Share, , Google Plus, Pinterest,

Print

Posted in:

How Machine Learning Already Drives Value For Marketers

For all of the hype around artificial intelligence, the current generation of machine learning technologies have primarily yielded business successes in two basic areas: classification and prediction.

Sure, DeepMind built a server cluster that can play Atari games. But chances are that getting a high score on Space Invaders isn’t the cornerstone of your business’ marketing strategy. And if it is, you better have a budget the size of Google’s to burn on it.

On the other hand, classification of customer segments and prediction of buyer behavior/demand are both fundamental tasks in marketing. Machine learning (also known as data mining) is already paying off for the marketers who are sophisticated enough to apply it to these tasks.

Essentially, machine learning is an approach to programming in which problems are solved by iteratively “fitting” algorithms to data sets over time (a process known as “training” machine learning algorithms).

In other words, in machine learning the model is the output rather than the input. This means that the technology can be applied to problems in which there are too many variables to build a usable model upfront.

For instance, one can predict a customer’s likelihood to buy a product based not only on historical purchasing behavior, but also geographical location, age, the weather conditions surrounding a store location etc.

Try coming up with an Excel formula that allows you to do that, and you’ll quickly understand the value of machine learning for marketers.

Let’s now look at how machine learning is already driving value for marketers, instead of simply guessing at the directions in which the technology will take us down the road.

Sentiment analysis – Sentiment analysis is the identification of human emotions in various kinds of data: text, images, video feeds etc.

Historically, the technology has been mainly applied to text, and many marketing departments are already planning and evaluating campaigns using data about customers’ feelings extracted from tweets, surveys, call center notes and other textual sources.

Advances in machine learning for image processing have now made it possible to classify emotions in facial expressions. The importance of video feeds and images in customer sentiment analysis will skyrocket over the next decade as the technology continues to evolve.

Machine learning is the foundation of sentiment analysis, because human emotions are too “fuzzy” to be neatly captured in mathematical formulae. While Isaac Newton could describe the basic workings of gravity with just a few variables, classifying whether someone is angry based on the detection of a flared nostril in a photo is a vastly more complex computational task.

Customer segmentation – You may be thinking, “It’s pretty easy to identify customer segments using pivot tables.” And it is. The question, however, is whether you’ve identified all of your customer segments using basic techniques like pivot tables and OLAP, or even the most important ones.

This is where machine learning comes into play. Using a machine learning-based approach that leverages algorithms like k-means, you can segment customers at different levels of granularity without knowing the clusters you want in advance.

Instead of assuming that you already know the most important traits of your customers to pivot on (age, geographic location, income level etc.), you simply input the number of segments you want, and the algorithms output the segments for you. It could turn out that education is a more important indicator of a customer’s likelihood to buy or churn than income level, but you would never have discovered this fact if you’d assumed you knew the relevant traits at the outset.

Thus one of the most important applications of machine learning in marketing is in segmenting e-commerce customers. In e-commerce, you’re frequently segmenting customers based on their behavior online. An example of such a segment is first-time customers who add an item to a shopping cart, abandon the shopping cart and then come back within a month to purchase the item.

This may be one of the most valuable groups of customers to target in your marketing efforts. But because the variables involved are so different from how we normally classify people, machine learning approaches can work better in segmenting e-commerce customers than human insight.

In fact, some CRM providers have gone on an acquisition spree to buy up machine learning startups in order to enhance the analytical capabilities of their products.

Salesforce, for instance, has acquired a machine learning-based platform called BeyondCore that can perform sophisticated, multidimensional analyses of patterns in customer data and other datasets with minimal human input. The product even leverages machine learning to transform its findings into sentences that can be easily understood by human beings (e.g., “European sales down in Q1 because of supply chain delays in Taiwan”).

Real-time personalization engines – Amazon’s product recommendation engine is one of the most infamous examples of how to use machine learning to drive revenue. You’re probably familiar with it: the engine leverages machine learning to recommend a product “B” based on your purchase of or interest in product “A.”

Such engines are now evolving far beyond simple product recommendations in order to deliver a digital experience tailored to a customer’s unique needs and characteristics.

The evolution of personalization engines has been driven by the machine learning-based approaches to customer segmentation that we explored above, as Gartner analysts Penny Gillespie, Jason Daigler and Magnus Revang explain in their “Market Guide for Digital Personalization Engines”: “Personalization engines also detect behavior patterns, identify customer locations and discover correlations in behavior among customers. As these similarities start to be recognized, recipients can be segmented.”

The analysts continue that personalization engines are currently restricted to basic use cases (search recommendations, product recommendations, personalized landing pages and dynamic pricing). However, as customer segmentation via machine learning becomes both easier and more precise, use cases will only expand.

If you’re interested in applying machine learning to problems in your own marketing department, you can start by examining the many free and open-source data science platforms out there (RapidMiner, KNIME etc.). While these platforms aren’t always user-friendly for novice analysts, they offer lots of built-in modules and algorithms for tasks like customer segmentation and sentiment analysis.

You should also explore native machine learning capabilities that are appearing in CRM and BI systems. While products like BeyondCore are still a long way from replacing the role of the human analyst, machine learning technology will inevitably automate the task of pattern recognition in business data in the decades to come.

Early indications point to this shift happening more rapidly in marketing and CRM than most other business disciplines.

By contrast, helming an in-house initiative to develop machine learning-based approaches to marketing problems will be prohibitively complex and expensive for the vast majority of businesses, at least until the technology has evolved further.