In my last post I wrote about the 5 key differences between targeting and personalization. Several comments were made about the distrust of automated systems. While I understand the concerns, I think it is important to know that automated personalization systems using machine learning are still under human control…at least for now.
First of all, I find it helpful to think of a personalization system as a combination of algorithms. An algorithm, by definition is a process or set of rules to be followed. Some of these algorithms are created and managed by machines while others are created and managed by people. In a personalization system, it’s the combination of human and machine algorithms that deliver results a customer will see that is appropriate for a business or brand.
For example, think of a personalized promotion on a website. If there are a large number of possible promotions to choose from, an automated system should be tasked with delivering the right set of promotions to each user. To accomplish this, people have to set goals for the machine. Are we interested in having customers click on the promotion, engage with site content or make a purchase? Once we pick an objective, the machine knows what to look for and the machine is learning and optimizing based on our human goals.
Next, people need to put some guardrails on the personalization system to make sure the machine keeps things in line with the brand or business objectives. These guardrails might tell the machine things like, “Only show this loyalty promotion to customers with loyalty points.” This is common sense to humans, but the machine needs to be provided with this information. Yes, the machine might figure out that customers who don’t have loyalty points don’t take the offer. But why even go there if you don’t have to. So, again people are in control of the machine learning system.
Finally, there are the numerous data points we have available. While machines can learn from all kinds of data, there are some data sources that are going to be easier to leverage than others. For example, if we already have marketing or audience segments, why not provide those data points to the machine. The machine can then learn which promotions are preferred by a particular segment. If we have location information, we can provide that data as well. The machine can learn if location impacts which promotions customers respond to and combine that with segments. This way the data we provide has an impact on the machine and how it learns. When we provide the machine with human-readable data, there is also the added benefit that outputs are more likely to be understandable to us mere mortals.
While automated personalization may seem scary, it is quite the opposite. The machines that we put to work are helping us to meet our goals, living within our guidelines and learning from the data that we provide. The end result for customers is a more personalized and relevant experience with higher engagement, increased revenue and loyalty. For online marketers and merchandisers, having machines help out means they can devote more time to strategic and creative tasks that still require their very unique and powerful human algorithms.