Machine Learning in Marketing? 4 Suspicions Challenged

By: David Verhaag | January 24, 2017 | Artificial Intelligence

The machines are learning. First, it was which movies we want to watch on Netflix. Then it was which products we want to purchase on Amazon. These are now classic examples of machine learning applied to solving everyday consumer problems like “What do I want to watch?” and “Doh! It needs batteries?!”. The artificial intelligence-enabled world is all around us and quickly evolving. Beyond movie and shopping habits, the machines are also helping us drive our cars, control our home’s temperature and lights, and give us appliances that talk back, warn us about the weather, remind us to pay our bills, and help us mix drinks.

Why should you care? The machines are also learning the roles of the marketer. Artificial intelligence is powering marketing decisions around which message will resonate with our audience, which channel we should reach them on, and what time of day we should try to engage them. The machines are learning and the marketer’s role is evolving.

AI and the modern marketer

To embrace the evolution of artificial intelligence-enabled marketing, Chief Marketing Officers need to understand the power and potential without simply granting blind trust that The Machine is going to make the right call on critical campaigns. Understanding requires admitting that for many of us, myself included, data science, machine learning, and artificial intelligence can all sound a bit like sci-fi. Because it is difficult, it can be dismissed as the tech terminology of the moment, something everyone has and is talking about simply because everyone is supposed to have it and be talking about it. Understanding the role of artificial intelligence in marketing doesn’t require we mind meld with our favorite data scientists, but it does require challenging some of our suspicions.

Challenging our suspicions can be uncomfortable. It can be a bit embarrassing to admit we don’t know what everyone else purports to.

Embracing that discomfort, I will share a few of my own early suspicions about data science and the role of artificial intelligence in marketing. Challenging our suspicions can lead to understanding, to acceptance, and to adoption of the power of AI which will drive better business outcomes from your marketing initiatives. Read on…

1. “Big data” is really just a bunch of little numbers

Data science often involves small numbers. Really small numbers. It also involves complex concepts for numbers like confidence intervals, covariance, and linear regression. It can get confusing, especially if you, like me, are not a “math person.” Not only can it be confusing but also a bit suspicious. How can a 1-2% bias in your data affect your outcome by a factor of two? Uh…if I can’t understand the math, how can I trust it with revenue critical campaigns?

Understanding the math behind machine learning and the implications for marketing is not as important as understanding the fundamental questions that technology can answer for us. Most of the Chief Marketing Officers I speak with intuitively understand the right questions to ask, for example, how do I proactively engage new users without overwhelming them?, and how I do encourage active users to complete the virtuous actions that drive revenue? The right questions are those that focus on the business outcomes of the marketing campaigns and messaging strategy. It is simply a matter of breaking those questions down to the foundational elements that can in fact be answered at an incredible scale and pace by artificial intelligence. CMOs who embrace the opportunity to focus on the right questions will find that artificial intelligence is their best friend in helping them focus on strategic decision-making and artistic license and less on tactical decision-making like which time of day to push messages.

In 2016, for example, Kahuna customers who leveraged machine learning-driven optimization in their conversion campaigns realized a 1.4x increase in user engagement over campaigns run without. The benefit is accelerated even further when leveraging machine learning and Kahuna’s new Experiences product. Customers who leveraged artificial intelligence in their Kahuna Experiences saw an increase of 2.6x in user engagement compared to campaigns run without.

2. “Interesting, not practical”

One of the classic scenes from Mad Men doesn’t involve the data scientists we have today but “our man from research.” In the scene, Dr. Greta Guttman is sharing her research with Don Draper and Salvatore Romano to make the case that they need to evolve their marketing message based on the health hazards of smoking and changing consumer appetites. Her research is not exactly embraced by the marketers. Don literally throws it in the trash can.

