A hotly debated area of marketing analytics is multi-touch attribution modeling. Before the advent of advanced browser and device tracking during the 2010s, last-touch attribution was the standard model used to deterministically measure the impact of marketing efforts. After tracking caught up with the evolution of the internet in Web 2.0, there was a rise in tools and reporting that were capable of stitching together unique data points in a user journey to conversion which led to multi-touch attribution modeling. These capabilities opened up pandora’s box of marketing measurement and with it a debate around the right and wrong approaches. The use of Last-touch modeling was written off by many analytics thought-leaders in favor of a wide assortment of multi-touch models. Fast-forward roughly a decade later and despite their criticism of last-touch, there is no resolution to the debate of what is the right model. With the rapidly evolving marketing analytics landscape, I don’t believe we will ever know or soon care.
Multi-Touch Attribution Modeling
To start, here is a refresher on multi-touch attribution modeling. Multi-touch is a type of attribution modeling that credits more than one interaction (touchpoint) in a user's journey from first interaction or exposure to the time they converted. This logic aligns with the technologies impact on our purchasing behaviors going from one device in one seating to multiple devices and a decision being made over time. Common credit-based multi-touch models are
- Linear = Credit is evenly distributed among touchpoints within a specified lookback window,
- Time Decay = Credit is skewed towards the last-touch point and decreases based on time elapsed and the number of touchpoints prior
- Position Based = Credit is skewed towards the first and last-touch point with credit evenly distributed between these points.
The other multi-touch approach that does not have a definitive position structure is data-driven also known as algorithmic, which is based on the analysis of internal business metrics and their relationship to the conversion value.
The Limitations of Multi-touch Attribution
With so many choices, the immediate problem this modeling creates is the analysis paralysis that comes from “choosing the right one”. As someone who went to great lengths researching and building an attribution model for an organization, I know firsthand how there is no real structure in the decision process with most solutions being subjective and black box. While we all hope there becomes a standardized model that you can adapt your business to, modeling has to be a reflection of how your business operates and your audience’s purchasing behaviors. Even after this analysis you still need to make a subjective decision on which model fits your data and the insights you are looking to reveal by using a model. Trial and error can help you find the closest fit with the numbers, but at the end of the day, it is just storytelling. Once you inject subjectivity into any decision-making reporting, biases will be present and it becomes more art than science.
Beyond the model selection dilemma are the physical limitations in getting an accurate representation of the user's journey. The explosion of phones and tablet usage has fragmented the once linear purchasing journey and our usage across multiple devices presents significant troubles in stitching together a user journey between them. You might discover a product on your phone on the subway. Research it more on your work computer and then a week later, purchase it on your personal tablet at home. Each of these touchpoints would be identified as a unique user since each device/browser has its own identity and there is no accurate cross-platform tracking solution to tackle this problem. The platforms that are most capable are Google and Facebook which stitch together journeys based on identifying logged in users across multiple devices, but even this approach isn’t reasonably accurate.
Does it Really Matter What Model You Choose?
Since most journeys don’t occur on one device, what would using a different attribution model prove given the unknown variable of touchpoints not included in the journey that influences the outcome of a conversion? Last-touch might be the enemy, which I can side with to some extent, but can we truly validate the impact of marketing initiatives knowing that potentially valuable touchpoints are missing from the multi-touch model? In some sense, we could end up overvaluing or undervaluing the measurable touchpoints by having only part of the picture, leading us down the wrong path with our marketing strategy. Instead of putting so much emphasis on the model selection process and gathering our pitchforks to drive out last-touch modeling, could a better approach be to compare multiple reports involving both last-touch and multi-touch reporting to determine the value of marketing efforts?
The value of an analytical effort is only as significant as the understanding by people who will use it. In my opinion, given the amount of effort, subjectivity, and complexity the model selection process can take, I feel as though there are more tried and true marketing tests that can be utilized to prove the value of a channel than a black box solution that attribution modeling can quickly turn into.
Death by Privacy
Before discussing alternative measurement solutions, I want to highlight that there are even more significant challenges already in motion that could signal the end of multi-touch modeling’s utility in the near future. Privacy and government regulation is one change that has started the shift in the landscape of marketing analytics. GDPR and CCPA created strict user tracking regulations with the most prominent change being the requirement of user opt-in for cookie usage in Europe and California. What has followed since this regulation is Apple’s App Tracking Transparency framework and SKAdNetwork which uprooted the global mobile app analytics landscape. The common idea shared by these initiatives is the restriction or complete removal of user-level tracking. Without this tracking, it is hard to have the ability to do proper multi-touch modeling. For web, the restriction of cookies and soon-to-be elimination of 3rd party cookie tracking will put a dent in creating a unique touchpoint journey. For mobile apps, Apple’s gatekeeping of attribution with SKAN has eliminated the ability to see the timestamps of ad interactions to create any touchpoint journey and Android is starting to migrate to similar, but less stringent privacy tracking rules.
The Future of Marketing Attribution Reporting
Marketing analytics isn’t for the faint of heart. The layers of complexity and constant evolution require you to be quick on your feet and not get bogged down in the details. Multi-touch reporting is a topic that I have felt has dragged on for a while and is starting to come full circle with the impact of privacy and regulation. The majority’s sentiment of “death to last-touch” is changing to “it’s the best we got” and that is fine.
Experimentation is the only way to make progress with data and in the wake of deterministic measurement limitations, probabilistic (statistical) techniques will have to be used to fill the void. Old school is the new school, with statistical techniques like Media-Mix Modeling (MMM) and Geo Experiments (GeoX) being utilized to fill the void left in proving marketing effectiveness. We were spoiled living in a world where we felt like one report could tell us everything we needed to know and now have to come to terms that things will never be as black and white. Report blending like last-touch reporting layered with the results of incrementality tests will be the only way to give us the most accurate depiction of what marketing initiatives are helping or hurting our bottom line. Reporting isn’t going to be pretty, but at least we know that we will be relying on more science than art at the end of the day, and with that shift comes objectivity, accuracy, and opportunities.