Since GDPR came into effect in May 2018, user journey tracking has been on the decline, and tracking all users on websites and mobile apps has become increasingly difficult. In addition, big companies such as Apple and Google restricted tracking across websites by prohibiting third-party cookie tracking, and Apple single-handedly killed user journey tracking on mobile devices with their iOS 14.5 update.
These restrictions initiated a kind of rethinking among marketing managers. On the one hand, performance marketing became more aware of the limitations of user journey tracking while, in parallel, realizing that the frameworks they used to work with weren't a realistic model of how marketing actually worked. Only including trackable touchpoints (e.g., Google Ads, retargeting, or affiliate marketing) in the calculation of CPO or ROAS is similar to pretending that those are the only causes of orders or revenues for an organization. So in some way, performance marketing was living the lie that all conversion or revenue was only caused by performance marketing, neglecting all other effects, such as branding or offline marketing.
Another factor motivating marketers to prioritize their branding activities is to lessen their dependency on the typical performance marketing channels, in particular Google Search Ads. Ever-increasing competition and higher click prices can easily lead to a race to the bottom in terms of performance and efficiency. Specifically for D2C brands, an effective way to differentiate and survive this performance battle is to increase brand awareness and focus on the upper funnel. Users with higher brand awareness will result amongst others in higher click through rates, increasing rankings and decreasing click prices.
Adidas courageously published the finding of an econometric modeling through which they found out that they had allocated too much budget to performance marketing channels in relation to the measured sales (77% of marketing budget but only 35% of sales) compared to their branding activities (25% of marketing budget but 65% of sales).
What’s usually also neglected in marketing measurement are non-marketing effects, so-called baseline effects, such as trend, seasonality, macro economic developments (recession, inflation) and changes in the other 3Ps of the marketing mix: product, price or placement.
Understanding triangulation in marketing measurement
Triangulation, within the context of marketing measurement, involves the strategic combination of multiple methods or approaches to validate and cross-verify data to help you get closer to the “truth.” It goes beyond relying on a single measurement technique; it provides a more comprehensive and accurate understanding of marketing performance by relying on the strengths of chosen methods. Triangulation most often involves two or more strong methods of measurement, such as marketing mix modeling (MMM) and incrementality testing.
Requirements for a future-proof marketing measurement system
A future-proof, holistic marketing measurement system including triangulation should account for as many of these effects on sales and orders as possible. It should also use all possible data sources and more than one measurement approach.
Marketing measurement requirements differ depending on the level of decision-making
Strategic marketing steering and planning requires a system that allows comparing different budgeting scenarios and possibly incorporates all input factors and effects. For tactical or strategic marketing controlling wholeness in terms of different factors influencing order and revenues has a higher priority than granularity. For this level of marketing decisioning for example integrating baseline effects such as seasonality and trend to understand incrementality is more important than to optimize ad spend on a SEA keyword level.
Operational marketing decisions demand detailed data about user journeys and sequences of marketing touchpoints to understand interdependencies. On an operational marketing optimization level measurement needs to be available on the most granular level, for example on a Google Click ID (every Google Ad click has a unique ID) which allows to feed attributed revenues back into Google’s bid optimization.
These different levels of decision-making require different types of measurement solutions and knowledge about the suitability and limitations of each. The three most common ones discussed in the marketing literature are the following approaches, and can be used as part of triangulation:
- Multi-touch attribution based on user-level tracking (micro-level data)
- Marketing mix modeling based on aggregated data (macro-level data)
- Experiments / AB tests / randomized controlled trials / incrementality tests
An integrative approach for holistic marketing measurement
Each of these three measurement techniques have their very own advantages and disadvantages and are suitable for different levels of decision making. A truly holistic marketing measurement system combines all three approaches and thereby 1) is also resilient to further tracking and data restrictions and 2) can quickly adapt to new possibilities for retrieving user-level data, such as server-side tracking.
This system allows for a coherent data story, preventing inconsistent insights between the different approaches (e.g. MMM showing a below ROAS for Google Search and MTA an above average ROAS) and benefiting from the advantages of each technique.
An integrative approach enables the methods to learn and benefit from each other in the following ways:
Marketing mix modeling can be optimized through integrating bayesian prior distributions or multi-objective optimization. A bayesian prior could for example be a ROAS of seven for Youtube, resulting from an incrementality test or another attribution model. Integrating such a prior allows the model to be trained from this starting point instead of starting without any prior knowledge, which in theory should lead to more accurate models.
Multi-objective optimization allows the marketing mix model to be optimized not only for minimizing the statistical error (e.g. regarding difference between predicted and actual conversions per day), but also for minimizing the distance between the model’s attribution (e.g. 750 conversions to YouTube) and an outside attribution (e.g. 800 conversions to YouTube). Outside attribution could be results from an incrementality test or MTA. The best model would hence be closest to the attribution results of incrementality tests or other attribution models, while reaching a desirable threshold of statistical accuracy.
The same works for MTA, which could learn from the results of marketing mix modeling, as well as experiments in the same way. Ideally MMM learns from MTA regarding typical performance channels, for example attribution between Google Search Non-Brand vs. Google Search Brand, and MTA from MMM for example when it comes to TV attribution. Both, MTA and MMM would incorporate results from incrementality tests and experiments, as those have the highest validity and informative value.
