Attribution For data-driven marketing

28 July 2020

The technological shift in recent years has hit the consumption and marketing world alike. Consumers have the potential to carry out activities on multiple devices, and in a world of shrinking barrier between offline and online, they expect a seamless experience, no matter through which channel they are navigating. As a result, this seamless experience is leading to a disrupting shift in the way businesses understand and measure their performance.

For years, businesses have relied on last-click attribution to measure performance. But with the explosion of channels, the power of interacting with mobile devices, the traditional linear customer journey does not exist anymore. Hence the question arises: how do businesses attribute credit to the right channel in the new digital landscape with complex customer journeys; and how can data be leveraged

– to understand the multiple journeys,
– to optimize marketing effectiveness and 
– measure campaign efficiency.

A Primer on customer journey and attribution

Attribution in digital marketing is the machinery that produces quantitative knowledge to interpret the quality of communication through various digital channels in a network of consumers and stores to achieve the business goal. The business goal might be online purchase (conversion), or any of the intermediate customer journey stopovers (micro-conversions) representing awareness, consideration, or preference, or even after-purchase targets such as retention and advocacy.

Marketing channels include paid and organic search, email, affiliate marketing, display ads, social media, and more. An online purchase usually comes after a sequence of user interactions through different channels with different touchpoints. It is the attribution algorithm job to understand the causation in the above- mentioned process and assign proportional credit to each channel (Figure 1). The attribution outcome can later be used as a source of insight for future ad campaign planning and budget spend optimization, improving customer experience, and shortening the buying cycle.

Data has changed the practice of attribution

In general, attribution models can be ordered into two main categories; rule-based and data-driven approaches (Table 1). Rule-based attribution as it sounds is functioning according to a set of pre-defined rules. Regardless of data, these rules would always be the same in any circumstances. Single-touch and multi-touch strategies are two subsets of rule-based attribution, in which the credit goes entirely to a single touchpoint or gets distributed among all channels in a certain manner. These models would be fulfilling provided that the buying cycle in the studied business is short, simple, and predictable, or the focus is on a certain stage of the customer journey. For instance, when studying channels contributing to awareness, giving all the credits to the first interaction in the journey sounds sensible. Nevertheless, even in the simplest and shortest journeys, rule-based models suffer a major drawback when it comes to considering the non-converting ones. Consequently, rule-based approaches fail to consider actual channels’ efficiency. Imagine a case in which direct-entry wins all the credit in a last-click attribution since simply it is the most accessible last interaction, while if also considering the non-converting journeys, the performance of direct-entry might be quite poor.

Out of the two major attribution approaches, only data-driven attribution can deal with all the aspects of a real-life large business customer journey by leveraging the capacity of big data.

The rapid growth of online advertising in recent years, as well as the amount of recorded user data, have pushed marketers toward developing innovative techniques of measuring the effectiveness of advertisement channels. The diversity of channels has increased the chance of users’ exposure as a marketing target and raised the complexity of customer journey by many folds. Figure 2 shows that a large part of a conversion value is sourced by multiple-touchpoints journeys in modern marketing models.  The users’ behavior also is another variable that is pretty much dependent on various external factors at society or individual level. All the introduced uncertainties make it almost impossible to understand complex customer journeys and evaluate channels’ effectiveness by conventional methods. The necessity of a more in-depth analysis of the customer journey has provided the ground for the emergence of the second major category of attribution models called data-driven attribution. A successful attribution model has to take account of different journey aspects, such as journey length, sequence, diversity, user behavior dynamics, and as well, the non-converting journeys. Out of the two major attribution approaches, only data-driven attribution can deal with all the aspects of a real-life large business customer journey by leveraging the capacity of big data.

From game theory to machine learning

Data-driven attribution solutions also can be achieved through different analytical or numerical techniques. Cooperative game theory (e.g. the Shapley value), stochastic model (e.g. a Markov chain), and machine learning (e.g. linear regression and natural language processes) are different methods used to solve modern attribution problems. Depending on the implementation limits, amount of available data, or data-specific questions such as channels inter-dependency one method might be preferred over the others (Table 2). However, most of the data-driven attribution approaches are fairly capable of integrating journey length and sequence, behavioral dynamics, and non-converting journeys into the model.

Rule-Based Approaches

  • Single-touch attribution: One touchpoint receives the entire credit
  • Multi-touch attribution: Multiple touchpoints receive credit corresponding to their position in the journey, based on a prefixed rule.

Data-Driven Approaches

  • Cooperative game theory: The importance of each channel to the overall conversion is evaluated in its competition with the rest of the channels in a deterministic manner.
  • Stochastic processes: Channels credit is allotted based on the corresponding probability of future conversion in a sequence of touchpoints.
  • Machine learning: Channels value is assigned after a numeric estimation of the conversion probability based on available observations.

Attribution, Prediction and Data Dynamics

Overall, data-driven attribution capabilities are far beyond what any rule-based solution may offer. It builds the model purely based on data and it can adjust to the data dynamics, unlike rule-based models that rely on human interventions or modifications. Nonetheless, every attribution model is prone to customer journey fragmentation in case of cookie-based website tracking and is likely to make fallacious allocations. Besides, no clear mechanism exists that explains how to put attribution insights into action. However, the outcome is a valuable source to enhance marketing activities and optimize the customer journey and ad campaigns, either intuitively or as an input to another learning and prediction apparatus.

Back to top