QuantifiedNation #6: Unconsented marketing measurement
One of the key milestones in the history of marketing measurement was the introduction of audimeters in select households. This created a representative sample for measuring ad viewership. These households provided consent in exchange for remuneration, incentivizing participation. Marketers then drew conclusions from this sample. We discussed this in Quantified Nation #2.
As we spend more time online, a larger portion of our behavior involves interacting with different servers around the world. Why track just a sample of these interactions? Technology allows us to collect data from every visitor to a website, put cookies on every ad seen. Most websites use JavaScript tags to send our browsing behavior to tracking technologies. The same applies to apps – what we see, what we tap, and what we type is recorded.
A new, overwhelming amount of information was made available to any businesses with a website and any digital marketing activity. Tracking user behavior to improve products, services, advertising, customer support, and many more has made measurement through individual tracking a pillar of successful businesses today.
This data offers precision in measurement and the opportunity for personalization. When used responsibly, personalization capabilities can exponentially increase the impact of digital marketing, especially advertising. Third-party cookies allowed marketers to understand user paths across different websites and their interactions with marketing activities, revealing which strategies were most effective.
Combining customer databases for refined campaign optimization and utilizing third-party data for improved targeting became daily habits for every data-driven marketer.
IDs and cookies everywhere
The use of personally identifiable information (PII), such as IDs stored in browser cookies or mobile device advertising IDs, has been a cornerstone of digital marketing measurement. However, this landscape has shifted dramatically. First in Europe, and now in many other markets, consented marketing measurement has become mandatory for any user-level tracking.
Europe's 2002 ePrivacy Directive mandated "clear and comprehensive information" to obtain user consent for cookies. This was a major step towards transparency and informed consent. The 2018 General Data Protection Regulation (GDPR) further strengthened these requirements. Despite this, surveys indicate that most people still don't fully understand cookies or do not actively manage their online tracking settings.
Traditional cookie-based marketing measurement (e.g., conversion tracking and digital attribution) has assumed near-complete coverage of users interacting with campaigns or websites. However, this ignores potential biases. Stricter consent enforcement exacerbates these biases, further impacting accuracy.
Moreover, factors like multiple devices, tracker prevention, incognito mode, ad blockers, and cookie deletion further complicate the picture. Despite these challenges, enhanced cookie-based methods remain a valuable tool for operational or directional digital marketing measurement. Fluctuations with total business metrics make everyone nervous, while neglecting the unknown unknowns. Modeling is a key process to make this data closer to reality
When we click an ad, whether we give consent or not, such click will be counted. Same happens in case we browse the website and purchase a product, such revenue will be counted. Now, the way we tie these two events together (marketing action and business outcome) determines how consented vs unconsented marketing measurement works.
Consented measurement
User-level methodologies that connect marketing actions to business outcomes traditionally rely on storing an identifier (i.e., a cookie) in the user's browser. As mentioned earlier, in the European Economic Area (EEA), linking these events mandates informed user consent.
The recent enforcement of the Digital Markets Act (DMA) has generated significant attention. This includes provisions such as prohibiting gatekeepers from tracking end-users outside of their core platform services for targeted advertising without explicit consent. As a result, advertising platforms like Google Ads as well as advertisers using it have implemented new consent management protocols (e.g., Consent Mode v2). There has been additional implications in products provided by these gatekeepers, it is fascinating to observe how these regulations reshape the technologies used by millions.
Unconsented measurement
Methodologies that tie marketing action and business outcomes using non user-level information to join both. Most common are comparing geographic regions with different marketing activities or observing the behavior of marketing actions and outcomes over time to find patterns.
These methodologies aim to provide a future proof estimation of the real impact of marketing. The requirement for individual data in consented measurement makes it much less appealing for a comprehensive view of marketing performance. Also, these might be the techniques that you need to develop more in the coming months/years. Let’s see two examples:
Geo experiments: you have your target audience segmented in different geographies and target these geographies differently. This requires a precise definition of the experiment, ensuring you are using comparable combinations of geographies.
