In the context of advertising, a significant transformation is taking place thanks to the growing adoption of artificial intelligence-based advertising campaigns, such as Google’s Performance Max. This paradigm shift has been followed by other platforms such as Meta, LinkedIn and Bing, which have launched similar initiatives, demonstrating the universality of interest in AI-driven solutions in the pay-per-click arena. These campaigns promise a revolutionary approach to online advertising, reducing advertisers’ workload to simple initial inputs such as the URL and desired marketing objective. The promise is that the platform’s artificial intelligence will be able to autonomously manage asset creation, choice of the most appropriate channels, optimization and bidding strategies, leaving advertisers to observe the results achieved.
These advanced advertising strategies are based on complex machine learning and predictive analytics algorithms capable of identifying the most conversion-prone user in real time and delivering highly personalized and timely advertising messages. Google’s official guide for PMax campaigns illustrates the concept of reaching the “right person, at the right time, with the right message and on the right platform.” This is made possible by the ability of AI algorithms to analyze vast datasets of user behavior, accurately predicting the most opportune moments for conversion.
Generative AI plays a crucial role in this process, integrating the ability to create dynamic and personalized content directly within the conversational experience offered by platforms. This approach enables the automatic generation of dynamic advertising assets that adapt in real time to the needs and preferences of target users.
Another key aspect of these AI-based campaigns is their cross-platform nature: they do not require advertisers to predetermine on which platforms to convey their messages, whether it is YouTube, Gmail, Discover or other channels. The algorithm itself determines the most effective channel for each message.
The complexity and effectiveness of the algorithms underlying artificial intelligence-based automated advertising campaigns is a topic of considerable interest and relevance in the context of contemporary digital marketing. These advanced technologies have demonstrated that they can deliver significant results in terms of increasing ROAS and reducing acquisition costs by operating on a model that involves extensive data collection and analysis.
Central to this process is the application of predictive models, which analyze users’ online behaviors, preferences, and paths to anticipate their future actions, including interest in specific products or services.
This real-time assessment of the opportunity for engagement is based not only on the probability of conversion but also on an assessment of the value a single impression can generate for the campaign.
This holistic approach makes it possible to optimize advertising resources, ensuring that the investment is directed toward the users most likely to take the desired action, so important questions have been raised by industry experts regarding their actual ability to generate incremental conversions.
Against this backdrop, recent studies have begun to question the incremental value brought by such campaigns, raising the critical question: do AI algorithms in advertising campaigns really bring measurable and meaningful incremental performance to businesses?
How to achieve incremental performance
Achieving incremental performance in digital marketing is a constant challenge for agencies and practitioners in the field. An experiment conducted by Pace, an agency specializing in Marketing Science located in Segeltorp, Stockholm, offered significant insights in this direction, demonstrating the effectiveness of an approach based on expanding the coverage of the advertising message by shifting the focus of campaigns from conversion to reach.
Traditionally, online advertising campaigns have tended to focus on conversion goals, targeting individuals already considered ready to buy. This approach, while effective in generating immediate sales, tends to make algorithms “lazy,” which limit their scope to a narrow target audience. However, Pace’s experiment revealed that shifting the budget from conversion-focused to reach-oriented campaigns resulted in a 50 percent increase in sales. This result underscores the importance of spreading the advertising message more evenly and inclusively, reaching potential customers beyond those immediately ready to convert.
The reach algorithm, in contrast to conversion-focused algorithms, takes a more statistical and distributed approach, aiming to effectively reach the maximum number of people on target. This results in a higher probability of reaching users who, while not initially considered close to conversion, may be positively impacted by the campaign, thus contributing to significant increases in sales.
Another crucial aspect observed in the experiment concerns the tendency of highly conversion-focused campaigns to favor transactional content and channels, neglecting more informative or narrative brand messages. This choice, although it may seem efficient in the short term, leads over time to target audience saturation and performance degradation, limiting opportunities to generate incremental revenue.
To overcome these challenges, it is critical to adopt a balanced approach that also includes awareness strategies, i.e., initiatives focused on the top end of the marketing funnel. This keeps attention to the brand alive among a broader audience while ensuring efficiency in budget allocation. The implementation of awareness-oriented campaigns requires the identification of a well-defined and highly qualified audience, underscoring the importance of sophisticated segmentation strategies and careful strategic planning to maximize the return on advertising investment.
