Marketing Attribution: Data-Driven Model Methodology

 

In the final installment of our three-part blog series on marketing attribution, our goal is to help you establish a methodology for implementing a data-driven attribution model that makes the most sense for your business, and how to think more broadly about the relationship between data analytics and marketing decision making. At this stage, let’s assume that we have already established the following:

• We’re tracking the right data and metrics
• Our data is organized and collected using the right tools and technology
• We’ve established clear goals and corresponding KPIs for measurement

With the above elements in place, it’s important to determine your ideal marketing attribution window. A marketing attribution window is a specific period of time during which you assign credit to a particular marketing activity for a desired action or conversion. The attribution window is the time frame in which you identify which marketing channel or touchpoint was most effective in influencing a consumer to take a particular action, such as making a purchase or filling out a lead form. For example, if a consumer clicks on a paid search ad, visits a website, and then makes a purchase, the attribution window is the period during which the marketer assigns credit to the paid search ad for influencing the purchase. The length of the attribution window can vary depending on the marketing channel, the industry, and the specific marketing goal. Related to our previous example, if your business is mostly transactional in nature then a good standard attribution window to begin with is a timeframe of 7 days. Businesses that feature longer customer journeys and lifecycles (i.e., B2B) are likely to require larger attribution windows. Tools like Google Analytics, which features a 30-day “lookback” window as a default setting for conversion analysis, can also help you determine the right attribution window.

Building a Data-Driven Attribution Model

Earlier in our blog series, we introduced some commonly-used attribution models that have been widely tested and implemented with success (i.e., first-click, last-click, linear, time decay, position-based). If your business has yet to experiment with marketing attribution in any way, we recommend beginning with one of those before diving deeper into data-driven modeling. At a minimum, you’ll be able to identify gaps in data and information that may exist within your organization and determine the steps needed to make sure your data is logged and organized properly to truly power a data-driven attribution model.

For those businesses with significant volumes of historical data and a strong technology infrastructure in place, developing a data-driven attribution model is much more realistic and will ultimately prove more beneficial. Building an in-house, data-driven model is no easy task and requires significant commitment in both time and resources. So where do you begin? There are two methodologies that have produced prominent data-driven models that have been widely adopted in the fields of data science and information measurement.

The Markov (Chain) Model

The Markov data-driven attribution model (or Markov Chain model) is a method of assigning credit to different marketing channels by using a mathematical algorithm that consider the sequence of touchpoints a user has with a brand before converting. In the Markov model, each touchpoint is considered a state in a sequence, and the probability of moving from one state to the next is calculated based on historical data. The model uses a matrix of transition probabilities, known as a Markov chain, to calculate the credit assigned to each touchpoint.

To illustrate how this model works, let’s say a customer sees a Facebook ad, then clicks on a Google search ad, and finally makes a purchase. The Markov model would consider each of these touchpoints as a state, and it would calculate the probability of moving from one state to the next based on historical data. For example, it might find that 30% of people who see a Facebook ad then click on a Google search ad, and 10% of people who click on a Google search ad then make a purchase. Using this information, the model assigns credit to each touchpoint based on how much it contributed to the final conversion event. For example, if the total credit assigned to all touchpoints is 100, the model might assign 25% credit to the Facebook ad, 50% credit to the Google search ad, and 25% credit to the purchase itself.

The advantage of the Markov model is that it takes into account the entire customer journey, rather than just the last touchpoint before a conversion. This can help businesses make better decisions about which marketing channels to invest in and how to optimize their campaigns for maximum impact.

The Shapley Valued-Based Model

The Shapley value-based data-driven attribution model is a method where credit is assigned to each marketing channel based on the incremental value it adds to the actual conversion event. Named after its creator, Nobel Prize-winning mathematician Lloyd Shapley, this model works by comparing the value of each channel when it is used in combination with other channels, versus when it is used alone.

