Incrementality

Attribution provides valuable, live data for marketing performance measurement and optimization. However, attribution can miss some important details. For example, many users who see your ads may not click, and in-game behavior from those users will not be attributed to your ad placements. Additionally, attribution will miss out on second-order effects like word-of-mouth, and statistics like ROAS have never told the whole story about the true value of your marketing efforts.

Traditionally, marketers have used methods like A/B testing to get deeper insights in to ad performance. However, in an increasingly privacy-focused world, it may not be feasible for some advertisers to share that level of PII with ad networks.

Matched Market Testing

Gamesight has developed a suite of tools to help marketers measure the causal impact of their marketing via geographically targeted lift experiments. Available for ad networks and strategies that allow for regionally targeted campaigns, Gamesight can help you measure the lift of marketing experiments in test regions compared to baseline performance in control regions. Currently, Gamesight supports incremental measurement at the state and country level.

Requiring only aggregate geographic data, incrementality testing does not required the sharing of PII with ad networks, and will even allow for the measurement of out-of-home and on-console advertising.

This is a four step process:

1. Experiment Configuration

Decide which goals you want to measure, what ad network you want to test, and the geographic resolution of your experiment.

2. Market Selection

Based on the length your experiment, the number of desired test markets, and other parameters, Gamesight will help you determine the best test markets for your experiment.

3. Campaign Configuration

Once you have determined how you want to run your experiment, tell Gamesight when and where your campaign will run.

4. Analyze Results

View the results of your experiment, including baseline quality and daily lift deltas.

Additionally, we provide recommendations for getting the most out of Incrementality.

How Does Incrementality Work?

When you want to run an Incrementality experiment, all geographic regions will be broken down into two broad categories:

1. Regions where you target your ad campaign - your test markets
2. Regions where you do not target your ad campaign - your potential control markets

We want to create a baseline from which we can measure your lift. For example, let's say you want target France with an ad campaign, and you want to measure a lift in installs. Unfortunately, once your ad campaign begins it's difficult to know what install counts could have been in France without your ad campaign. Additionally, install counts in your control markets (Japan, the UK, and Canada, for example) may not look exactly like France.

With machine learning, it's often possible to create a weighted average of performance across all other countries that looks very similar to France. We can call this artificial baseline "synthetic-France", and it would be influenced by all control regions, including Japan, the UK, and Canada. By not running ad campaigns in these countries, you can be sure that your new "synthetic-France" is a reasonable baseline from which you can measure lift.

This is called a Synthetic Control Method or SCM (synthetic-France, in our example, being our synthetic control), and these kinds of analyses are very popular in economics and policy research. After all, you can't run an A/B test to measure the impact of environmental policy on consumer behavior! Consequently, SCMs are also useful for marketing lift analyses where PII is restricted and you want to know the causal impact of your ad strategy.