marketplace. Traditional attribution may not connect those interactions, particularly when the purchase happens in an environment where customer-level tracking is unavailable.

GeoLift can evaluate a halo effect by comparing geographic order outcomes between regions where the advertising runs and regions where it is withheld.

The order source being measured must provide sufficient geographic order data. The relevant advertising spend and order data must also be available to the analysis.

For example, a business may want to test whether advertising aimed at its direct-to-consumer storefront also contributes to marketplace sales.

A reliable positive result indicates that the tested advertising contributed to sales beyond the destination most visible in attribution.

Use this information when evaluating the channel’s total business impact rather than relying only on the sales directly attributed to the campaign.

A useful question can still produce an inconclusive result if the experiment does not have enough signal.

Before launching, review Triple Whale’s recommended:

Campaigns with very low spend can be difficult to measure because their impact may be smaller than normal changes in revenue or conversions.

If the planned experiment is not sufficiently powered, consider:

Use the recommendations shown during setup rather than choosing a holdout percentage or duration from a generic example.

GeoLift works best when the advertising intervention is the main meaningful difference between the test and holdout regions.

Other major changes during the experiment can make it difficult to determine what caused the result.

While the experiment is running, avoid:

For example, if you substantially increase the tested budget halfway through the experiment, the final result may reflect two different spending levels rather than the original plan.

When possible, avoid running the experiment across:

These events can add noise or create another explanation for the observed result.

If a major event is planned, account for it before scheduling the experiment. If an unexpected event occurs, document what happened and when so it can be considered when interpreting the result.

You do not need to stop normal business operations during a GeoLift experiment.

Routine activity is less likely to interfere when it:

Regular email, SMS, website, and merchandising activity can continue when it remains consistent across the geographic groups. Avoid introducing a new region-specific promotion or campaign that affects only one group.

Ending an experiment early reduces the amount of evidence available and can make a real effect more difficult to detect.

Let the experiment complete its planned duration and cooldown period before interpreting the final result.

An inconclusive result does not mean the tested advertising had no effect.

It means the experiment could not separate the estimated effect from normal variation with enough confidence.

Common reasons include:

Before repeating an inconclusive experiment, review the test configuration and any events that occurred during the measurement window. A longer, larger, or cleaner follow-up test may produce a clearer answer.

A GeoLift result applies to the activity and conditions included in the experiment:

Do not automatically apply one result to every campaign on the channel.

Incrementality can change as budgets, creative, audiences, competition, seasonality, and customer behavior change. Use each experiment as a strong decision signal, then continue validating important assumptions over time.

For help interpreting a completed experiment, see Reading Incrementality Test Results.