The system is also able to recommend for which channels to run experiments and incrementality tests. For example, based on which channel or marketing tactic has the highest parameter variance, highest disagreement between models etc., the system would be able to feedback to marketers for which marketing tactic the degree of confidence is the lowest. This would greatly facilitate designing future incementality tests, the results of which could then again feed back into the MMM or MTA components of the system.
Seamless data integration, automation and abstraction
To make the communication between the three components of this holistic marketing measurement system, or triangulation, work we need frictionless data integration, automation of data flows and model updates, and a certain level of abstraction and transparency.
For seamless data integration all data sources should be integrated through standardized APIs or user interfaces. Tracking data, user touchpoint data as well as aggregated will be made available through data storages such as AWS S3 or Google’s Big Query. Data from marketing tactics, where data is not readily available through APIs, e.g. Out of Home, Print campaign, PR activities or competitor activity data can be included through a standardized data uploader within the system’s dashboard. The goal is to minimize manual intervention and to build automated data flows wherever possible, e.g. through out of home campaign data that is automatically transferred from the respective agency into the system’s data storage.
Once the data is made available, automated data flows transform and map them in the required way, so that they can be fed into the respective modeling techniques (MTA & MMM). Models are trained and validated automatically and scoring recalculated if needed (e.g. when new TV airing data uploaded for the past for weeks).
The user of the holistic marketing measurement system does not need to know nor understand (although they are allowed to if they want to), which measurement insight and performance attribution has been calculated through which modeling technique. For example, a marketing manager does not need to know that recommended budget allocation for TV ads has been generated by an interplay between incrementality test results and MMM. He/she also doesn’t need to know that the attributed revenue per Google Click ID (Gclid) fed back into Google Ads for Gclid that weren’t tracked is estimated and intrapolated.
This system is resilient in that it doesn’t rely on user-level tracking data, but can still leverage it where possible and additionally allow for smart budget allocation on all levels of decision making. For strategic decisioning it greatly enhances monthly or quarterly media planning through scenario optimization. On an operational level it enables data activation by automatically exporting attribution data on the most granular level (e.g. GClid) into the respective ad platforms.
Use Cases
To exemplify the advantages of the discussed holistic marketing measurement system, two use cases mitigating the impact of tracking restrictions will be provided in this section:
Filling the gaps
Due to tracking restrictions (e.g. ITP or Consent Management Systems) conversions tracked and attributed based on independent MTA (e.g. within Funnel measurement) or attribution within an ad platform might not sum up to the total number of conversions for that marketing tactic. For example Google might only attribute it’s Non-Brand campaign with 2000 conversions, even though the MMM modeling resulted in 3000 conversions for Google Non-Brand, because Google couldn’t reliably track all conversions. A particular Non-Brand campaign “Winter 2021 Ugly Sweater” might have a share of 25% of these 2000 conversions. These 25% can then be applied to the 3000 conversion attributed to Google Non-Brand, resulting from the MMM part of the system. Thus, the overall and more realistic conversion attribution for this particular campaign is 750 conversions.
However, by using the triangulation method to combat this, all bases are covered and new tracking restrictions become less of an issue by canceling out any weaknesses and leveraging the strengths of chosen methods.
Mobile app attribute on a source level for non-trackable channels
A common and more specific problem for mobile app marketers is how to evaluate their mobile app campaigns on iOS devices. Due to the IDFA tracking restrictions enforced through Apple, there are many blind spots regarding the performance of campaigns on, e.g., Facebook or Instagram. The above system would allow us to measure and extrapolate the performance of trackable user journeys. For example: Only 10% of all iOS journeys are tracked (10% opt-in rate). 20% of the conversions of these opt-in user journeys are attributed to a particular Instagram campaign, based on MTA. This distribution could then again be applied to the more holistic, albeit less detailed, result of the MMM, resulting in a holistic mobile campaign attribution.
Conclusion
Some vendors and marketing organizations have developed a rather dogmatic approach. They might, for example, claim that everything except incrementality testing is flawed and that modeling user-level data is not worth it. Possibly they have lost confidence in processing user-level data because the expectations were overinflated, in the sense that MTA has overpromised to create 100% marketing transparency. This, of course, was never the case or possible because many input factors were not trackable. In addition to this, less and less previously trackable activities are not trackable any more (display ad impressions).
User-level data holds a lot of valuable information, which should always be included if available since it can drastically increase the quality of marketing decisions being made.
There are also some promising developments regarding user-level measurement data. Google’s server-side tagging and Meta’s and Mozilla’s Interoperable Private Attribution Initiative (IPA) are just two of the most recent developments.
Two years ago, client-side tag management became widely popular with marketing organizations, website administrators, and web developers. It is now possible for marketers to become easily capable of implementing new javascript code on their website. It’s not impossible that the same could be true for server-side or backend tracking in the future. After all, there’s a huge motivation for sophisticated marketing organizations to generate as much valuable marketing data as possible to optimize their marketing performance.
Try this next:
