Often an overlooked piece of the marketing puzzle, attribution requires the development of a program that provides insight and feedback on which pre-revenue activities were most impactful in developing a pipeline or assisting in a revenue transaction. Establishing and tracking these contributions is crucial since it helps companies to effectively structure content calendars, assign budgets, and invest in—and deprecate—specific marketing activities for improved business outcomes.
In other words, attribution clearly identifies where you should invest resources and what you can expect the returns from those investments to be. A one-size-fits-all method of attribution doesn’t exist, since companies have unique marketing and sales tactics; however, there are commonly accepted practices that most companies utilize as a starting point.
Anonymous Touch – Pageviews or other anonymous digital events that are assigned an anonymized first-party cookie, to be later stitched with a known/identified user and subsequent events
Known Touch – Meaningful and quantifiable interaction with a campaign where the person shows explicit interest, like a hand raise. Typically this is something like a web form fill. These are often measured in ‘responses’ (e.g., Responded, Attended, etc.)
Single Touch Models
First Touch – This approach assigns 100% of the conversion value to the first known touch—most often, the first form that a website visitor submits (otherwise known as the identification event). Through this model, a record with the person’s information is entered into a company’s MAP/CRM system, giving credit to the source of the information.
Last Touch – This model assigns 100% of the conversion value to the last known touch prior to an opportunity being created (e.g. Primary Campaign Source in Salesforce). The Last Touch model, which is one of the most prevalent models used today, analyzes a person’s record, and gives all of the credit to that person’s most recent campaign response.
Last deliberate touch – Very similar to the Last Touch model, this approach excludes channels such as direct/none and referrals, since they don’t provide any insight into where or how a company should allocate budgets. These factors are only included if they exist as part of a specific marketing program.
Data Strategy & Pipelines
Turn user activity into actionable insights while reducing cost, complexity and privacy risk. Learn More →
Linear – With this model, each known touch along a visitor’s journey receives the same amount of credit. For example, if a person’s interaction with a site has 10 known touches, each of them would earn 10 percent of the credit. This approach is a good starting point for companies that need to determine which of their known touches are most valuable but still want to see the full picture when it comes to influenced campaigns.
Time Decay – This is a useful model for evaluating the effectiveness of a known touch, especially if the customer lifecycle is lengthy and the act of purchasing comes with a lot of deliberation. By taking this approach, companies assign a value to the most recent known touch and use an algorithm to detract percentages of that value based on how far in the past each earlier known touch occurred.
U-Shaped – Using this model, the first and last known touches share an equal—but disproportionately large—percentage of the credit. With a U-Shaped model, the first and last known touches are each typically assigned 40 percent of the credit, with the remaining known touches splitting the final 20 percent. (Companies can modify these percentage totals based on what makes the most sense.) The U-Shaped model works best for companies that rely heavily on of top-of-the-funnel marketing, since they’ll learn how a person was first introduced to their brand and what persuaded them to become a customer.
W-Shaped – As you might expect, with this model, the first, last, and chosen middle touches are all evenly split the credit (e.g. a campaign identifying a person record, the campaign before MQL is created, and the campaign before SQL is created). The W-Shaped model is often used when a company has a specific set of conversion-focused materials but is also pushing other types of advertising. The large, conversion-focused campaigns would split the attribution, while other materials are seen as assisting components to those campaigns.
Markov Chain – Markov chains determine the synergistic effect that different platforms have on one another by analyzing how people transition from one source/medium to another. With this type of attribution, statements such as, “if we remove channel X, overall conversions would decrease by Y percent,” can usually be made. For companies that are spending a lot on tactics that don’t directly generate conversions, this approach is especially useful, since it reveals what kind of impact removing one of those tactics will have.
Shapley Model – This model uses game theory to identify the marginal contributions that each channel is making toward an overall conversion. It focuses on every possible combination of channels that contribute to a conversion, identifying the ROI for each combination, and it uses this information to determine how much of a share each channel should receive. This model works best when it can evaluate the entire funnel, looking at actual revenue generated. And it’s an extremely useful tool for companies that need to analyze hard dollar amounts.
Custom – As its name suggests, a custom model can be anything that a company believes aligns well with its specific needs or structure. The previous models described are only suggestions. Some clients find they aren’t the most effective methods to use. If you’re considering a custom model, it’s a good idea to have a data engineer guide the structure of that custom model, as things can get complicated, especially at scale.
Self-Reported – A qualitative approach to attribution in whereby lower funnel conversions (e.g. eCommerce transaction, Demo Request etc.) are asked to self-identify where they first heard of the brand (e.g. LinkedIn, Podcast, Analyst, Paid Ad etc.). As any form of digital attribution is nearly impossible to perfect, coupling the process with self-reported attribution can provide a broader picture of where success is coming from. This is especially true for scaling companies and those with large founder-led growth motions, but less so for larger brands (e.g. Snowflake, Microsoft etc.).
Determining which attribution model is right for your company can be a complicated task. Ultimately, the purpose of these approaches is to assist a company’s growth, providing insight as to where investments should be made—both in terms of the right tools and the right employees. With this in mind, here are a few questions to ask yourself as you begin the journey toward improved attribution:
What Is The Source And Quality Of The Data?
Attribution models are data-dependent. Consistent and accurate collection of data is paramount to your program’s success. A customer journey can take many months (or years) and span dozens of touches, both anonymous and known. Many times the systems in a given marketing technology stack are not configured to produce the data required to satisfy any of the models above with any consistency. Before embarking on a given model, ensure each touch, campaign, and activity type is defined and tested against multiple use cases.
What Is Our Current Model? How Would A New Model Improve Outcomes?
This may sound simple, but ensure the existing model is clearly defined and data collection processes are documented. Too often marketing professionals and many corporate executives become enraptured with the newest trends in attribution and are drawn to a new method simply because it’s popular. Before you implement any changes to your attribution model, make sure you can clearly answer these two questions. Those answers will provide a lot of clarity and will dictate the approach that you take.
Who Will Be Using The Insights?
Remembering who the end-users are in this process is paramount to the success of any data project, including attribution. If you put a model in place that is too complex for the end-users to utilize, no meaningful insights or change will occur. Pay special attention to the amount of time that is necessary to produce meaningful insights, not to mention the amount of time that is needed for new team members to become competent using the data.
At the end of the day, attribution is all about tracking user interactions with your company and assuring that those interactions are as efficient and effective as possible, especially as it pertains to customer conversion. There is no such thing as a “best” model. A business-to-business technology company will likely use a very different model than one that operates as a direct-to-consumer operation.
The most important takeaway is that attribution is an ever-evolving practice. A company may change models over time, simply because the way it does business has changed. If you’re wondering when the time is right to reevaluate your attribution models, consider that approach if it feels like your corporate budgets aren’t carrying the same weight that they once did. It could be that your attribution model is no longer portraying accurately how much each channel is contributing to the bigger picture. A scalable program can only be built when you have a foundation of solid data from which to start.
If you are ready to talk about taking the next step, we would be happy to connect with you!