A new era of ad measurement?Rebuilding strategies with media mix modeling | Deep | Campaign Japan Japan

Media mix modeling (MMM) is making a comeback in the digital marketing field as the privacy environment changes due to the abolition of cookies and mobile IDs.

The top three advertising platforms Meta, Google, and Amazon are also investing in MMM as a statistical tool to measure advertising effectiveness to recognize the influence of competing channels. MMM uses statistical data to analyze a wide range of media channels across digital and physical advertising spaces, taking into account seasonal variations, promotions and other factors.

Most advertising platforms rarely use tools like MMM to cross-evaluate different advertising media. This approach therefore departs from common practice. Traditionally, brands used statistical models to build MMMs to assess the correlation between media spend and sales growth.

However, privacy regulations have severely limited measurement capabilities in recent years, requiring even major platforms to adopt measurement tools to demonstrate attribution.

Meta got into this trend three years ago with the introduction of its MMM tool Robyn. This open source MMM tool does not rely on cookies or personal data. This comes at a time when Facebook was suffering serious damage from Apple's App Tracking Transparency (ATT) framework, which made it difficult to track conversions.

Amazon probably believes that first-party retail data will also give it an advantage in MMM analysis. Google also appears to be exploring the development of an MMM tool for brands to use as a performance evaluation tool for cross-channel advertising campaigns.

Campaign explored where the renewed interest in MMM is and how major platforms are getting into MMM analytics.

Will history repeat itself?

MMM has been around for decades before real-time bidding (RTB) became popular on the internet and in ad tech. You can think of MMM like solving a puzzle that is missing or broken.

Today, big retail media networks like Google, Facebook, Amazon, TikTok, Apple, and GrabAds and Carousell Media Group have a clear incentive to tackle MMM. They need to prove that their advertising channels are providing marketers with the right audience reach and driving sales.

MMM also provides valuable insights for marketers to better understand their customers.

Niraj Nagpal, an ad tech and martech consultant, said proving their contribution brings more ad spend to the platform. He explains that in some cases, the company could also aim to win orders for cloud services such as data, storage, and computing.

In the long run, these MMMs will solidify the position of major platforms as diversifying media strategies and must-haves for IT providers.

“When data resides in separate silos, whether it's across media channels or in cloud environments, it becomes increasingly difficult to complete the puzzle,” Nagpal told Campaign.

“Most marketers are willing to accept the imperfections and biases of the data from these environments (usually privacy-sensitive, hashed or statistical versions of personally identifiable information) while increasing their advertising budgets. We can only hope that only a select few advertisers will be able to invest the huge sums of money, data science teams, and cloud and clean room technology needed to aggregate the data needed to build MMMs. Masu”

(Top left to bottom right) Laraine Chris, Liam Brennan, Niraj Nagpal, Di Wu

Recognizing these issues, Nagpal will work on an integrated data strategy using data lakes and data warehouses, and will work with partners who can operate data in accordance with privacy regulations to build a robust first-party data strategy. We recommend this to marketers.

However, he cautions that this process is not straightforward and achieving ideal conditions is quite difficult.

The first step to achieving pure unbiased MMM, even if you aspire to it, is partnering with an agency team that can properly manage data storage and derive insights from incomplete data sets. It's something to do.

“Advances in machine learning and generative AI have made it much easier to derive insights from complex data sets. Even media teams without AI knowledge can perform input tasks such as data conversion, cleaning, and consent management. We can now focus on the story and value of our data, rather than worrying about it,” Nagpal explains.

MMMmethod and accuracy

Marketers must ensure the accuracy of MMM measurements at all costs, even in an environment where user-level tracking is increasingly limited.

There are several easy-to-understand approaches to ensuring MMM accuracy, said Di Wu, vice president of data science at Jellyfish.

One is to combine MMM with experimental methods to measure media incrementality. Experimental data can be used to correct the MMM output results or be treated as prior probability values ​​for Bayesian-based MMM modeling (the concept of probability, interpreting probabilities based on knowledge and beliefs rather than frequencies), and can be Accuracy can be increased by incorporating global results into the model.

