The original promise of Multi-Touch Attribution (MTA) was that it excelled at offering a detailed perspective on how customers engaged with various touchpoints during their journey; aiding businesses in making informed decisions about marketing budget allocation; providing granular insights by tracking individual touchpoints across digital channels; gauging the effectiveness of marketing campaigns and helping identify the most influential channels and optimizing conversions. But can MTA still claim to do all these things?
The challenges that have beset MTA in recent years have gradually eroded its technical effectiveness as well as its standing as a key player in marketing measurement. Tech giants like Google, Apple and Facebook have made significant changes in response to privacy concerns, affecting the way advertisers track and target users across devices and presenting roadblocks for MTA with obtaining complete and accurate consumer journey data. These sound like significant blockers, but have these challenges indeed put an end to MTA or have they opened the door for some MTA solutions to develop alternative forms of tracking and measurement. Let’s look at each of them in turn:
Walled Gardens
Walled Gardens, or ecosystems, like Meta, control access to data and interactions within their boundaries and have caused a significant challenge to MTA, making it virtually impossible for marketers to track and attribute conversions across channels. Many of the biggest platforms, like Meta, YouTube, SnapChat, TikTok and others, don’t provide any of this vital user-level information. Marketers are then not able to evaluate impartially the effect of marketing campaigns on these platforms and instead must rely on the data reported by the platforms themselves – which again is problematic because of the double-counting of conversions (as Meta, Google, TikTok may claim the same conversion).
The technology behind the Walled Garden platforms comprises data silos where data is generated within the platform making it hard to integrate with data from other external channels. Walled Gardens typically offer last-click attribution models, attributing conversions solely to interactions internally; this undervalues the contributions of other channels and touchpoints outside the Walled Garden and ultimately affects resource allocation and marketing decisions.
Some MTA solutions overcome the Walled Gardens challenge by leveraging probabilistic modeling techniques to infer user behavior and attribution paths when deterministic identifiers are unavailable or are limited within Walled Garden platforms. These models use statistical algorithms to analyze patterns and correlations in available data to attribute conversions accurately. The solution can be a probabilistic approach, such as that of Synthetic Impressions, which take conversions reported by the platforms and probabilistically match them with conversions measured in web analytics (we will come back to Synthetic Impressions in a minute). It is a shift from having MTA working with purely deterministically connected data points to probabilistic matching.
Let’s dive deeper for a moment and see how this new MTA technology can successfully reconstruct the customer journey. Essentially, there are two steps to this process: first, how it creates the customer journey, and secondly, how it attributes credit across the touchpoints.
Synthetic Impressions explained
Our journey begins with a potential customer seeing an ad on TikTok on Monday, which gets their attention. On Thursday, they click on a YouTube ad, but they don’t make a purchase. On Friday, they see an ad on Facebook. Then, finally on Saturday they visit through a Google Ads Brand Search and make a purchase worth $100 at 1:30 pm.
Depending only on the click is misleading as it doesn't incorporate the effect of seeing an ad without clicking on it, which is very common with upper-funnel marketing activities. In this case, Facebook and TikTok would not get any credit. If these impressions are ignored, marketers will use incomplete data to manage their marketing budgets – wasting money and missing out on improving their performance.
As explained earlier, Walled Gardens platforms like Facebook, Pinterest or TikTok, don't provide detailed impression-level data for individual users, so it’s not possible to include them in the customer journey using traditional MTA methods. To overcome this, some MTA solutions have developed Synthetic Impressions methodology whereby granular data is downloaded directly from the ad platforms and incorporated into the existing click based customer journeys using an AI pairing algorithm. Specifically, data is downloaded from Facebook and TikTok with hourly granularity. Meta reports are analysed showing a post-view conversion for a given creative worth $100 in that 1-2pm window. Simultaneously, they analyse Google Analytics data for the same timeframe and identify a number of conversions – one of which is also a $100 transaction.
After matching these conversions, a Synthetic Impression for Facebook is added into the customer journey – following the same process with TikTok data to include a TikTok Synthetic Impression and creating a complete overview of the customer journey which can now be used for calculating the real impact of each touchpoint.