To illustrate how this model works, let’s say that three marketing channels are involved in a customer journey: email marketing, social media advertising, and search engine marketing. The Shapley model would first calculate the value of each channel when used in combination with the other two channels, then compare that to the value of each channel used alone.

For example, the model might find that when all three channels are used together, the conversion rate is 10%. When email marketing and social media advertising are used together, the conversion rate is 6%. When social media advertising and search engine marketing are used together, the conversion rate is 8%. And when email marketing and search engine marketing are used together, the conversion rate is 7%. Finally, when each channel is used alone, the conversion rate is 2% for email marketing, 3% for social media advertising, and 4% for search engine marketing. Based on these calculations, the Shapley model would assign credit to each channel based on its incremental value in the conversion event. In this example, social media advertising would be assigned the most credit, followed by search engine marketing, email marketing, and then any remaining credit would be assigned to interactions that were not attributed to any specific channel.

The advantage of the Shapley model is that it takes into account the interdependence of marketing channels and the incremental value each channel adds to the conversion event. This can help businesses make better decisions about which marketing channels to invest in and how to optimize their campaigns for maximum impact. However, it can be more computationally intensive than other attribution models, especially when dealing with a large number of marketing channels.

In Summary

By basic definition, the primary goal of implementing a marketing attribution model is to better understand the buying journey of your customers so that you can better align your marketing strategy and tactics to positively influence that journey and maximize conversion potential. As we know, the customer journey itself continues to grow more complex with time. Depending on your source, the average number of touchpoints in a typical customer journey varies quite a bit. For example, Salesforce says that 6-8 touchpoints is a good average, HubSpot settles on 8 touchpoints as a good benchmark, and data from Google suggests that it may be as many as 20 touchpoints prior to purchasing a candy bar and more than 500 touchpoints for customers purchasing a flight. The important thing to remember is that each touchpoint can make or break a customer experience and a conversion for your business. As a marketer, if you’re able to apply the information garnered by proper data collection on how customers are interacting with your paid, earned, or owned media, that information can ultimately be used to turn an interested consumer into a paying customer. Whether you decide to implement and test a proven attribution model like the Time-Decay model as a baseline, or you possess the data and capabilities to develop a more complex data-driven model, there is no question that some form of marketing attribution model will be necessary to ensure your marketing and advertising tactics properly align with the respective buying journey of your target audience today, and in the future.

A Final Word

When developing this series on marketing attribution, one research paper that we found quite useful is titled “Bridging marketing theory and big data analytics: The taxonomy of marketing attribution” by Dimitrios Buhalis and Katerina Volchek, originally published in the International Journal of Information Management in February 2021. The research and subsequent findings of the authors was motivated by their attempt to create a more universal theory and practice of attribution models through data collection and analysis, taxonomy development, and refined terminology. The final product is quite dense, but it offers a conceptual framework of marketing attribution while organizing models into Attribution Capabilities and Attribution Facilitators, which may be of further use in developing your own attribution model.

At Location3, we’ve been testing marketing attribution since the initial launch of GA360 and have successfully implemented a variety of attribution models on behalf of our partners since then. Today, we’re leveraging data analytics tools like Google Analytics 4 and Campaign Manager 360 to help a cross-section of our partners implement baseline attribution models. As the digital marketing and ad tech industries continue to enhance protection of consumer privacy and move away from third-party audience tracking and toward first-party data, we’re optimizing our own technology stack accordingly. We’re also actively supporting partners with implementation of technology like Customer Data Platforms (CDP) and other powerful data analytics tools that will provide better marketing attribution, while empowering our teams to create more sophisticated measurement systems, better understand incrementality and diminishing returns in marketing, optimize media budgets and properly align with every customer journey we aim to positively impact.

If you’re as fascinated by marketing attribution and data science as we are at Location3, we welcome the opportunity to speak with you and learn more about your data goals, success stories, challenges, and ideas for the future.