Understanding The Limits Of Attribution

While there is much discussion about attribution and its benefits, marketers seldom discuss what it can’t do. Unfortunately, this creates a perception with many marketers that attribution will solve all their multimedia problems and that their media mix problems will go away.

While I believe that the evolution of attribution and media mix modeling technology is one of the more significant online marketing developments in recent years, I also think that the technology in its current form has a ways to go before one can confidently say that the problem is solved.

Multi-event attribution aims to distribute the credit of a conversion to all advertising touch points that influenced the conversion. The distribution of credit aims to be proportional to the level of measured influence that the touchpoint had on a conversion.

An example is shown below:

Here, one wants to measure the influence on a display impression after a search click. One can then run an experiment where a pool of users is served a display ad and another pool is served a public service ad (PSA) as a control.

In this example, the conversion rates for the PSA view-throughs is 30% and for the ad viewthroughs it is 45%; hence, the incremental jump in conversion rate is 15%. One can then determine the attribution weights at 2/3 for the search clicks and 1/3 for the display click.

While this methodology is appealing, two basic problems make it intractable.

1. The ability to serve ads to a user. With display one can decide when and where to serve ads, but this is not possible on other channels like search and social.
2. Determining the weights for every possible funnel is practically impossible. If one wanted to determine the weights for all possible 5 event funnels, across 3 channels then 5^3=125 experiments would have to be run.

Due to these limitations most marketers adopt a simple attribution rule like first click, last click or even distribution. While theoretically incorrect, they provide a level of insight of channel interaction. This is useful in understanding how consumers behave. Currently, most solutions that I see in the industry stop here.

However, the value of such a solution is limited because it doesn’t answer the media mix problem, e.g. given the channel interaction what is the right allocation of media budget that will maximize the overall return on investment for the marketing dollar. It should be noted that some companies actually solve this problem algorithmically, but we will not delve into algorithmic methods here.

Second, attribution methods can at best directionally guide your budgets in the right direction, but they cannot guarantee the best media mix immediately. This is because channels interact and when you shift the budgets of one channel significantly it will have an effect on the performance of the other channel.

As a result, attribution and algorithmic technology will at best be locally accurate but claims of magically finding the globally optimal media mix should be treated with suspicion. Realistically, a good attribution and media mix approach will directionally indicate where to shift budgets and over time will converge to an optimal solution.

Third, attribution analysis requires one to use long look back windows  to capture the entire multi- channel effect of a different sales funnels. However, this often means that the behavioral insight you get from analyzing the data is often out of date with the present market conditions.

I offer the following tips and recommendations as you consider different attribution platforms, solutions or methods:

• Simple attribution will give you insights about your customers and how they come down the purchase path. However, they will not give you clear insight as to how to shift the media budget to maximize ROI. This requires an optimization layer on top of the attribution technology.

• The media mix recommendations coming from any attribution method should be considered directional. Further, do not change budget allocations more than a few percentage points at a time. If you shift budgets and find that your performance improves, shift it again and measure performance. While a large shift may improve performance significantly, it is a risky approach as the media mix models could be off.

Despite the present short comings, marketers will do well to use attribution techniques to measure, analyze and optimize their marketing campaigns. The science of attribution and optimization is continuously evolving and I expect that in a few years there will be solutions that come close to solving the media mix problem. Until then, use the technology and the science with a good dose of judgment.

Some opinions expressed in this article may be those of a guest author and not necessarily Search Engine Land. Staff authors are listed here.

About The Author

Siddharth Shah

Siddharth Shah is head of web analytics, digital strategy and insights at Adobe. He leads a global team that manages the performance of over $2 BN dollars of ad spend on search, social and display media at Adobe.