Once the customer journey is complete, a good model is used to assign credit to each of the impressions and clicks. (If a legacy MTA solution were to be used, all $100 from this conversion would be attributed to the last touchpoint before the purchase, in this example, the Google Search Ad. This clearly oversimplifies the customer journey and overlooks all the other touchpoints that contributed to the final decision/purchase.) Here, an AI Data-Driven Attribution model uses machine learning to provide reliable results for the entire marketing mix – including upper-funnel activities. It evaluates every interaction along the customer’s journey, uncovering how various touchpoints (including impressions) contribute to conversions. Based on these learnings, the model then estimates the probability of a user making a purchase after viewing the ad on TikTok. Based on its learnings, the model calculates that the probability of purchase increased by 0.5% after seeing the ad. Continuing along the customer’s journey, the model then evaluates the probability of conversion for a user who initially viewed a TikTok ad and subsequently, a few days later, clicked on a YouTube ad, spent a few minutes on the e-shop and visited 7 different pages. The model estimates the additional impact on conversion probability is +2%. In this way, the model systematically analyzes and attributes the probability of purchase increase for each touchpoint.
Now that we understand how each touchpoint increases the probability of conversion, we can proceed to the final step - channel scoring. During this phase, the model assigns credit to each touchpoint based on how much it increased the probability of a conversion. In our example, TikTok is attributed 10%, YouTube 40%, Facebook 20%, and Google Search 30%. It is important to note that this is just an example, and that in different scenarios the same channels will have different percentages assigned.
So we can see that Synthetic Impressions methodology in MTA provides the value of each touchpoint which allows users to create any number of custom dimensions on different levels of granularity, making the view suitable for any reporting across the marketing team.
New Privacy Laws
Another significant impact for some MTA practices has been new privacy laws, such as GDPR in Europe and CPRA in California. These laws mandate obtaining explicit consent from users before collecting and processing personal data; this restricts availability of third-party data, which most legacy MTA solutions rely on for tracking user interactions across various touchpoints and provides the conversions they require. Additionally, these new laws emphasize the principles of data minimization, which require organizations to limit the collection and processing of personal data to the strict minimum; this poses a further obstacle for some traditional MTA solutions as it leads to a reduced pool of data available for MTA analysis – ultimately impacting the granularity of attribution models. But have these new privacy laws been a showstopper for MTA?
Again, not really. More future-proof conscious MTA practices are evolving their technology by developing AI solutions that perform probabilistic path-stitching whereby parameters such as IP address, device type, geo location, type of browser, etc. can be measured; then based on these identifiers, path-stitching technology attempts to build a connection between visits which did not share the same cookie but, much like the estimation approach of Synthetic Impressions, will assume it is the same user and connect the visits into the same journey.
Third-party Cookie removal
One of the most difficult challenges for traditional MTA solutions has been the removal of third-party cookies. A fundamental tool for old-school MTA, they are set by external domains and were primarily used for tracking user behavior across multiple websites for advertising and analytics purposes, enabling the attribution of conversions as users navigate through various websites and platforms. Now blocked by many browsers, (and soon to include Chrome), legacy MTA increasingly struggles to connect the user interactions across the web accurately, leading to incomplete attribution.
Some MTA solutions, however, have turned their attention to first-party data from owned channels. While valuable, first-party data offers only a limited view of the customer journey, especially if users interact with multiple touchpoints outside owned channels. Browsers such as Safari are now deleting first-party front-end cookies after 24hrs, exacerbating the difficulty of measuring customer journeys even more.
The demise of third-party cookies has indeed stirred the waters for MTA, but it has also been the catalyst for alternative forms of future-proofing technology for MTA tracking and measurement. These might include implementing cross-device tracking techniques to connect user interactions across different devices without relying on third-party cookies at all. Techniques such as deterministic matching (e.g., through user logins), or probabilistic path-stitching which, as explained earlier, uses statistical models to identify patterns in user behaviour attributing conversions accurately across devices without the need for third-party cookies.
So let’s take stock. It seems that, despite these challenges, MTA stubbornly retains its value – adapting its technology to overcome each hurdle that stands in its way by continuously evolving with new machine learning techniques and future-proofing new, dynamic technologies.
So, amidst attention-grabbing proclamations in marketing discourse, such as "MTA is dead!", the reality is that MTA is very much alive. It is weathering the storm, tackling all the hurdles that have beset it, defiantly adapting to each challenge by developing effective, robust technology. While MTA is not flawless and cannot capture every nuance of a customer journey, it remains a potent tool for comprehending customer behaviour and refining marketing strategies and it would be a misjudgement to accept overly simplistic claims declaring its demise.