Netflix Has Adopted Machine Learning to Personalize Its Marketing Game at Scale
Imagine having a conversation with your financial advisor. He can tell by your tone of voice or facial expression if something makes sense or if you disagree with something, and he can adjust accordingly. But wouldn’t it be odd if your advisor started making suggestions that were completely irrelevant to your financial situation? It would also be troublesome if he offered suggestions that you’d already discussed in a previous conversation or, worse, had already acted on.
You would be beyond frustrated to have such an experience when speaking face-to-face with an advisor -- or anyone trying to help you -- yet for years, we have accepted such lack of personalization as par for the course online.
How many times have you encountered such an irrelevant experience? I can think of many examples. Business-to-business (B2B) technology sites have suggested that I download a whitepaper I just downloaded. Retail sites have sent me emails recommending women’s shoes when I have never shown any interest in women’s apparel. I’ve seen retargeted ads across the internet that show me products I viewed at some point but quickly decided weren’t right for me. Once you start noticing these unpersonalized experiences, you spot them everywhere.
In the last couple years, though, a few forward-thinking companies have changed our digital expectations. The mass adoption of services like Netflix and Spotify have created a new expectation in the mind of the consumer -- one-to-one personalization. The technology that Netflix and Spotify are using to create truly personalized experiences is a type of artificial intelligence (AI) called machine learning.
The term “machine learning” sounds highly technical -- and it certainly can be -- but its applications can be highly valuable to marketers. While you don’t need to know all of its technical intricacies, it does help to have a general understanding of the concept. To explain what machine learning is about as it relates to one-to-one personalization, let’s start with an approach that preceded it.
Rule-based experiences work in marketing -- but on a very limited basis.
It might help to more firmly establish how machine learning differs from other marketing strategies first.
The way that most marketers have delivered personalized experiences in the past is through rules and segmentation. Segments are created manually when a marketer decides to group customers, site visitors or app users by some predetermined criteria.
For example, a B2B marketer can segment site visitors by industry. Then, the marketer can manually set up a rule to display some kind of experience to those different segments. The marketer could select a relevant piece of content such as a whitepaper or a case study to promote to each person depending on which industry segment he or she falls into. This creates a personalized experience, because a person who falls into the financial services segment would see a different piece of content than someone who falls into the healthcare segment.
This approach can work, but it’s very limited.
The problem is that the rules are written by humans based on what they believe to be true. In the previous industry example, the marketer must decide which piece of content to promote to each segment. But each person is unique. Even within an industry, visitors could be at different points in their journey, or they could have different interests or content preferences. Their intent from visit to visit can also change. A few segments and rules cannot take all of this information into account.
There’s simply too much data for a human to sort through without the help of machine learning.
Machine-learning algorithms create unique experiences for individuals.
Rather than rely on a marketer’s manual effort to create different experiences for groupings of people, machine-learning algorithms provide a scalable way for marketers to create unique experiences for individuals.
With machine learning, instead of giving the computer lots of rules to follow, we’re programming it to learn everything it can about a person and select the experience most likely to appeal to that person. And for machine-learning personalization to be most effective, marketers should be able to build their own “recipes” that tell the computer what types of information to consider when determining someone’s digital experience.
A customizable recipe begins with the selection of one or more pre-programmed base algorithms. These algorithms can be simple, such as displaying items that are trending or recently published, or they can be more advanced, like collaborative filtering or decision trees.
Then marketers can put their own spin on those algorithms by telling the machine to include or exclude certain variables or to boost individual preferences -- depending on the needs of their businesses. With this approach, marketers maintain control of the individualized experiences their visitors receive.
Human vs. machine decision-making.
When you’re speaking with someone in person, you decide what to say next and recognize when to stop talking based on what you know about this person from previous encounters. Even if this is the first time you’ve spoken to this person, you have general behavior guidelines from similar social circumstances to work with.
Machine learning does the same thing. This learning is based on recognizing and remembering situations to form a fluid pattern that governs behavior.
Machine learning can determine what messaging you should display, the offers you should showcase, the next-best action you should suggest, the navigation options you should provide, the search results you should show, the timing and contents of an email you should send, or the most relevant products or articles you should recommend -- all based on what you’ve learned from the current and previous encounters with the person.
Machine learning uses real data to make these decisions, in the moment, just as a person would use it to make decisions in a conversation.
Ultimately, humans should have the last say in machine-learning marketing.
All that said, as beneficial as machine learning can be, humans shouldn’t hand over total control. A human’s ability to define, test and refine algorithms and points of interactions is vitally important to the overall objectives you have for one-to-one personalization and the brand experience you want to deliver.
Machine learning has some obvious advantages over traditional rule-based approaches. Most prominently, you can take into account far more data, uncover critical insights, and allow for individualized personalization.
You don’t have to understand all the technical nuances of machine learning. But understanding the basics of machine learning and its benefits will undoubtedly make you a better marketer.