Attribution

Attributing value to marketing campaigns

Attribution modeling is a process of assigning, or attributing, credit to specific campaigns for their role in driving conversion events (valuable actions taken by potential customers like purchases/revenue, leads, site visits, etc).

In fact, every approach to marketing measurement shares that same goal - it's all about understanding the actual business value driven by each individual campaign/channel, and comparing that value with the investment's cost in order to calculate ROI. Without an accurate understanding of ROI companies are unable to make optimal decisions about which investments to continue and to scale up, where optimization is required to reach profitability, and which investments should be cut and redeployed elsewhere.

Attribution modeling is a specific measurement method that tracks individual people, across the marketing touchpoints that influence them and the resulting conversion actions they take, and assigns credit for those conversion actions to one or more of the preceding marketing touchpoints.

Attribution modeling, over the years

Attribution modeling approaches have evolved consistently since their application to digital marketing, and for a long time it was all improvement. But over the past decade third-party (3P) cookies have been blocked in more and more situations, which has created major problems for attribution modeling.

Last-click attribution (LCA): In the early days of Google Search and PPC advertising almost all attribution models were 'single-touch', meaning that all of the credit for each individual conversion event was assigned to a single marketing touchpoint. The vast majority of those were 'last-click' attribution models, assigning all credit to the most recent touchpoint that the customer clicked on prior to converting.

Today, even within large and relatively sophisticated organizations, last-click attribution remains an overly common approach, despite not accurately representing the increasing complexity of customers' decision-making processes and the multiple marketing touchpoints that typically influence their decisions.

Multi-touch attribution (MTA): Over time more and more companies sought to go beyond single-touch attribution, in order to better represent the real world, by dividing the credit for each individual conversion up across multiple marketing touchpoints. This is what's known as multi-touch attribution (MTA), or sometimes 'fractional attribution'.

Companies, through the digital marketing platforms they adopted, could typically select from a range of options for splitting conversion credit across touchpoints. These options included 'linear' (even distribution of credit across all touchpoints), 'U-shaped' (more credit assigned to the first and last touchpoints with the remaining value evenly distributed across the touchpoints that took place in between), and 'time decay' (assigning credit to touchpoints based on how recently each took place), and are collectively referred to as 'rules-based' MTA models.

Data-driven attribution (DDA): Multi-touch attribution modeling enabled more realistic ROI calculations than single-touch, but it still assigned credit in a somewhat arbitrary way based on the rules of the selected model. To address that, digital platforms (and even individual marketing organizations) began to develop data-driven attribution (DDA) models, a more advanced form of MTA.

Data-driven attribution models apply statistical modeling and machine learning to customer journey data, including both converting and non-converting paths, in order to determine how much credit each touchpoint should receive. In theory, data-driven attribution promised to deliver the ultimate measurement solution, accurately identifying the impact of each marketing touchpoint and assigning credit across all digital channels and advertising ecosystems.

Unfortunately for marketers, it would never come to be.. DDA is still the most advanced form of attribution and therefore still has an important role to play, but its ability to track complete, cross-channel customer journeys was tied to the use of third-party (3P) cookies.

The deterioration of third-party (3P) cookies: In the mid-2010s the industry was all-in on multi-touch, data-driven attribution. AOL acquired Convertro, Google acquired Adometry, and briefly it appeared that attribution would be able to deliver accurate, cross-channel measurement all on its own.

But soon after that, Facebook (now Meta) and Google 'walled-off' their respective advertising ecosystems and Apple started implementing privacy-focused policies like ITP and ATT, significantly limiting the ability of third-party (3P) cookies to track individual users' marketing touchpoints and resulting conversion actions. Together these changes largely broke the ability for any attribution platform to track the complete, cross-channel journeys that influence customers' purchase decisions.

Today, even with the most advanced data-driven algorithms, it's truly impossible for attribution modeling to accurately value the impact of marketing investments on its own. No system is able to connect individual users to the ad impressions they've been served across both Meta and Google, and in many cases Apple's platforms prevent tying conversions even to the most recent preceding touchpoints.

Attribution modeling still plays a critical role in digital marketing measurement and optimization, but it will never be able to provide comprehensive, cross-channel measurement on its own.

The ideal role of attribution modeling today

Given the growing limitations around tracking individual users' behavior, attribution modeling cannot be trusted as an organization's primary method for measuring the impact of marketing investments. At the same time, there is no other approach that provides the granular, real-time measurement that is necessary to drive automated bidding and short-term optimization decisions, both of which are crucial to effective digital marketing.

The result is that attribution modeling is a necessity, but one that can lead companies to make bad investment decisions if trusted as a standalone source of truth. Fortunately, by leveraging incrementality measurement alongside attribution, and by being very thoughtful and intentional about the types of attribution models employed and the role each plays, marketers can significantly improve measurement accuracy and therefore inform better investment decisions.

Incrementality-calibrated attribution: Attribution modeling must be leveraged for its unique strengths, specifically the granular and real-time insights it provides. But marketing teams must also employ more accurate, though not sufficiently granular or real-time, measurement methods like incrementality testing to account for the large and growing limitations of attribution modeling.

This leads to a practice of tuning, or calibrating, attribution models over time based on the results of periodic incrementality tests, typically at the channel or sub-channel level.

Incrementality results, when based on rigorous and conclusive tests, should be treated as accurate assessments of the overall impact of the channels/campaigns included in the test. Comparing those incrementality results to attribution model outputs, for the same scope of channels/campaigns and over the same time period, informs a calibration multiplier. That multiplier can then be used to adjust or calibrate the attribution model's results for those channels/campaigns going forward (or at least until a more recent incrementality test concludes covering the same scope).