It is easy to be suspicious of data that points us in a non-intuitive direction. Data scientists, like Dr. Guttman in Mad Men, are often the voice validating what the machines are telling us. Fortunately, today’s marketers are not quite as oblivious to the power of research and innovation. Kahuna customers are already using machine learning to optimize more than 50% of their messages. Tomorrow, the most agile will embrace path optimization which leverages machine learning to drive the consumer’s journey. The opportunity for marketers is to understand that artificial intelligence is constantly changing and constantly evolving and what might seem “interesting, not practical” today will be the foundational technology of tomorrow. Seasoned mainstream marketers partner with Kahuna to understand and leverage the ever-changing evolving world of machine learning.

3. “Data cleansing” just sounds suspicious

Break out the scrub brush—my data is dirty as hell. The term data cleansing has always set off alarm bells for me. So if the data doesn’t agree, you simply cleanse it? Maybe bring in big Vinny to do a little “cleaning” and rough up the data set to ensure it tells the story the CMO wants to hear? Data cleansing is not as suspicious as it sounds. It is the “process of detecting and correcting (or removing) corrupt or inaccurate records from a record set, table, or dataset.” It is a critical step in ensuring that as marketers we are listening to the right message from our data.

We all know that our marketing data sets may contain outliers, test data and bad data. That single user who is visiting your site at 3am daily to make a purchase can give you the impression that after hours shopping is an area to focus versus an individual’s random shopping habit. The campaign you ran last fall that went to the wrong audience but, because of election results, ended in 20% uplift can give you the impression that the audience engages on a particular channel at that time of day. That campaign you ran internally to resolve the push receiver issue doesn’t suggest that “test” is the ideal campaign message since it drew such great user engagement.

Sometimes the data needs to be reviewed and “cleansed” to remove inaccurate or misleading data points. That shouldn’t lead you to believe that “cleansing” is removing the facts that don’t tell the story you want to hear. Rather, cleansing data is an important step your artificial intelligence-enabled marketing platform will take to ensure that you don’t misread noise for signal—that we don’t manually misdirect our campaign strategy at corner cases that don’t regularly occur or adopt messaging that won’t resonate simply because of random test data results. Misreading noise for signal can derail your marketing plan. AI and the process of data cleansing can help prevent this.

4. His “unsupervised learning algorithm” needs supervision

Unsupervised learning is a type of machine learning algorithm we here at Kahuna use to draw inferences from data sets consisting of input data without a known outcome, e.g. which message variant is going to resonate with the selected audience. This is common A/B testing but performed by Kahuna’s machine learning algorithm to dynamically adjust the delivery to maximize engagement. As the algorithm learns which message, channel, or path is performing best, the overall performance of the campaign improves, the end result being far better campaign performance than a non-AI driven campaign strategy.

Supervised learning, on the other hand, relies on known outputs, e.g. we already know which channels or messages we should use to train the algorithm to reach the desired outcome. Supervised learning is helpful if you already know the answer but often in marketing, we don’t. Or, we simply assume we do. The power of machine learning, and the basic tenant of unsupervised learning, is that the algorithm will show us the optimal path, the right channel, and the right message.

The optimal path, channel, and message might be counter-intuitive. To fully embrace the evolution of artificial intelligence-enabled marketing, marketers and the CMOs they report to need to understand that the machines are learning—and at a faster and more accurate rate than the well-intentioned marketers on the team. Through the process of drawing inferences and testing results, the learning algorithms that Kahuna uses can quickly determine the optimal message, channel, and path. No guessing required.

Final thoughts

The machines are learning. To embrace the evolution of artificial intelligence-enabled marketing, Chief Marketing Officers need to understand the power and potential of machine learning in their marketing technology. Luckily for most of us, that understanding doesn’t require an advanced math degree. It does, however, require embracing the fundamentals of machine learning and the acceptance that those data scientists might be right after all.

Sloan

Author: David Verhaag

David Verhaag joined Kahuna in March 2015 as the VP of Customer Success. Prior to Kahuna, he was the VP of Customer Success at HireVue, where he drove market leading product adoption and best-in-class net promoter scores. David also spent 8+ years at SuccessFactors where he held a variety of leadership roles, including Director of Customer Value, where he led the growth of the global Customer Value function. David received his Bachelor’s degree in Business Administration and Human Resources Management from Eastern Washington University.

View all posts from this author