Econometric models: find statistical relationships between marketing variables, mostly between marketing spend and outcomes. The most notably example of an econometric model applied to marketing measurement is the MMM (Marketing Mix Model) where its open-source options have grown in popularity (see below on the recently announced Google’s solution, Meridian)
You can measure your marketing without the consent of the user by using these two techniques because they do not require to tie marketing activity and business outcomes at the user level.
The latest innovations in marketing science allows you to build these techniques on strong foundations. If used in the right way, the ability to have a sophisticated understanding of the incremental impact of marketing on business outcomes.
This will help you reach greater results with your marketing and fall out of the attribution trap.
Hot takes
Google’s launch of Meridian
Meridian is an open-source Marketing Mix Model (MMM) solution built by Google. Open-source means you'll soon be able to download the Python library and use it directly on your own systems.
MMMs are a powerful tool for gaining a holistic view of marketing effectiveness across all your channels. To get the most out of them, you'll ideally have at least two years of significant marketing investment data across multiple channels.
Meridian employs Bayesian causal inference, a sophisticated approach that combines existing knowledge, data-driven insights, and uncertainty quantification. It then uses Markov Chain Monte Carlo (MCMC) simulations to estimate coefficients and parameters, including those critical for adstock and saturation curves.
Those are great and unique features but there is much more:
As expected in a bayesian solution, you can add prior knowledge. The great improvement here is that you can directly add ROI priors without distributions and from any source. If you have the ground truth through a sustained use of incrementality, that would be perfect. If you just have an intuition about ROI, well you can test it as well. This feature is the implementation of the research published recently called “Media Mix Model Calibration With Bayesian Priors”.
You can add reach and frequency data to those relevant channels, because 1000 impressions to 100 people is not the same as 1000 impressions to 500 people. Very few MMMs done by the best providers of the world account for reach and frequency today so this is a very good step forward in the improvement of modeling quality. This feature is based on the 2023 article Bayesian Hierarchical Media Mix Model Incorporating Reach and Frequency Data.
Last but not least, the ability to control for organic demand when measuring paid search or other lower funnel channels. It is a great challenge to distinguish changes in business outcomes driven by a demand change or a paid search spend change. This is better understood when adding query volumes as control variables when measuring the impact of these channels. This feature is based on the classic article linked in the last section of this article, scroll till the end to read it.
As you can see, there is a lot of research and innovation in this solution so we are all looking forward to try it.
Blog post - Documentation and access
AI-Augmented Predictions
The legendary expert on forecasting methodologies Philip E. Tetlock has just co-signed an article on “AI-augmented predictions”. The research focuses on how LLM can help human forecasters to get better results. The experimental design and the overall approach is fascinating. However, as usual in academic papers, it requires some effort to get to the conclusions. “This suggests that, at least at the time of this paper’s writing, LLM cognition may synergistically improve human cognition in the domain of forecasting when used as a human tool, even when LLM cognition by itself is somewhat ineffective”. You can read it here
Industry updates and upcoming events
ThinkBox TV Masters Summer 2024 course will start May 28th. You can enroll as of May 1st clicking on this link. This free course focuses on all the elements of successful TV advertising, from planning to measurement. Even if it’s just focused on 1 particular media and based in the UK, it has great content and it’s a worth investment if you work on audiovisual advertising around the world.
These events were mentioned in previous editions of the newsletter, but worth refreshing as they are coming soon:
I-COM Summit 2024, to be held in Málaga (Spain) in May. Registration
MeasureCamp Czechia, to be held in Prague (Czech Republic) in September. Registration
Ehrenberg Bass institute is touring the world with their “How Brands Grow for Executives”. There will be physical sessions in Singapore, Europe and USA. Price tag for a workshop is nothing to laugh about, but the content is surely top quality. Link to tegistration and more info.
ESOMAR will host a 3-day event on Insights in Bogotá in April 7-9th, including a dedicated session on media metrics on the last day. Link with details of agenda and logistics.
Chart of the week
Some new drinks becoming popular.
Nevertheless, these specialties are dwarfed when we incorporate coffee into the equation (link)
Oldies but Goldies
The 2018 “Bias Correction For Paid Search In Media Mix Modeling” paper (link here) laid out a range of solutions for the complex interrelations of consumer queries in Google search, paid and organic results, which typically pose a challenge on Media Mix Modeling.
La que más me gustó hasta la fecha!!!! top.