MTA, MMM, or Incrementality Testing? How to Measure and Optimize Advertising Campaigns with an AI-driven Framework

During my last call with an experienced Head of Digital, I was asked the most important question right at the end: “What is the most effective method to measure and optimize our marketing activities? Should we favor technologies based on Multi-touch Attribution or Marketing Mix Models?” My answer was immediate: “Both, and more.” Here’s why.

The world of advertising is undergoing significant transformations, such as the announced privacy restrictions and the emergence of new communication channels. These developments greatly complicate the task of answering one of the most pressing questions for every CMO: “What is the actual return on investment of my advertising campaigns? How can I optimally allocate my investments across different channels and marketing activities?” Currently, the market offers several tools based on three main approaches:

• Multi-touch Attribution (MTA)
• Marketing Mix Models (MMM)
• Incrementality Testing (IT)

However, each tool provides only a partial answer to these questions, and industry experts agree that the combined use of these techniques represents the most effective strategy for measuring and optimizing marketing activities. This integrated approach, often referred to as “Triangulation,” is considered essential for an accurate assessment of marketing performance.

Let’s examine the strengths and limitations of each method, highlighting how an effective combination of MTA, MMM, and IT can offer the most promising solution to successfully tackle current market challenges.

Triangulation in Marketing Measurement

Multi-touch Attribution (MTA)

Multi-touch Attribution (MTA) is the standard method for measuring the effectiveness of online marketing. This technique relies on the use of browser cookies, UTM codes, tracking pixels, and other similar identifiers to track the marketing sources that lead to customer conversions.

Strengths: MTA allows for very detailed and almost real-time measurement and optimization, essential during the more tactical phases of campaign optimization. This approach facilitates quick adaptation to market dynamics and emerging trends, effectively addressing crucial questions like: “How should I distribute the budget across my campaigns and ad groups?” and “Which copy or creative asset works best?”

Challenges and Limitations: When used in isolation and not in combination with other tools, MTA is limited: it focuses on specific objectives and risks losing sight of a comprehensive understanding of marketing investments across the entire funnel. Moreover, with the gradual phasing out of cookies, the accuracy of these techniques is likely to decline in certain contexts.

Marketing Mix Models (MMM)

Marketing Mix Models (MMM) are statistical modeling techniques that analyze historical data to determine which marketing mix channels contribute to generating sales, in order to reallocate the budget to the most effective channels.

Strengths: MMM allows for the measurement and optimization of investments across both offline and online channels through a holistic and long-term approach. These models answer complex questions such as: “What was the return on my TV campaigns on sales?”, “What was the impact of promotions?” and “How much budget should be allocated next quarter to achieve my revenue goals?”

Challenges and Limitations: Like all models that describe complex phenomena, MMMs require a lot of data to provide very accurate estimates. One of the basic principles of Machine Learning – the Curse of Dimensionality – tells us that as the model’s parameters increase, the amount of data should grow exponentially. Unfortunately, this is often not possible, as brands, for various reasons, do not have large amounts of good quality historical data. Consequently, an accurate MMM can require significant data pre-processing and model fine-tuning by Data Science teams to make the most of the limited data available.

Incrementality Testing (IT)

Conversion Lift Studies and Incrementality Tests are structured experiments to quantify the effectiveness of marketing strategies on sales. Methods include Geo-testing and holdout tests. A typical example of these tests involves activating a marketing channel in a selected set of regions or countries, then comparing the increase in sales in these areas with those where the channel was not activated.

Strengths: CLS allows for targeted tests that provide answers to unresolved questions from previous methods, offering essential generalizable information for future strategy definition. Through these tests, it is possible to empirically and accurately validate the effectiveness of marketing actions.

Challenges and Limitations: Conducting accurate tests requires the application of a rigorous scientific method that appropriately balances test reliability and associated risks. In other words, it is necessary to define tests with a sufficiently large and representative sample over a sufficient period while minimizing the “waste” of budget that results from pursuing clearly sub-optimal strategies on a portion of advertising campaigns.

How Do MTA, MMM, and IT Integrate?

Despite the evident strengths of Multi-touch Attribution (MTA), Marketing Mix Models (MMM), and Testing (IT), none of these methods, if used individually, can serve as a definitive solution for strategic decisions. Often, in online discussions, there is talk of “Triangulation,” treating these methods as separate sources from which to draw for a more robust final estimate, mediating between the various suggestions and measurements they offer.

However, we propose an alternative that surpasses this vision: the synergistic and coordinated use of these techniques, enriched by a scientific approach and supported by artificial intelligence, can indeed lead to revolutionary transformations in the sector.

Interaction between MTA, MMM, and IT

Interaction between MTA and MMM: The mutual influence between MTA and MMM is evident. Continuous campaign and ad group optimization driven by MTA significantly impacts the performance of each channel, thus also influencing the long-term forecasts of MMM. Conversely, budget strategies based on MTA need to be aligned with the strategic plans and high-level directives established through MMMs.

At AD cube, we have elevated this concept: we use impact estimates of each channel at various funnel levels to achieve balanced campaign and ad group management and optimization across the entire funnel.

Integration of MTA and IT: Designing tests in marketing is not simple, and relying exclusively on GEO-testing to evaluate the impact of specific channels and performance is not always practical. Therefore, it is often necessary to select test groups and control groups based on other characteristics and time series analyses obtained through MTA approaches.

Integration of IT and MMM: “What is the real impact of a certain channel on sales?” This fundamental question, often asked during Marketing Mix Models (MMM) analysis, does not always find accurate answers, especially when the available data on some channels is limited, as in the case of short-term activities or those related to special events and promotions. In these contexts, Testing methods can be strategically adopted to quickly acquire data on the effectiveness of such channels. For example, implementing a Geo-testing test can reveal the impact of a channel on a representative sample of users. These results can then be generalized to provide a more accurate estimate of the effect on the entire population, thus compensating for the intrinsic limitations of MMMs due to data scarcity.

At AD cube, we use Active Learning algorithms to identify the areas of greatest uncertainty in MMMs, where additional data needs to be acquired through tests. We then apply Transfer Learning techniques to incorporate this new information into our MMMs, obtaining more precise and reliable estimates. This approach not only improves our ability to predict and optimize performance but also supports a continuous cycle of learning and strategic adaptation.

The triangle composed of MTA, MMM, and Testing, powered by advanced Artificial Intelligence approaches, constitutes an excellence framework in the modern advertising landscape. It provides increasingly precise answers to complex questions, allowing for the development of flexible and dynamic optimization strategies with a long-term perspective aligned with the overall business objectives.