Home Blog Modeled Attribution: ...

Modeled Attribution: a contemporary framework for measurement

Filippo Trocca

We’ve talked before about Marketing Mix Modeling and how the future of marketing measurement must necessarily shift from a privacy-first model.

We’ve brought this hot-topic to the center of debate on a number of stages at the most recent industry events, from IAB Forum to Search Marketing Connect, and some of our guests in the halls have raised some questions about the term MMM itself.

“Marketing Mix Model” is, in fact, considered to be an outdated definition, harking back to the past, to econometric models, to offline… true, that’s exactly the matrix, but the layers of technology that can be activated today and the synthesis with Multi Touch Attribution are evolving those patterns toward the measurement of the future, Modeled Attribution.

Privacy, big tech and modeled attribution

In the current privacy-centric scenario, traditional methods of marketing measurement and analytics are no longer viable. So the pressing question is: What are the next key steps for companies that want to effectively measure their marketing activities, understand how best to allocate budget, and choose the best performing channels based on their real contribution to achieving business objectives?

User-level measurement has always driven companies’ strategic choices because it allows them to accurately understand the impact of marketing campaigns in order to make targeted optimization and budget decisions. Until now, however, in an era that puts privacy at the center, to the point that the measurement methods used for years in the ads world, for example, are anachronistic and no longer compliant.

Major Big Tech companies such as Google and Meta have implemented enhancements to preserve user-level measurement as much as possible. These include the Enhanced Conversion and Conversion APIs, which enable more accurate attribution of conversions to different marketing campaigns.

It’s not a complete solution, however, because it only covers a portion of the missing data; and that’s where Google’s Consent Mode comes in, which leverages modeling techniques to account even for users who opt out of consent for marketing and analytics.

It is possible that there is some skepticism about relying on “modeled” data within reports. However, it is important to note that this is not new. In fact, modeled conversions have been present in tools such as Google Ads and Facebook Ads Manager for many years. And, paradoxically, the requirements for modeling will only increase as known user datasets continue to decline.

Although large vendors do not make it very easy, with the right expert support it is possible to compare unmodeled results with modeled results. This allows for more informed decisions about the numbers and their degree of accuracy.

Rather than running away from modeling, marketers should try to understand it better and embrace it wholeheartedly, because it is the near future we can envision.

Econometric models and attribution: “back to the future” of measurement

Why, then, does talking about Modeled Attribution mean talking about econometric models and attribution at the same time?

Attribution has been the focus of heated debate among marketers for years now, all the more so today as we think about how to untangle the many closed ecosystems and privacy restrictions.

Given the inevitable gaps in known data, a user-level attribution model is now essentially impossible, unless we consider a specific subset of channels that do not cross so-called walled gardens, precisely.

Otherwise, creating a robust custom user attribution solution at the channel level is utopia.

That said, obviously every company in the world will still need to accurately measure the performance of its media mix and make effective budget decisions.

Interestingly, the optimal next-generation solution is precisely a combination of two historical approaches.

Modeled Attribution, in fact, takes the best of Marketing Mix Models, understood to be traditional econometric models that reason about aggregate-level data sets rather than user-level inputs, which is perfect in today’s cookieless Web, and Multi Touch Attribution, combining them to provide a view of the entire marketing performance funnel, while respecting privacy, without worrying about considerations such as user consent or how to navigate walled gardens.

External factors, such as trends and seasonality or competitor activity, can also be included to increase the accuracy of the model and isolate the specific impact of individual media campaigns.


Modeled attribution and marketing mix model: yes, but what about granularity?

So when people say that the Marketing Mix Model concept is outdated, they are actually pointing out one of its disadvantages: MMMs, by their very nature and logic, return aggregate results, with a very low level of granularity (e.g., data on the TV channel VS data from the Digital VS data from the PR&print world) and with a limited frequency, every 3-6 months or so.

This is no longer the case today with MMM’s “new deal,” which is Modeled Attribution: the latter, in fact, can leverage direct connections with each of the marketing platforms in use in the company to obtain daily input at the most granular level possible. This makes the data much more usable for tactical planning and budget decisions.

It can also integrate within it the results of incremental Conversion (Conversion Lift)i tests, so it can be able to calibrate the model even better.

The con is that the initial setup requires precise planning and experience, then experienced Data Scientists who can master the models and a Data Analytics team that can guide companies in structuring a true Data Strategy.

However, this brings significant competitive advantages in the long run, as modeled attribution not only provides all the details typical of Multi Touch Attribution, but is also more stable over time in view of further industry changes, which is possible precisely because of the power of modeling, which reasons through AI and ML.