Marketing attribution is a topic that remains a thorn in the side of most organizations. Are we still no better off than John Wanamaker back in the 1800’s when he purportedly said those famous words: “Half the money I spend on advertising is wasted; the trouble is I don’t know which half”? The trouble is that the goal posts keep moving, and just when you think you might be able to get your arms around the subject, the landscape changes.
Marketing attribution is both a business problem and an analytical problem. The business problem is simple: “How do I best spend my budget?” The analytical problem is a bit more complex: “How do I develop a methodology that delivers some valuable insight to solve the business problem with the data, time, and budget available?” Historically the answer to that question has been to use whatever methodology the technology we are using can give us, and that’s generally been the “last-click” attribution model. In a recent survey conducted by Econsultancy in association with Adobe, only around a quarter of client-side respondents claimed to be doing any form of marketing attribution analysis that wasn’t “last-click.” So lots of companies potentially need to solve this analytical problem.
To address analytical problems I’m quite a big fan of the CRISP-DM process. Although originally designed specifically for data-mining projects, I think it encourages good discipline and business process around any analytical problem. I thought it would be interesting to apply it to the marketing attribution problem.
The process is made up of six stages:
- Business understanding
- Data understanding
- Data preparation
Here we’re trying to determine what the business problem is that we’re trying to solve for. In this case it’s generally a marketing mix optimization problem. But what are the boundaries and what are the constraints? Are you only talking about digital channels here? If so, which ones? What about non-digital media such as TV, press, radio, etc.? Do they need to be included in the mix as well?
Dependent of the definition of the business problem, the next stage is to understand the data. Where is it and what does it look like? How good is the quality? Typically most companies will have web analytics and campaign-tracking systems at the core but there may be other data available that will need to be integrated into the approach in some way. Based on what you find here it may be that you need to revisit the business problem definition and adjust it accordingly.
This is generally the “heavy lifting” element of any project. Data need to be prepared for analysis and modeling and quite often you will find that there is additional processing and data management that may be required. You may need to extract data from different systems and load them into another type of database. You may need to integrate different data sources together to get a comprehensive view of all the potential marketing touch points. You may need to create new data collection methods to capture data that currently doesn’t exist. Never underestimate the time and effort involved in this phase!
Once that heavy lifting has been done, it’s time to start modeling. Typically in the CRISP-DM view of the world this would be some kind of statistical modeling, but in our marketing attribution view of the world the models are generally not statistical in nature but generally different views of the data – e.g., first click, last click, linear attribution, and so on. It’s likely that as the business problem becomes larger and more complex, more sophisticated analytical approaches are going to be required. Integrating offline marketing channels into the mix might mean, for example, the need to use inferential modeling techniques such as regression analysis or econometrics. These techniques are generally not cheap, but if the marketing investments are large enough, then they should be able to generate some real ROI.
Part of the modeling process is evaluating the results. This is evaluating the model on two levels. Is it a good model from an analytical or statistical perspective and is it a good model for the business – i.e., is it fit for purpose? Here you have to go back and look at the problem that you’re trying to solve and evaluate which model is the right one for you. Something that is too simplistic will provide you with a sub-optimal solution to your problem. Something that is over-engineered will not give you the kind of ROI that you’re looking for. There’s no point in having some complex econometric model if people only generally visit your website once before they decide to buy or not.
There’s no point in doing any of this stuff if the results aren’t deployed back into the organization! It’s something that needs to be considered upfront in the business understanding phase. How is the model going to be used? How will decisions be made? How often are you going to review the methodology? Markets move, things change, so the models you use need to be regularly reviewed.
I’ve come to the conclusion over the years that there is no “right answer” to the marketing attribution question. Every organization is different – priorities are different, strategies are different, and so the business problem (while generally consistent) is often articulated and defined differently. Every marketing attribution model is likely to be a customized solution in some shape, form, or other. Hopefully this approach can give you a framework for thinking about the problem.
Businessman With Business image on home page via Shutterstock.
Content marketing includes so many tasks and processes that it can be very overwhelming. You need a solid toolset to keep it ... read more
Can 360 video advertisements drive more engagements than regular video ads? Google partnered with Columbia Sportswear to find out and discovered it's not always the case.
With a little help from some algorithms and machine learning, marketers may soon be able to tap into the power of Hadoop, ... read more
If you don’t have a boss that expects you to deliver results reports on your programs today, you will in the future. ... read more