By Justin Norris July 6, 2018
Sometimes I think asking, “Which attribution model do you prefer and why?” would be a great (marketing) conversation starter. From single-touch to complex regression-based analysis, some marketers are passionate about a particular method while others are still contemplating which is the best option. The topic sparks an interesting discussion.
Of course, all models are simplified approximations of an infinitely complex reality, and, no attribution model is perfect. Attribution models attempt to estimate the influence of your various marketing campaigns on human behavior that is unpredictable, irrational and fluid in nature. There’s no way of actually knowing that your white paper or webinar was responsible for 33% of the purchasing decision and therefore should receive a third of the credit. But, even with the flaws of attribution, applying the appropriate model, understanding the data it’s generating, and applying the directional insights will help you make better marketing decisions.
In this post, we’ll explore the different models and why you might use each one.
Before settling in on a particular attribution model, assessing your needs and being realistic about what you want to accomplish will assist in your model decision.
Once you know what you want to achieve, then you can select a model that’s appropriate for you. (And I should note, unless otherwise stated below, all models discussed are as defined within Bizible’s platform.)
Stemming from the philosophy that a sale cannot happen if a customer doesn’t know you exist, a first-touch model applies 100% of attribution credit to the first tracked marketing interaction, which may occur before the person even enters your marketing database. The model itself is simple, and data analysis is less complicated. In a simple sales cycle with a quick or transactional sale, it’s very easy to see marketing effectiveness and contribution to revenue. The challenge with a first-touch model is data collection, because you need a way to capture and store the anonymous first touch and then associate it with the person when they eventually enter your lead database. You can solve for this challenge with custom tracking script and Bizible tracks this out of the box.
In more complex sales cycles, first-touch attribution acknowledges the brand awareness stage, highlighting which of your early marketing efforts were most successful at attracting new customers to your product or service. If you seek to gain insight into top-of-funnel activity, then a first-touch model can be useful in providing answers. If you want to know marketing influence in later stages of your sales cycle, a first-touch model falls short as it only tells part of the story by overvaluing early-stage efforts and ignoring subsequent campaigns.
Going beyond brand awareness, a lead creation model attributes 100% credit at the point a customer is interested enough to provide contact information and essentially, becomes a “lead.” For example, if a customer visits your website three times and on the fourth occasion, completes a form for more information, the marketing effort that drove the fourth visit would receive 100% of revenue credit. The philosophy here is the campaign that converted a prospect to a lead is the most significant. Many organizations often start with a lead creation model because it provides an excellent introduction to attribution and the set-up is relatively straightforward.
Like first-touch, this single touch, simple model does not provide a good representation of longer and more complex sales cycles; for that, you need a multi-touch model.
A Linear or Evenly-Split model gives equal weight to every touchpoint with the rationale that every marketing effort is essential to moving a prospect through the sales pipeline. The challenge with this model is it oversimplifies the marketing process and fails to take into account the context of when the interaction occurred when giving credit.
For example, let’s say a person enters your database, consumes a few blog posts and then – a few months later – attends a VIP dinner and soon after is added to a new opportunity. With an evenly-split model, the casual content consumption that did not occur in proximity to any meaningful funnel event would get the same amount of credit as the high-touch dinner that likely made a much bigger impact on the sale. If you relied on this model exclusively, you might easily draw some inaccurate conclusions about relative channel importance.
Nevertheless, a Linear model can still provide some insight into which marketing programs are impactful. If you are tracking attribution using Marketo and Revenue Explorer, this is the only multi-touch option available.
U-Shape is a simple multi-touch model that distributes credit between the early-stage touches to provide a more balanced view of which channels are generating new names in your database. In this multi-touch model, 50% of the weight is assigned to the first touch and 50% to the lead creation touch. The philosophy behind it is to emphasize lead generation while also sharing credit between the various touches required to grow your database. For this reason, I prefer it over either a First-Touch or Lead Creation single-touch model for evaluating lead generation activities.
A W-shaped model is very similar to a U-shaped model except it acknowledges a third milestone, opportunity creation. Each primary stage of the sales cycle, first touch, lead creation and opportunity creation, is attributed with 30% of revenue and the remaining 10% is split between the other touchpoints. A W-shaped model is one of the most popular attribution models as it gives marketers a well-rounded view of the marketing campaigns leading up to the opportunity creation stage.
What’s missing in a W-Shaped model is insight into any activities that occur after the opportunity is created. For example, let’s say you organize a special event for customers and later stage prospects and then several opportunities close soon after. With a W-Shaped model, the significant investment in this event wouldn’t receive any credit.
Similar to the W-shaped model, a full-path model also acknowledges major milestones in the sales cycle, now extending all the way through the revenue stage. Each significant stage receives 22.5% of the credit with the remaining 10% spread across touchpoints in between. The Full-Path model is obviously more complete than the W-Shaped model and is arguably more sensibly-weighted than an evenly-split / linear model, as the touchpoints that occurred in nearest proximity to important funnel events get a much higher percentage of credit. This can produce reports that better reflect the “actual” impact of these important activities while still giving credit to everything. For businesses with a complex sales cycle who want full visibility, a Full-Path model is a smart choice and remains easy and simple to implement.
A more advanced multi-touch option within Bizible is the Custom model. With this model, you can define custom stages in the sales cycle in addition to those included in the Full-Path model—a common one to add is an “MQL” stage. You can then define your own percentage weightings for each stage based on your unique business model.
This model offers more flexibility and requires some extra configuration. Its relative freedom also brings a certain level of risk, as the marketer might have inaccurate assumptions about the relative weightings that the different stages should receive and thereby create misleading distortions in the model.
Companies may want to run a Full-Path model first, then as knowledge of their unique sales process deepens, transition to a Custom model to achieve a more tailored approach.
This model uses the same stages as the Custom model, but in this case, the machine makes recommendations for weighting credit between the various stages, representing the relative importance to winning a deal based on three criteria:
The algorithm is not random—Bizible based it on millions of touchpoints and buyer journeys. Notice in this example, that the product demo stage is now receiving 10% credit, demonstrating the significance of this event in the sales cycle. You can use the insights from the Machine-Learning model to refine and alter your Custom model, ultimately producing a machine-learning influenced model that incorporates human insights specific to your organization.
In a Tactic-Weighted model, credit is allocated based on the importance of the specific marketing tactics involved. For example, attending a webinar may get more credit than downloading an e-book, and attending a prospect VIP dinner may get even more.
This type of model—or one that blends it with a position-based model defined above—makes a lot of sense to many marketers, who intuitively know that spending four hours at a high-touch event naturally carries more weight than casually perusing web content.
This is an advanced model that is not available out of the box in any platform I’m aware of, but is something an analytically-mature organization could engineer within a BI tool.
One of the nice things about Marketo’s acquisition of Bizible, is marketers now have more model options to choose from, single-touch to multi-touch, simple to complex. To some, the options may seem overwhelming. My advice: take an inventory of your needs and start with what’s attainable. Remember, you can always transition to a new model as your knowledge and understanding grows. No model is perfect, but attribution will help you gain insight into your customer journey and the relative influence of your marketing efforts.
In my next post, I’ll address how to leverage your attribution data to fine-tune your marketing strategy.
Join us for an insightful and informative webinar, Bizible Essentials for Marketo Users on July 10 at 1:00 ET. We’ll explore the differences between Revenue Explorer and Bizible, the solutions Bizible offers and the impact on your daily operations. Reserve your seat here.