Marketing Mix Modelling: A View From Metas Rasheeqa Jacquesson

Rashika Jaxon, MEA Marketing Partner at Meta.

Today's digital landscape gives marketers access to a seemingly endless amount of information. But are they using this data to its full potential? For many organizations, this is not the case. Over the past decade, direct marketers have used various data sources, such as cookies and mobile device identifiers, to measure ad performance. This data is increasingly difficult to access, making measurement systems such as channel-level reporting and cross-channel attribution less effective.

Digital brands accustomed to rapid and frequent optimization based on granular data sources need to rethink their measurement strategies. This category includes disruptive brand marketers – startups that were born online and deliver value to customers in new, innovative ways. Now is the time for marketers, including digital natives, to upgrade their existing marketing analytics. By 2023, marketing mix modeling (MMM), a statistical analysis of marketing and business, will become a relatively promising measurement solution for brands seeking actionable insights into the effects of cross-channel marketing.

As privacy laws change how marketers collect and use data, how brands can use marketing mix modeling to gain insights and measure performance.

Modeling the marketing mix in the new era

 Using innovative machine learning algorithms, MMM lets you promote actionable ideas with the level of detail and speed you want. By not relying on individual data, it becomes a single, comprehensive and consistent measurement system in an evolving data ecosystem. It's great to see marketers moving towards a sustainable MMM – developing skills and hiring talent for internal management. A good indicator of this change is the growth of the Robyn MMM open source community on Github and Facebook . Over 1,000 marketers, analysts and data enthusiasts are already connecting, discussing and learning as they progress through their MMM journey.

In a recent study by Accenture, they conducted several experiments using a custom MMM to test its suitability for disruptive brand marketers who need a powerful and cost-effective system to inform detailed media optimization. .

Their findings show that MMM offers the following benefits when used with advanced machine learning techniques and innovations:

1) Reliable MMM is available for traders of all sizes and categories: The success of MMM is primarily measured by the ability of the model to predict the dependent outcome. Accenture's experiment, using 1,200 data characteristics from 5 different sources commonly collected by traditional brand marketers and disruptive marketers, demonstrated high predictive accuracy at R-squared values ​​of 90% and meeting industry standards at or below 5% absolute. necessarily average. percentage of errors (MAPE).

2) MMM can provide detailed and actionable results: Thanks to advances in machine learning, MMM can now use techniques such as the gradient descent algorithm to decompose the data and generate actionable insights based on these variables. In the Accenture study, the model broke down two years of data down to the day-of-the-week level, an important metric for marketers changing their daily budget (as seen in Figure 1). This is an example of how MMM can provide actionable and detailed insights that marketers can use to optimize their marketing activities.

3) MMM demonstrates cross-channel synergy without user tracking: Disruptive brands often seek to understand their customers' conversion paths in order to optimize their cross-channel marketing efforts. Marketing mix models, when integrated with advanced machine learning techniques, can provide similar insights into cross-channel synergies. Accenture's experience provided a clear picture of cross-channel impacts. The results are summarized in the “grid” of contributing factors in Figure 2 below.

A method that will consistently meet basic measurement needs

MMM integrates and evaluates all online and offline marketing activities, builds a picture of the relationship between them and extends tracking of factors such as promotions, seasonality or competition. Today, MMM requires fewer resources and budgets to implement and uses aggregated data to generate fast and detailed cross-channel insights, making it suitable for brand marketers of all sizes, including those who focus on direct advertising. Even new brand marketers who make frequent business changes and tend to use a variety of free marketing tactics benefit from the comprehensive overview provided by MMM.

Four best practices for efficient, accurate and consistent results:

1) Decide on the main objectives before the simulation: Given the variety of questions the MMM can answer, it is important to create a training plan for the MMM and focus on solving one question at a time. Aligning key objectives is a fundamental first step in the MRM process. All subsequent steps in the MMM construction process benefit from a clear understanding of the main objectives of the MMM.

2) Make sure the data is relevant and complete: Create separate need variables for each strategy that marketers want to quantify. In addition to developing media-related variables, it's equally important to include a comprehensive list of non-media variables that can affect a brand's business results. These variables vary from brand to brand, but some common ones include economic factors, seasonality, competition, etc.

3) Choose an MMM option that answers your questions: The most common questions are usually answered by various MMM options on the market, whether it's an open source solution like Robyn, an affiliate solution or a self-service MMM. SaaS solution. When evaluating MMM options, ensure their capabilities effectively address the questions identified in Step 1 so brands can generate actionable insights from MMM.

4) Regularly update and calibrate MMM to reflect business changes: Investing in a data infrastructure that allows new data to be automatically fed into the MMM model will help marketers and modeling professionals achieve long-term effectiveness in updating MMM models. It is important to establish a calibration framework and choose the most reliable MMM model. Increasingly, research is the industry's gold standard for measuring ground truth. Marketers should conduct additional research on their marketing channels alongside MMM guidance to improve model accuracy and build confidence in MMM adoption.

Flexible measurement solutions require a solid foundation

MMM and its evolution are here to stay. It's time for new brands to stop waiting and start the MMM measurement journey. The best way for marketers to understand the effectiveness of ads is to use rapid simulations combined with cause-and-effect testing. Performing marketing mix modeling now, before the door closes on individual-level data, will help prepare brands for future changes in online privacy. Invest the time and resources to build it now; pay those who do.