Multi-touch attribution is only going to get harder due to platform changes and a focus on privacy but there are a few approaches used within MTA that can produce actionable insights.
Attribution is an analytical method that takes a lot of user-level data and tries to measure the impact of specific tactics on a positive outcome, such as a sale. In its algorithmic form, it is supposed to be an improvement on quaint methods like last-click-takes-all, which are obviously wrong, but very convenient. The purpose of attribution is to give fair credit to the tactics – placements, creative ideas, formats – that work.
The term “attribution” refers to several types of models: Sales Attribution, Location Attribution and Multi-Touch Attribution (MTA). When people say “attribution is dead,” they’re usually referring to MTA and not the other two types of models. Sales attribution and location attribution are continuing to gain adoption within the industry as more media is executed through addressable channels and consumers increase mobile engagement and retailers seek to monetize their sales data.
Multi-Touch Attribution isn’t dead, it’s just hard
Multi-Touch Attribution (MTA) is not “dead” but it has always been hard to accomplish. True MTA was always an aspirational goal, as no single approach, or vendor captured all the touchpoints in the consumer journey. Vendors like Adometry (now Google 360) had specific limitations in mobile exposure due to the inability to tag on Safari or iOS. Thus, some brands were analyzing data on a sample of one percent of site traffic.
MTA is only going to get harder due to platform changes and a focus on privacy. Platforms like Google, Amazon and Facebook have restricted cross-platform tagging for their proprietary solutions, while party vendors (like C3 Metrics, Nielsen, Neustar/Marketshare and Visual IQ) are pixel-based solutions with limits to where their pixels catch consumer signals.
First, Google removed third-party tracking from YouTube and then Facebook, always restrictive in tagging, to sunset the ability to DCM tag on its site. In addition, as a reaction to GDPR, they closed many other linkages to their site. One major example was Google’s announcement of eliminating the Google ID from DCM records and log files, forcing consumers who wish to track in Google’s ecosystem into their Ads Data Hub product. Apple rolled out ITP 2.0, and Mozilla followed suit in Firefox that drops third-party tracking pixels for privacy and speed purposes.
But it’s going to get even harder. Emerging high-growth media channels like OTT, ATV and podcasts have yet to have a consistent measurement solution. California passed its interpretation of the EU GDPR, called CCPA, which comes into effect in January 2020 so we anticipate more platform reactions that close more tracking abilities.
Some MTA models are still viable
But it’s not all bad news. A few approaches used within MTA can produce actionable insights. One example is through reimagining Media Mix Modelling (MMM) by applying a channel/partner based approach. Instead of modeling broad level digital, social and mobile channels, this approach goes deeper into comparing the likes of Google-owned and operated, individual publishers, Facebook and Twitter, and calculating their media elasticity in that way. Another approach is to leverage experimental design and conduct incrementality tests using Ghost Ads or Randomized Control.
In the vein of utility, the native platforms that offer their proprietary attribution, such as Facebook, Google and Amazon, do provide value. However, expectations should be set on the tracking limitations of each solution.
In summary, as George Box said, “All models are wrong, but some are useful.” While attribution has never achieved the promise that it was supposed to solve, sales attribution and location attribution models continue to be adopted as they connect deterministically digital media activity to a business outcome. While MTA will continue to be challenged, keeping lowered expectations of what insights MTA can provide, balanced with an understanding of the data limitations from platform solutions, can still yield insights.