- 16th April 2018
- Posted by: Manolis
In a matter of months, machine learning shifted from cutting-edge science to a tech-industry buzzword. Specifically in advertising and marketing circles, it’s gaining a reputation as a magic bullet; however, machine learning is simply a technique to use computational power to solve specified and difficult problems. Across the board, marketers have to focus on identifying their campaign goals and then find the right tool to reach them.
Machine learning can be a powerful tool for those capable of implementing it correctly. The reality is, very few marketers actually use machine learning, and for most situations it’s like using a bazooka to swat a fly—while it can solve an array of tasks, sometimes it’s overkill. Let’s explore the criteria needed for machine learning to be successful.
Identify whether machine learning is the right fit
Machine learning is an advanced capability that gives computers the ability to learn without being programmed. Put simply, computers can construct algorithms that learn from data. You present a specific problem to the machine when you know there is a clear answer, and you train the machine to solve that problem through repetition.
For example, last-click conversion, while not the most accurate attribution model, is the simplest. But in a multi-touch attribution model, deciding how much credit each of the dozens or hundreds of advertising events in a sequence should be given for a final conversion is a complex challenge. With thousands of conversions, this is not a problem to assign to a human. But this is perfect ground for unleashing the power of machine learning.
There are many practical applications for machine learning in advertising. Solving for click through rate, audience selection, and ROI, for example. In reality, advertising has no shortage of data or challenges to solve regardless if you sit near the publisher or the advertiser.
Confirm your project meets the right criteria
Similar to any data science or research work, the larger the sample size the stronger the insights gained. Machine learning requires very specific questions, massive amounts of data and time to solve a given problem. If you don’t have these assets readily available, you’ll end up wasting a lot of time. You’ll be better off using other tools to solve the problem.
As humans, we have the ability to process all kinds of disparate and jumbled data sources at once. The human eye, for example, allows us to process three-dimensional moving images and understand our environment through the visible spectrum. How we interpret visual information so quickly is an incredibly complex process. Machine learning isn’t quite there yet.
Interestingly, a recent study from IDC supports the notion that marketing professionals struggle to identify appropriate use cases for it, or choose to use a more “old school” methodology, with only 14% of marketers actually implementing the technology.
Although machine learning can be used to solve designated problems, it requires a vast amount of resources and data scientists to manage the problem solving. While it is a powerful tool, there are cases where less complex simple programming rules can deliver the results you need. However, if your engineering team cannot simply code a rule—such as identifying gender by email—or the task is just too time intensive, that is when machine learning should be considered.
As impressive as machine learning is, ask yourself what problem you need to solve, does your project meet the needed criteria, and then focus on figuring out the right tool for the job.