Advertisers will spend about $300 billion for digital ads in the US in 2024, a number that’s growing by roughly 11% every year. Each of those dollars is spent with one or more KPIs in mind. As a CMO or performance marketer, you know that every dollar counts. But the burning question remains: How effective is your advertising spend, and how can it be optimized for better efficiency?

While digital marketers possess sophisticated tools to segment audiences, the key to efficiency lies in knowing the incremental value each ad dollar buys in each segment. While incrementality can’t be measured with perfect fidelity, we can approximate it to a useful degree. Using such estimates of the incremental return on ad spend per segment, advertisers can gain efficiency by reallocating resources from less effective segments to those with higher returns.

The problem marketers face is that good estimates of incrementality are hard to come by. Many advertisers instead base their decisions on average performance, as measured through rules-based attribution models such as last-touch, linear, time-decay, and U- or W-shaped. All of these models share two common problems. First, they give credit to touchpoints based on the order they come in rather than the value they bring. Second, even if they measure average performance accurately, the best performer on an average basis could be mediocre or worse on an incremental basis. The main benefit of these rules-based models is that they’re easy to understand and implement.

A step above the rules-based models, we have measurements based on sophisticated analyses of observed performance. Examples of these include several models based on the work of Lloyd Shapley. These models involve defining user-paths as a sequence of advertising touchpoints. Credit is then allocated based on a comparison of paths with vs without the specific ad segment being analyzed. While considerably more complicated than rules-based models, these have the advantage of being measurements of performance based on data. A disadvantage of these techniques is that they can’t distinguish between ads that are driving value (causation), and ads that are present when value is created, but aren’t driving value creation (correlation).

Finally we come to active experimentation. In this category, we divide the population eligible to see ads into test & control groups. We then apply some treatment to the test group, and compare the performance metrics of the two groups to measure the effect of our treatment. While this can be used to measure many things, our focus today is on incremental performance. For this analysis, the test group should acquire traffic either more or less aggressively than the control group. We then analyze the difference in cost & KPI between the two groups to estimate incrementality, or the marginal cost per KPI. This technique gives advertisers the best information to use in strategic decision-making and budget allocation.

In finance, the term ‘arbitrage’ is used to describe opportunities to profit by simultaneously buying and selling the same asset at different prices. With good incrementality measurements, advertisers can effectively realize arbitrage too. Using ImprezzAI’s incrementality measurements, advertisers would know the marginal cost per KPI their various ad segments bring in. Then, they’d benefit from arbitrage by moving spend from underperforming segments into overperforming ones until all are equally efficient. This would let advertisers buy more KPI for the same budget, or spend a larger budget at the same unit economics.

We invite your questions and feedback. Share your thoughts with us, and let's explore how we can make your advertising efforts more impactful.