The abundance of industry insiders using the terms AI and machine learning interchangeably is a chief cause of confusion about these two complex, interrelated topics. Generally, AI is a superset of algorithms, and machine learning is a subset of that.
Simply put, AI encompasses any system that can take an action based on rules, as well as any system that can learn, adapt, and take action based on that learning. One simplistic example in marketing is “if a site visitor downloads this whitepaper, then trigger an email inviting them to register for the upcoming webinar on the topic.”
AI has long meant rules-based systems. The original users and developers basically hard-coded this rules construct. But rules are fixed; there’s no learning. These systems are “intelligent” to a point, but rules-based systems are not adaptive. Humans have to adjust them.
Machine learning is an adaptation mechanism. Marketers look at data or patterns and try to learn from it. One of the main goals of machine learning for marketing is prediction – to try to figure out what customers and prospects are going to do based on what they’ve done before and the information you have about them. More specifically, it’s about trying to recognize patterns of what customers have done and are doing to predict what they will be doing.
What’s the Value of Machine Learning?
When enterprises use machine learning to predict something, they can focus on the most important variables, and cut out the “noise” of the information that’s not relevant. Remember, modeling is all about learning the underlying processes that generate the data and behaviors. Marketers can use that relevant insight to personalize campaign elements such as offers and pricing, or encourage current and prospective customers to take an action. They can even use the information to optimize inventory.
Another aspect of machine learning is optimization. There are myriad ways to do something and there may be multiple “best ways,” each with different positives and negatives. So, marketers can use machine learning to optimize on top of that insight. For example, machine learning techniques can optimize the channel mix based on what marketers are learning from their KPIs.
Beware: Anyone who always insists on only using one specific learning algorithm, or machine learning technique for optimization can negatively impact results – look elsewhere for your machine learning support. Marketers need to find what techniques work best for their data today, and know that their data can and will change over time. The best practice is to let the data speak for itself, and select the best model(s) for that data regardless of type. Linear regression, for example, may be easy to use and understand, but it is not necessarily the best approach to use all the time. In the real world, most of the underlying processes are non-linear, so applying linear techniques can lead to the right solution to the wrong problem. Marketers need to try multiple techniques to find the one that works best for the data at hand. This doesn’t have to be a manual, time-consuming task. It can be accomplished in an automated fashion, using optimization techniques to find the best algorithm, tune the knobs, and adjust the relevant parameters.
As with AI, however, too many marketers may believe that machine learning can recognize and predict things perfectly; that it’s a do-all tool for everything. Marketers need to manage expectations of what they’ll get out of machine learning. The available data may not be sufficient to create a model that can predict perfectly. It might only be possible to get 80 percent of the predictions correct, and that’s great if that’s all the information available within the data. Machine learning isn’t going to solve every problem. It’s just like predicting the weather.
What Are the Most Common Uses for Machine Learning for Marketers?
Marketers can use machine learning techniques to learn what’s inside the data, how items such as behaviors and channels affect each other, and recognize customer behavioral patterns.
Consequently, one common use of machine learning is segmentation – not just to learn who’s in what segment, but also why they’re different, and what the boundary areas are that put people in the different segments. Then marketers can use the resulting models to help figure out which actions, offers, or content are likely to move customers into one segment or another. With that kind of information, it’s possible to gain tremendous market share by improving the entire business construct in terms of retaining existing customers you have and obtaining new ones.
Remember, though, that what worked today may not work exactly the same tomorrow. Because people are adaptive learners too, there’s a constant dynamic shift. Fashions change, trends come and go – the world is a dynamic place. People move from one segment to another or may move from the center of one segment to a boundary. This dynamic is why real-time decisioning is so important. Creating predictive models can be time consuming and the data may be outdated by the time marketers use them. Ideally, marketers could use machine learning to automate the process. It’s like having an easy button. A marketer could say, “Hey, my data changed or my approach changed; I want to do this again,” then press that button and immediately, automatically, here’s a new set of predictions and a new set of models. Even better yet, encapsulate the learning process so that it automatically re-learns on a schedule or through some trigger event. That automation allows operations to run faster with fewer humans involved, so marketers can do their jobs more efficiently.
Machine learning is only as good as the strategy and data behind it. So, marketers need to decide what they want to achieve by using it; what they want to automate or improve. It’s best to start with a small project that will deliver a manageable response and trackable results.
Also, it’s important to minimize the data thrashing (i.e., moving data between systems). Anytime marketers have to go through multiple steps to get and pre-process (e.g., clean) the necessary data, they’re losing time, which can negatively impact getting a solution to market in a timely manner. They’ll also lose out with poor data quality. Marketers must ensure that their data is high quality, deduped, and meaningful enough to provide the insight they’re looking for.
Once marketers have that strategy in place and the data to support it, they’ll get the most from machine learning by using automation to personalize marketing campaigns and content. And, by making it easy for users to use it; for example, by providing scripts or macros.
Remember, managing expectations is huge. Don’t expect absolute perfection – the only perfect models with perfect predictions are created in statistics classes using perfect data. And, don’t expect machine learning to take over marketers’ jobs. Instead, it’s going to make their jobs easier and make analysts, data scientists, and marketers more efficient and effective—and more likely to positively impact the business.
Be in-the-know with all the latest customer engagement, data management, and Redpoint Global news by following us on LinkedIn, Twitter, and Facebook.