Evolving privacy regulations and the resulting restriction on the use of third-party audiences have presented companies with the challenge of reconsidering and placing greater value on their first-party data. This data, consisting of all company-owned information collected directly from interactions with customers and users, is a crucial resource for developing targeted and personalized marketing strategies. Although in the past the abundance of third-party data available for purchase may have obscured the value of first-party data, today it becomes imperative for companies to fully leverage this information to build a competitive advantage.
First-party data is not limited solely to users’ personal information but includes a broad spectrum of data generated from interactions with the company’s digital channels, such as website browsing behavior tracked through tools such as Google Analytics. This data, when used ethically and in accordance with current regulations, can turn into powerful levers for optimizing advertising campaigns and advanced targeting.
A key element in the effectiveness of strategies based on first-party data is the implementation of a persistent user identification system, known as “persistent user ID.” This technology overcomes the limitations imposed by the reduced effectiveness of traditional tracking cookies and pixels by providing a more stable and reliable method of recognizing users across different touch points and sessions. Persistent user ID can be derived from a variety of sources, such as email hashes or identifiers generated through probabilistic fingerprinting techniques. In addition, leveraging retrospective user ID can retrieve and associate a user’s past actions, significantly improving targeting accuracy.
The use of this technology facilitates the creation of audience clusters based on specific behaviors, interests, and previous interactions, enabling the use of artificial intelligence algorithms to further enrich and segment this data. This approach not only improves the personalization of advertising campaigns but also enables targeted reach strategies to intercept new, high-potential customers by leveraging first-party information to identify profiles similar to those of existing customers.
Despite perceptions of a supposed end of lookalike-based strategies, companies continue to benefit from evolved methods such as predictive audiences or audience expansions, which represent the natural evolution of the lookalike concept in a context where first-party data play a central role. Platforms such as LinkedIn and Google have introduced advanced algorithms that leverage first-party data to enhance these strategies, demonstrating the effectiveness of such approaches in the context of modern advertising campaigns.
In parallel with the transformation of targeting strategies, it becomes critical to adopt new methods for performance measurement and conversion attribution.
In this context, Marketing Mix Models emerge as an effective solution for evaluating the incremental impact of campaigns, offering companies the ability to accurately measure the return on their marketing initiatives, as demonstrated by the experiment conducted by Pace.
Marketing Mix Model
Marketing Mix Models (MMMs) are a fundamental pillar in the field of marketing analysis, offering a detailed and quantitative view of the effectiveness of different marketing actions on key performance indicators, such as sales. These models address a historical dilemma in marketing, eloquently expressed by John Wanamaker: “Half my advertising budget is wasted, but I don’t know which half.”
Through the use of MMMs, companies seek to unravel this uncertainty by optimizing the allocation of advertising resources to maximize return on investment.
In recent decades, the digital marketing industry has witnessed a significant evolution, driven by the increasing adoption of advanced statistical models. This trend is based on three main pillars: a remarkable expansion of computational capabilities, unprecedented access to vast data sets, often available in near real-time, and an increasing commitment by companies to a culture of data-driven decision making. In this context, MMMs have evolved, integrating established econometric methodologies with innovations from artificial intelligence and machine learning.
The combination of econometric models, originally developed in the 1960s, with ML and AI algorithms represents a significant breakthrough. This hybrid approach not only minimizes the risk of human error, ensuring more accurate and reliable analysis results, but also drastically reduces the time required for data processing and interpretation.
Whereas in the past, analyzing the results of a Marketing Mix Model could take up to six months or even a year, the adoption of ML and AI techniques now allows these analyses to be performed on a monthly basis. This transforms MMMs into extremely powerful analysis tools capable of delivering operational insights in significantly reduced time, and facilitates the adoption of agile marketing strategies, where decisions can be made and adapted quickly in response to changing market dynamics.
How Marketing Mix Models work and what information they give us
Marketing Mix Models represent an advanced and fundamental analytical methodology in the field of quantitative marketing, designed to measure and isolate the effect of various factors on a brand’s sales, thus enabling precise identification of those components that contribute to increased sales. This analytical approach allows companies to understand in detail what portion of sales is attributable to specific marketing actions and what elements are not under their direct control.
The versatility of MMMs makes them valuable tools for analyzing a broad spectrum of variables that influence sales, both in the context of online and offline marketing. They consider not only directly quantifiable marketing activities, such as advertising on various channels, but also those that are less tangible and difficult to attribute to a specific cost, such as promotions and events. Also important is the ability of these models to take into account exogenous factors-such as inflation-that can impact sales regardless of the marketing strategies implemented.