However, Wu cautions that simply aiming for the highest statistical score such as MAPE (mean absolute percentage error) or R2 (coefficient of determination R-squared) does not guarantee the most accurate MMM.

She says that effective use of MMM should be aligned with business objectives and does not recommend major shifts in media investment that would surprise stakeholders.

“Understanding your business guides your model expectations, provides your business with metrics that focus on statistical measures, and helps you choose the model that makes the most sense for your business.Business Considerations There needs to be a balance between accuracy and practical accuracy,” Wu explained to Campaign.

Lorraine Criss, Periscope's chief operating officer, said the key is to focus on collecting quality data, and automation will help improve accuracy and speed.

Brands can ensure accuracy by focusing on the quality of data collected and aggregated and incorporating appropriate attribution systems for analysis.

“MMM typically uses methods such as multivariate regression analysis.Meanwhile, to measure the effectiveness of MMM modeling, it is necessary to collect data sources related to a brand's marketing activities, such as advertising expenses and ROAS. '' Chris told Campaign.

Beyond the ethical aspects of user tracking, a second limitation of MMM is scale, says Liam Brennan, managing director at The Responsible Marketing Agency.

Brennan explains how the amount of data needed for analysis depends on the size of the audience, and the tendency to over-emphasize short-term metrics, such as focusing on sales over branding.

“MMM is by no means perfect, but with the right, accurate, and scalable data, it can better represent the short- and long-term effects of advertising efforts and help marketers make faster, smarter decisions. It helps us make decisions,” Brennan said.

Platform-specific data sources and transparency

Data collection is a key issue when building an MMM, so platforms such as Google, Meta, and Amazon provide data feeds for MMM purposes.

Wu points out that these feeds have several advantages. For example, the platform offers more detailed daily data such as daily moving averages (DMAs) for each local market. It also provides historical data spanning several years, allowing for comprehensive analysis.

Google, for example, provides data going back three years, which marketers can download all at once.

Accessing this data is relatively easy. Meta, for example, provides user interface (UI) and API options for downloading data, making it very useful for MMM practitioners. Marketers can download it manually or schedule it to be retrieved automatically.

Additionally, these data feeds maintain a consistent naming convention, making it easy to identify different types of campaigns and ads. For example, Google's data also includes bumper and non-skippable ads on YouTube.

“However, it is important to note that these MMM data feeds are limited to data within their respective platforms. To build a comprehensive MMM model, you need to collect data from other platforms, including offline media. We also need to collect data on

“Additionally, data on sales, competitive research, events, and promotions should also be collected to provide a complete picture of marketing performance.”

Chris doesn't want to think about what's going on behind the scenes at Meta, Google and Amazon. In any case, providing measurement tools seems to be the focus among them, she said.

However, whether marketers will leave evaluations to these platforms is another matter.

Chris emphasizes the need for marketers to ensure accuracy and think carefully when undertaking data projects, based on comprehensive planning and actionable intelligence.

By doing so, marketers can establish strong governance and avoid shortcuts that can be costly in the long run, setting brands up for true success.

“It's important to remember that MMM is designed as a data-driven solution for marketers to quantify the impact of their marketing and branding investments and improve performance,” says Chris.

“It allows us to understand the incremental impact of our marketing efforts and use that measurement to answer a variety of marketing strategy questions. It’s about thinking strategically. There are also clear advantages to building it in-house.”

“MMM is not always the answer. What an MMM model can do is form an important part of an overall measurement framework. It should not be the only tool for demonstrating marketing effectiveness. '' added Chris.

A.I.and the role of machine learning

MMM is undergoing a renaissance in the context of AI and ML (machine learning), and technological advances could bring huge benefits.

For example, machine learning is cleaning data from diverse sources and automating necessary data tasks such as formatting and margins, resulting in significant time savings and improved accuracy.

Advanced machine learning algorithms can uncover complex patterns and interactions in data, such as how weather affects sales or how seasonality affects advertising budgets, providing insights that human analysis can't capture. Also provided.

Traditional MMM methods require time to build and test models, so machine learning improves efficiency by quickly evaluating various combinations of variables, enabling rapid model development and updating with new data. It brings great benefits to the optimization of marketing strategies.

