Self-driving cars, Facebook auto-tagging photos, Netflix recommendations, and targeted advertising—what do all of these have in common? These technologies have all undergone significant advancements in recent years due to an explosion of computing power and advancements in computer’s ability to learn, or “machine learning.”
While it sounds like a futuristic term, machine learning is the science of getting computers to act without being explicitly programmed. For example, let’s imagine a CRM program where data has been collected on customer’s interests, demographics, and engagement with previous campaigns. Based on previous interactions with customers, we can create predictions of how these customers will interact in future campaigns.
While the technology has existed for quite some time, significant advances in scale and computing power have allowed this technology to flourish. Companies including Amazon, Google, IBM, and Microsoft have all developed user-friendly machine-learning capabilities to complement their growing web service and cloud offerings. While some user interfaces are more intuitive than others, the goal is to allow users to upload data and allow the computer to extract valuable insights.
The marketing field is certainly taking notice. Marketers who have begun to use these technologies are asking questions such as, “What type of user will click on this ad?” or “How likely is this user to return to my site?” One popular use of the technology is to determine the probability that a user will respond to a direct mail or email. Based on previous information gathered and past user behavior, machine learning can identify who is most likely to engage in certain activities. Instead of blasting a direct mail out to 10,000 people blindly, we can really hone in on the users that we think are going to respond and customize a solution for them.
Another use is detection of click fraud in online advertising. Marketers certainly do not want to pay for 1,000 clicks when 980 of them are spam. While there can be numerous types of fraud, a computer can differentiate these types of spam and determine if a “real” person actually clicked on their ad. These technologies can realize significant savings for advertisers, and certainly distinguish advertising platforms and publishers.
Of course, there are still significant challenges to overcome. In the case of ad fraud detection, because click-through rates tend to be quite low, a significantly large amount of data is needed to accurately predict user action. Another issue is the growing complexity of these machine-learning models. As predictions tend to become more accurate, the complexity of how the computer arrives at an answer is increasingly unclear. Most recent machine learning algorithms have been labeled “black boxes,” as computers are performing millions of abstract calculations that are too vast for the user to analyze.
As machine learning solutions become user friendly and easy to implement, marketers should certainly start thinking of how they can apply machine learning to find new insights about their business.
CONTINUE THE CONVERSATION:
Questions? Comments? You can contact the author directly at firstname.lastname@example.org.
Please allow 24 hours for response.