One of the biggest challenges in marketing is quantifying the influence of each customer touch point. Even marketing teams that can map the entire customer journey have trouble identifying the exact moment or event that compelled a prospect to become a customer. This blind spot has always foiled retail marketers, but the issue has become more urgent as CMOs demand greater attribution and measurement for every ad dollar spent.
With the customer journey growing more complex every day, due in large part to the proliferation of screens and cross-channel marketing, retail brands now require robust attribution models to understand how different channels contribute to desired outcomes. Fortunately, attribution models have come a long way since marketers first realized that the last customer touch point does not necessarily tell the full story. Algorithm-based multi-touch attribution (MTA) models that have surfaced over the years undoubtedly perform better than click-based deterministic attribution models, but there is one model grounded in game theory that has proven to be the most accurate for marketers.
The Decline of Click-Based Attribution
Attribution is as much of a data challenge as it is a marketing challenge. Customer journeys today often resemble a roller-coaster, with the customer’s interest rising and falling after every cross-channel impression, each accumulating a trail of data along the way before that final conversion. While click-based attribution models yielded some value for early direct-to-consumer brands, they don’t make as much sense for retail brands that depend on both in-store and online shopping, not to mention offline marketing.
Once offline touch points enter the equation, attribution through click-based models becomes virtually impossible. Even the first MTA models that attempted to measure multiple channels were not sophisticated enough to handle the chaotic customer journeys seen today, especially when the data was incomplete or siloed. The U-shaped attribution model, for example, arbitrarily gives the most weight to the first and last touch points, while the intermediate events always receive less weight. A customer journey that started from a mobile ad, followed by a trip or two to a physical store during the research phase before completing the conversion online through an ecommerce store, is difficult to map, let alone derive insights for the post-purchase experience.
Advertisement
To optimize attribution models for the modern retail experience, marketers are turning to a concept in game theory that can help measure the impact of every channel at each stage of the customer journey.
Game Theory Meets Attribution
In game theory, a “cooperative game” is a game in which players cooperate with one another in coalitions to achieve a shared aim. Despite having a shared aim, each player may contribute unequally to the outcome of such a game — some players may deserve the bulk of the credit, and others may barely increase the coalition’s probability of winning.
In 1951, Nobel Prize-winning economist Lloyd Shapley introduced the Shapley value, a weighted average of each player’s marginal contribution to all the possible coalitions of players in a cooperative game, in order to calculate the fair distribution of gains and costs among all the players. Applied to MTA, the players in a cooperative game represent various marketing channels, and the shared aim is producing a conversion, whether that is persuading a customer to make a purchase, sign up for a loyalty-rewards program or agree to receiving promotional emails. As a result, the Shapley value can be used to estimate how much each channel, from email and social to physical stores and in-person events, adds to the probability of converting a customer.
Like any model, the utility of Shapley values in retail marketing rests on a business’ data infrastructure. If a retail brand has invested in technologies that pull real-world customer data from multiple sources, then it is likely that brand can leverage Shapley values to draw correlations among different touch points and measure the impact of each channel.
AI-Driven MTA at Scale
Data-driven retailers are already leveraging Shapley values at scale through machine learning tools such as long short-term memory (LSTM) neural networks to analyze historical data from the lookback window. Some MTA models can index each marketing channel to show on which days within the lookback period the user interacted with a given channel, allowing for long-term dependencies among these data points, such as a sequence of customer interactions days apart from each other, to detect meaningful patterns over relatively long periods of time. This enables marketers to calculate the total attribution percentage for each touch point. The neural network can thus learn from the historical data and estimate the probability of conversion for all existing combinations of touch points and sequences of interaction across the entire customer journey.
For example, a company’s customer data platform (CDP) might contain historical data from five different channels in a marketing campaign: direct website traffic, paid search, social media, display ads and email. The Shapley value-backed LSTM model will compute the probability of conversion for each touch point combination and sequence, so if one path to conversion runs from direct traffic to social media to email, and another from social media to email to direct traffic, the LSTM model will provide different probabilities of conversion for both journeys — along with all the other combinations of those five channels.
The intelligence gleaned from a Shapley values analysis frequently torpedoes a brand’s prior assumptions about their own customer experience, as marketing teams are finally able to get a detailed understanding of the entire customer journey. By finding the total attribution percentage of each channel, for example, retailers can determine which channels are generating empty traffic, and which ones are driving most of the real conversions. And by examining the daily breakdown of each channel’s contribution to conversion, retailers can see which channels are having a lasting impact on lead generation over the entire lookback period, and which channels only performed well toward the end of the window.
Customizing Models
Once marketing teams deploy a tech stack that has the machine-learning capabilities to run MTA models at scale, they can then customize their models according to what makes intuitive sense to their business. Retailers should ask what conversion rates are like throughout the customer journey, from the introductory phase to purchase, and all the intermediate steps and touch points in between. By doing so, retailers can decide whether they should recalibrate their attribution models to fit specific needs.
Once retailers are running real-world data into their models, they can begin to measure key performance indicators and the channels that are contributing the most in the cooperative game known as retail marketing.
With insights from Shapley values MTA, marketing teams can attribute credit for outcomes across multiple channels with unprecedented precision. Marketers can then develop strategies and campaigns that reflect the reality of actual customer journeys rather than unverified assumptions about user behavior.
Dilyan Kovachev has more than 10 years of experience as a data analyst, performance optimization manager, and more recently a data scientist in the fields of finance, martech and education. Kovachev is as passionate about helping people as he is about writing code and crunching numbers. As a Senior Solutions Engineer at Treasure Data, Kovachev helps internal teams and clients achieve their goals by demonstrating the platform’s technical functionalities, working with clients on their platform integration, and building custom data-driven solutions for various marketing use cases and business problems.