By isolating the effects of each variable, MMMs provide valuable insights into the incremental contribution of specific marketing actions to overall revenue. For example, a company might find that 92 percent of the increase in turnover is attributable to marketing print activities, thus obtaining a quantitative assessment of the effectiveness of that channel. At the same time, the models are able to “cleanse” the effect of non-marketing related variables, providing a clearer and cleaner view of the impact of marketing actions on sales.
A crucial benefit of using MMMs is the ability to determine the optimal allocation of the advertising budget across marketing channels, thus maximizing incrementality. This results in more efficient management of marketing resources, enabling companies to optimize returns on advertising investments and strategically tailor their campaigns according to the effectiveness of each channel.
In summary, MMMs offer:
- A detailed analysis of the contribution of each marketing channel.
- The ability to cleanse the effect of exogenous variables, offering a more accurate picture of the impact of marketing activities.
- Insight into channel saturation, allowing companies to identify when further investment in a given channel may not translate into a proportional increase in sales.
In addition, MMMs are designed with a “privacy by design” approach, as they are based on time-level aggregated data (daily, weekly, or monthly) rather than individual data. This aspect makes them particularly valuable in the current environment of increasing privacy concerns. Their ability to operate without relying on individual-level data allows these analytical models to continue to be leveraged even in the face of the challenges posed by limiting the tracking of user data.The relevance and effectiveness of MMMs are further highlighted by the investment of large technology companies, which have begun releasing open source codes to facilitate the adoption and implementation of these models. Such is the case with Google, which recently released Meridian. Google’s move in making an open source Marketing Mix Modeling (MMM) tool available highlights its understanding that restriction in access to key data, within exclusive ecosystems, hampers advertisers’ ability to accurately assess the effectiveness of digital ads.This effort demonstrates the growing importance of MMMs as an essential tool for analyzing and optimizing marketing strategies in the digital age.
Lift Experiment
In the area of advertising performance analysis, digital marketers use a variety of tools to quantify the effectiveness of their campaigns. In addition to traditional Marketing Mix Models, there are experimental methods such as Lift Tests (Lift Experiments), which offer the possibility to assess the incremental impact of an advertising campaign in a preventive and accurate manner. These experiments, focused on measuring the incremental effect, make it possible to compare the results achieved by the campaign with those that would have occurred in its absence.
The adoption of advanced statistical models is crucial in the implementation of Lift Tests, as it allows the difference in performance between the group exposed to the campaign and the control group to be accurately determined. The selection of these groups is based on geographic targeting criteria, a practice that allows for strategic distribution of campaign visibility. Before starting the experiment, a preliminary analysis is conducted to determine how the variable of interest is distributed geographically, and accordingly decide which geographic areas to include in the treatment sample (exposed to the campaign) and which to exclude (control group).
To exemplify, one might decide to launch an advertising campaign in specific regions of Italy, while others remain without exposure to the campaign. During the treatment period, data on conversions in both the exposed and control groups are collected and analyzed. At the end of the experiment, the application of statistical tests makes it possible to assess the actual significance of the observed difference in conversions, confirming or not the presence of an actual increase.
Lift Tests prove particularly useful for two main reasons: first, they allow the incremental effect of a campaign on specific metrics such as conversions, sales or revenue to be tested on a limited budget before investing more substantial resources.
Second, they provide an opportunity to compare the effectiveness of different advertising strategies, for example by evaluating the incrementality generated by campaigns with reach targets versus conversion-focused campaigns using the same advertising content.
Moreover, the end-to-end approach of geo-experiments ensures a comprehensive assessment of a campaign’s potential, optimizing budget allocation in an effective and privacy-friendly manner, since they are based on aggregated data.
The use of Lift Tests in the context of digital marketing strategies therefore represents a significant evolution in the ability of companies to measure and optimize the effectiveness of their advertising campaigns, driving evidence-based decisions and increasing the return on advertising investment.
As for Lift Tests conducted directly by platforms, focusing exclusively on their channels, they make it complex to compare the performance of multiple campaigns and different channels. In addition, these tests are susceptible to technological updates and changes in individual platform policies, which can compromise the accuracy of the results. As a result, it is advisable to rely on external partners for these tests.