However, Chris cautions that AI alone cannot solve all problems, and the most technically sophisticated models can come at very high costs.

For example, to ensure compliance with privacy regulations such as the General Data Protection Regulation (GDPR), businesses must also take steps to protect user data while implementing MMM.

“The trend is increasingly towards privacy, and one of the key reasons why MMM analysis is gaining popularity so quickly is that it does not require personal data,” explains Chris.

“MMM is a marketing measurement tool that does not rely on personal data, but with up-to-date information and simulations, it provides reliable capabilities for data-driven decision making.”

Jellyfish uses a variety of media measurement methods, each with its own purpose, to get a comprehensive view of its marketing efforts, Wu explains.

Jellyfish uses multi-touch attribution (MTA) for continuous optimization of digital media, determining how much credit to give to different parts of marketing. You can make informed adjustments to your advertising strategy.

In addition to MTA, Jellyfish conducts incrementality experiments and conversion lift research to answer questions and find opportunities to increase revenue. We conduct these tests on key platforms like Meta and Google and measure incremental impact using geoexperiment techniques.

“MMM tools support channel decisions and advanced budget allocation. Unlike MTA, MMM does not rely on individual user data. Based on aggregated data, you can choose between different channels, strategies, and custom segments. We will consider it,” Wu said.

“We also incorporate external factors that can impact the business, such as competitor promotions, COVID-19, inflation, and industry trends. and is also used to predict outcomes based on the model for future quarters.”

“We employ these measurement methods to ensure the accuracy and objectivity of our marketing analysis. These tools are not intended to replace each other, but to provide a comprehensive understanding of marketing performance. Each needs to work together,” Wu continued.

Future trends and developments

Although MMM is a methodology that has been established by marketers for many years, it now demands more and is evolving daily to meet new challenges and expectations. Therefore, the future of MMM seems quite promising.

Marketers will seek more granular insights, including analysis by audience, campaign, and tactic, to better understand the impact of each element of the marketing mix.

Because MMM models are sensitive to small changes in inputs and initial settings, marketers can explore the art and science of ensuring that models deliver results in line with business expectations while seeking meaningful insights. It is necessary to strike the right balance.

In MMM, bottom-of-funnel tactics are prioritized. Because it's difficult to quantify the impact of branding ads like CTV ads and organic content, marketers need to find other ways to evaluate efforts that focus on brand building.

Additionally, as mentioned above, MMM models are generally built on historical data, which can quickly become outdated as marketing strategies evolve rapidly. Marketers need to be agile and adapt to campaign launches to maintain business relevance and business expectations.

“Traditionally, building MMM models required time and manual effort, making them less agile in the fast-paced digital marketing environment,” Wu explains.

“In the future, we expect MMM to bridge the gap in granularity, speed, privacy compliance, and branding measurement to deliver valuable and actionable insights even in a rapidly changing digital marketing environment. Masu”

Chris points out that future MMM trends will focus on reinventing MMM through AI and machine learning. MMM's history dates back to the late 1960s, but today's automated data flows have made it much easier to collect data from diverse marketing channels and information sources, she says.

If you have historical ad spend and conversion data, you can build a model with relatively little investment.

“Typical options include leveraging solutions from media agencies, independent providers, or building solutions in-house. If you are a well-known brand, you may see scale benefits. However, if you have a large budget, This field is quite viable without it,” Chris explains.

“However, as with anything, it requires thorough research, a considered strategic approach, and proper due diligence.The industry has evolved significantly when it comes to advertising measurement. Innovative solutions are being offered one after another.

“Meanwhile, advertisers are also increasing their investment in data science teams. This trend is similar to disruption during programmatic ad buying. “We've democratized access to quality traffic so that you don't have to be a well-known company with the deep pockets to buy quality inventory at market price,” Chris added. Ta.

Additionally, Chris says industry standards for MMM metrics and benchmarks are needed to ensure consistency and comparability across brands and platforms.

“That's why the Marketing Science Institute (MSI) has launched an initiative to develop industry standards for marketing mix modeling,” Chris concluded.