Smart Attribution Modeling
Depending on the size and scope of your advertising and marketing spend, you may have spent time and effort thinking about attribution modeling. Different organizations have very different approaches to attribution.
To this end, developing a valuable attribution model that serves your goals and your business can take many forms. Herein, I’ve put together some criteria that’ve been used effectively by a number of organizations we’ve worked with to inform decision-making and use of attribution methods and models.
First Things First: Determine Your End
The most important questions senior marketers need to ask going into an attribution initiative, at any level of investment, include:
- “What is the purpose for attributing (estimating) media value?” You may be surprised how often that answer is ill-defined. Make sure you can answer, in simple business outcome terms, what the purpose of your attribution is. All else fails if this step is missed.
- “How logical, defensible and credible is a potential attribution methodology?” While attribution, by its nature, is rarely deterministic, it is requisite that a methodology is credible and has robust basis, or a raison d'etre, if you will, if it is to add value. The understanding individuals often develop is an appreciation that the assumptions underpinning any attribution strategy are tenants of the strategy itself.
The right answers for any brand depend on keeping the end in mind and knowing the expected outcome. So the logical starting point is defining your purpose for attributing media value, as described in that context. For example, “to get the best ROI from our advertising investments.”
3 Strategic Attribution Model Levers
In the spirit of keeping it simple, we think in terms of three strategic attribution levers that an organization can benefit from. These strategic levers are used to inform both the attribution model selection and the weighting of channels. They are as follows:
- Engagement: Measures a customer’s depth of interaction and potentially, the relationship with the brand.
- Recency: The amount of time lapsed since the last touch. For example, all other things being equal, a touch yesterday is more valuable than a touch 45 days prior.
- Intent: Identifies a need the user has or information the user is seeking. Intent is specifically valuable in search, and sometimes in social media. Lead generation programs demonstrate intent, as well. The point of considering “intent” is that it prequalifies traffic in a meaningful way. If the consumer exhibits intent-driven behavior — that should be weighted heavier in your attribution thought process.
While the decision to “attribute” always means judgment is incorporated, the credibility of the attribution is higher when media touches are evaluated within the three strategic levers and should always be based on the nature of the interaction — or lack thereof. If a user did not engage with an ad, then the amount of interaction is lower or even zero.
The following chart breaks out major channels and how you might evaluate each of the strategic levers discussed above.
The ‘Bonus’ Lever: Measurability
Measurability is the “fourth” strategic lever, and can be considered optional for very large brands utilizing traditional non-digital channels extensively. A channel that has evidence associated with its performance is one that can be weighted accordingly. When a channel is measurable, the weighting in the attribution model can be scaled to leverage the predictability of that channel; thereby, improving the efficacy of the attribution. It is a reality that some channels however, will have hard measures, while others require more assumptions and inferences.
Brands should give thoughtful consideration to not inadvertently “reward” a channel, simply because it is hard to measure — and, by the same token, not unnecessarily punish them, either.
Over- or under-weighting channels that have weak evidence of conversion value can actually reduce the performance of the overall media mix.
Viewability and Display-Weighting
While reach, frequency and targeting are hallmarks of display advertising, it has the widely known challenge of “viewability.” Viewability is when an ad is served (and paid for) but a consumer does not see it.
When the objective is to improve the ROI of the media mix, ads that are never seen (un-viewable) should be accounted for in the attributed value of the channel.
One way marketers simplify account viewability concerns is by deducting the percentage of ads that can never be seen on a percentage basis when weighting online display in the model. Bear in mind, “viewable” generally means that only part of the ad was viewable for 1 second. Specific viewability metrics should be discussed and negotiated with media outlets or networks you work with.
How Much Is Viewable or Unviewable?
A recent study done by Google identifies that many display ads are never viewed; therefore, the weighting of display ads should consider this reality (opens as a PDF).
Here are some of the issues with viewability that should influence the weighting of display.
- 1 percent of all impressions measured are not seen, but the average publisher viewability is 50.2 percent.
- The most viewable ad sizes are vertical units. Above the fold is not always viewable … Worth considering when weighting display.
- Page position isn’t always the best indicator of viewability.
- Viewability varies across industries. While it ranges across content verticals and industries, content that holds a user’s attention has the highest viewability.
- The most important thing is to give viewability consideration and weight based on your own experience.
Frequently Used Attribution Models
Let’s summarize the most popular attribution models in order of frequency of use, and as based on field experience. There are many more models you may consider, and this list is not intended to be exhaustive.
- Last Click: 100 percent of the sale is credited to the last click, given its immediacy in driving the sale.
- Linear Attribution: Equal weighting is given to all touchpoints, regardless of when they occurred. Its strength and weakness is in its simplicity. Not every touch is equal and for good reasons that we’ll describe in some detail below.
- Time-Decay Models: The media touchpoint closest to conversion gets most of the credit, and the touchpoint prior to that will get less credit. This is the best of the simple approaches. It does not, however, account for brand discovery.
- Position Model: Position model utilizes intuition and assumption to spread the weights of touches over time, heavying up the first and last touches, and considering the middle touches to spread the difference evenly across them. To be clear, this model presupposes “zero” brand awareness — and, therefore, that every customer “discovered” the brand from a (display/banner) ad impression, for example. Blanketing an audience in advertisements can provide great reach and frequency. It also sets a lot of cookies, which can be used to set the first “position.”
Pointers for Getting Started
The closer you can get to individualized attribution vs. broadcast attribution, the stronger the returns. For example, attribution by segment can provide insights you miss when measuring the aggregate.
Channel measurability should be weighted accordingly. Non-measurable channels should be measured by depth of observable engagement.
The Time-Decay model is widely considered a good place for brands to use when getting started in media attribution. Brands can simply insert logical and evidence-based assumptions and customize the half-life of decay based on the Three Strategic Levers described above.
Follow-up discussion and analysis can refine your thinking and allow you to provide a rationale that helps achieve the most credible, logical and valuable attribution capability.