How Machine Learning Can Help With E-Commerce Attribution

Martech companies often throw around the terms artificial intelligence and machine learning, and marketers are left wondering how to sort through the hype and understand if and how these technologies help. This problem is especially pronounced in attribution, which is crucial to optimizing marketing campaigns but suffers from doubts about its very possibility. Claims that “attribution is dead” are perennial.

It's not impossible for e-commerce companies to achieve a meaningful level of attribution that can help them optimize marketing and drive more conversions. And machine learning, by which I mean models that leverage historical data to project and improve future outcomes, can help by sorting through an amount of data that no team of human analysts could delve through on its own.

Machine learning can help e-commerce companies in three ways. It can analyze what for most marketers is an overwhelming amount of marketing data; develop a dynamic attribution model tailored to each site; and forecast revenue based on visitors’ purchase probabilities. Let’s explore all these capabilities to understand why attribution isn’t dead and how machine learning is increasing its utility for marketers.

Make Sense of an Overwhelming Amount of Data

The dream of attribution is to understand not only how marketing influenced a purchase but the many touchpoints a customer encountered on their path to purchase and even how much each of them influenced the final decision. Critics are correct that putting this together would be impossible — if humans had to pore over all the data and if an effective attribution model required understanding exactly how each and every customer made their decision.

But neither of those standards needs to be met to achieve an effective attribution model that can help companies understand who is visiting their sites, who is making purchases, and why.

First, machine learning can help marketers analyze millions of customer journeys, recording the many touchpoints they encounter and then using analyses of purchase probability across channels to draw conclusions about how much each touchpoint affected the final purchase. This would be impossible to do manually. However, with a model that can analyze an individual company’s data while learning and improving over time, an analysis of this type and scale becomes possible.

Second, critics of attribution, especially multitouch attribution, which attempts to understand the customer journey across multiple touchpoints, often point out that models, at best, estimate how much each touchpoint influenced a customer; they can’t decipher exactly how much paid search, email or social drove a purchase. This is true but irrelevant to the usefulness of attribution.

Marketers don’t need to understand exactly why each individual customer made a purchase to gather intelligence that helps them better calibrate their strategy and drive more revenue. They just need to understand holistically how effective each channel is and how different campaigns and creative perform within those channels. Machine learning-enabled multitouch attribution can deliver precisely that, helping marketers understand where to dial up efforts, by how much, and which channels need to be re-examined.

Develop an Attribution Model for Your Specific Site

Another aspect of attribution that would be impossible without machine learning is site-specific modeling. It’s one thing to analyze a large dataset of customer behavior across e-commerce sites and arrive at an approximate understanding of how effective various channels are in a competitive set or industry. It’s another to create a model that understands how effective each channel is for a specific company.

With machine learning, marketers can analyze a given seller’s individual customer journeys and show, for example, what percentage of site visitors driven by email make a purchase, how much revenue to expect based on visitation rate over time, how different creative and campaigns affect purchases, what the returning customer rate for each channel is, mean number of days between first and subsequent orders, and so on.

Altogether, this data helps marketers quickly assess which channels they should lean into and which they’re undervaluing. It can also help them flag channels declining in efficacy to see whether a different channel strategy can optimize results. But this is only possible to do on a per-site basis because machine learning enables custom models based on individualized analyses at scale.

Forecast Revenue for Each Site Visitor

In e-commerce, the basic equation for success is driving traffic, conversions and conversions for larger sales at that. However, not all customers are created equal, and not all marketing channels and tactics drive equally high-value customers. The granularity of machine learning-powered attribution allows marketers to solve this problem.

If marketers are able to track their site visitors and customers across channels to understand not just traffic and conversions but how big the sale is and how many times customers driven by each channel tend to return, they can optimize their marketing mix for more than the first purchase. They can lean into channels, campaigns and creative that drive the high lifetime value customers on which prosperous brands are built.

Even better, the scale and granularity of machine learning can help marketers gauge how much time it typically takes a customer to make a second purchase and when re-engagement comes off the table. This helps marketers time campaigns optimally so that if their subsequent purchase rate drops off dramatically after, say, day 90, they can get in touch with first-time customers before then to boost retention.

Rely on Machine Learning to Capitalize on Effective Attribution

Due to signal loss, elevated concerns about privacy, and the turn against the so-called 360-degree view of the customer, many marketers have soured on the possibilities of attribution, especially multitouch. Similarly, many wonder whether technologies like AI and machine learning are overhyped.

While not a miracle worker that can show exactly which factors drove each customer to make a purchase, machine learning-enabled multitouch attribution is more than able to help marketers more effectively strategize. Machine learning and attribution aren’t a pipe dream. They're real technologies and products used by thousands of marketers. They are a reality — and one that's only growing more effective with advances in machine learning technology.

Phil Dubois is the CEO and co-founder of AdAmplify, a provider of the next generation of marketing attribution software.