Accelerate Client Growth With Ecommerce Platform Apps
Ask almost any OPM about their biggest challenges and one of the ...
Anyone who has spent any time modeling in Excel has made the mistake of building a formula that refers to itself. When you succeed at creating this formula, Excel kindly notifies you that you have made a circular reference error in your formula. For me, this happens at the precise moment that is the intersection between when I think I have actually cracked the code to get THE ANSWER, and the moment when I realize that I had dramatically overengineered the calculation I needed to approach plausibility.
This is a good guiding principle we employ at DMi Partners on the complicated process of attribution: Avoid analyses where the stated or implied goal is to get THE ANSWER. Instead, focus on ways to use attribution to identify plausible insights that might otherwise be lurking in the background.
One of the first errors marketers often make when they set up an attribution analysis is confirmation bias. Determining what you think the analysis is going to show at the outset and building an attribution framework to support that hypothesis is the easiest way to waste time and effort with attribution. There is no causal framework to most attribution analyses, and therefore layering correlative decisions to yield a given answer leads to all-too-common fallacies in analysis. Instead of starting with the answer, a good attribution analysis starts with a clearly defined question. That question should be important regardless of the potential answers (note the plural – there is no singular answer to attribution).
Once the question has been clearly defined, it is important to consider all of the data points that can play a role in answering that question, and the assumptions for each of those data points. Making these two complete lists will ensure thorough consideration of the plausibility defining the framework for the attribution.
The last step is to take note of the information missing from the dataset you are utilizing (with every dataset, there is omitted data). This data can be the result of time, actions (including lack of actions), sources, technical limitations, and others. Identifying this list will help ensure that the model is cognizant of the data points that will not be a factor in the analysis.
Using this methodology to set up an attribution analysis does not guarantee productivity or success. It does dramatically decrease the likelihood that an attribution analysis will lead a good marketer to chase THE ANSWER that turns their marketing fortunes into gold – only to be greeted by the ignominious circular reference error.