Kitmondo process and packaging machinery lined up in a row.
PHOTO: Kitmondo PPM

Unless you’ve been living in an area with no Wi-Fi or mobile phones, you’ve likely heard about the promise of artificial intelligence (AI) improving marketing outcomes. Naturally, with hype comes reality, but AI has no shortage of challenges. 

Forrester reports that 45% of decision-makers say trusting the AI system is either challenging or very challenging. And perhaps an even more disturbing fact uncovered, according to a report by MMC Ventures, is that about 40% of Europe’s 2,830 AI start-ups do not use any AI programs in their products.

Why Machine Learning Matters to Marketing

What’s really happening with AI for marketing? If your marketing team decides to explore weaving intelligence into marketing tasks, what will you be working with, what terms should you know and how will you actually apply intelligence to marketing? 

For starters, let's get this straight. Don't think about AI if you're marketer. Well, not exactly. Think machine learning, a subset of AI, according to Jim Sterne, founder of the Marketing Analytics Summit and board chair emeritus for the Digital Analytics Association. “Artificial intelligence is the buzzword. Machine learning is a real thing,” said Sterne, author of "Artificial Intelligence for Marketing Practical Applications." He added, “Machine learning is the next level of software.”

AI is an "umbrella term" that covers things like robotics, self-driving cars, voice recognition, natural language processing and computer vision. Within that comes machine learning, “and that’s the stuff marketers should know,” Sterne said. “The other stuff is about communicating to customers with chatbots or robots or doing phone calls and natural language processing.”

Machine learning, however, is where marketers can apply actions like identifying a target audience, segmenting potential customers in order to send different messages to them and engaging people. 

Related Article: How Machine Learning Is Upending Marketing

What Is Machine Learning?

Now let’s dial it back in machine learning. What actually is machine learning? According to an MMC Ventures report, "The State of AI 2019: Divergence," machine learning is “modern AI” that allows software to learn through training itself instead of following a predetermined set of rules. Marketers are all too familiar with rules-based marketing decisions. “By processing training data, machine learning systems provide results that improve with experience,” according to the report. 

Machine-learning algorithms also use statistics to "find patterns in massive amounts of data," according to a report by the MIT Technology review. Machine learning consumes data and behaviors and ultimately makes an educated guess on the next-best action or appropriate messaging.

Machine Learning Marketing Actions

Machine learning when it applies to marketing has three actions: detect, decide and revise, according to Sterne. “Machine Learning is the ability for the system to look at a dataset and discover the most predictive attributes for a given outcome (detect),” Sterne wrote in a Medium post. “It can then infer rules about the data — from the data — weigh the attributes, and suggest a course of action to best achieve that given outcome (decide). Finally, the machine can look at the results of that action and alter its opinion about the attributes and their weights (revise).” 

Machine learning builds models of engagement through data. “As the data comes in, it changes its mind,” Sterne said. “And this is where the learning comes in.” How? A written piece of code helps the machine look at the data, and the machine creates its own structure and rules about the data.

Example: a marketer asks the machine to send out emails that will improve the open rate. The machine has thousands of subject lines to choose from, and it knows the particulars about the email recipients. It then learns that for this micro-segment, that message is better. And for this other segment, this other message is better, and it learns.

Related Article: How Machine Learning Is Transforming the Way Marketers Engage With Customers 

Knowing When to Use Machine Learning in Marketing

Machine learning at the end of the day is just another tool in the marketing technology (martech) suite, Sterne said. As with any tool, discretion is needed to determine when it's best to use it. “The marketer’s responsibility is to know when the tool is useful, when it's overkill, when it's just too expensive and where it can solve a problem that has not been solved before," Sterne added.

You will need data to start. “We need more than big data,” Sterne said. “These machines need a huge amount of data in order to come up with anything useful. So there has to be a lot of data, and it has to be a high-transaction environment. Think advertising online display ads. It’s a perfect environment, where there is lots of data, a huge number of transactions every second, and it’s very low risk. If the machine puts the wrong display ad in front of you, there's really no downside, except you spent hundreds of a penny.”

An email list of 5,000? Not a time for machine learning. However, an email list of 5 million people can be a good time for machine learning implementations. 

Marketing Automation Parallels

To put this all in marketing terms, machine learning can be seen as a continuation of marketing automation, according to Kelly Jo Sands, chief CRM and martech officer for Ansira. “Intelligent machines are really enabling marketers to make human decisions at scale,” Sands said. “It’s mimicking human behavior. Marketers have a lot of data at their fingertips that they can analyze and make a decision, but that becomes increasingly complex. It’s not unlike a parallel to marketing automation where marketing automation was brought to bear the market to really automate manual and intensive processes for the marketer. And machine learning is almost a continuation of that.”

Only with machine learning, algorithms are given data and asked to process it without predetermined roles, Sands added. “Oftentimes as marketers, we are speaking from kind of a gut validation, but we can lead ourselves down the wrong path, vs. letting techniques and advanced systems, like AI and machine learning, help guide us to maybe an answer based on the data that we didn't know.”

Related Article: Establishing AI Ethics in Marketing

Coding, Mathematical Models, Statistical Analysis

Marketers planning to make investments in machine learning will also almost certainly come across the need to apply coding, mathematical models and statistical analysis. For coding, expect to start with programs like Cobol, Fortran, C++ and Javascript. “You write code, and the machine executes,” Sterne said. Check out this list of programming languages for AI

Mathematical models are straightforward, according to Sterne. You apply formulas and values and play the “what if game to your heart's content,” he added. A mathematical model describes how something in the world works in mathematical terms; a quantitative way to describe systems, according to Lean Systems.

With statistical analysis, this is where PhDs come into play. “You want people who are doing predictive work, building complex models, and running them to evaluate their efficacy,” Sterne said. “And if they're good, then they're useful. It does take humans to do iterations to get a model that's good enough. And when it is good enough, it can have a big impact and be very valuable.”

Sterne recognizes that most organizations can’t afford data scientists like, say a Google, Amazon or Facebook. The latter has its own data scientist research department, after all. That’s why he comes down heavy on the "buy side" when it comes to building or buying for machine learning in marketing. 

Getting Familiar With Intelligent Terms

You’re not done with the terms you’ll need to know if you’re a marketer investing in machine learning. Here are some important terms to know and what you’ll likely run into investing in a machine learning marketing program.

Decision Trees

Decision trees are learning algorithms generated from the training data to solve classification and regression problems, according to Jinde Shubham’s report in "Becoming Human: Artificial Intelligence Magazine." It considers several factors that lead humans to making decisions. What makes decision trees special in the realm of machine learning models is their "clarity of information representation." 

Random Forest

Random forest is what gets built after a decision tree. “Random forest is a flexible, easy to use machine learning algorithm that produces, even without hyper-parameter tuning, a great result most of the time. It is also one of the most used algorithms, because its simplicity and the fact that it can be used for both classification and regression tasks,” according to a report in Toward Data Science. “Random forest is a supervised learning algorithm.” It is an “ensemble of decision trees."

Support Vector Machines

“Support Vector Machine” (SVM) is a supervised machine learning algorithm that can also be used, like with decision trees, for both classification or regression challenges. Mostly, it is used in classification problems, according to Analytics Vidhya. 

Deep Learning and Neural Networks

Deep learning comes out of neural networks, a set of algorithms modeled loosely after the human brain, that are designed to recognize patterns. “They interpret sensory data through a kind of machine perception, labeling or clustering raw input. The patterns they recognize are numerical, contained in vectors, into which all real-world data, be it images, sound, text or time series, must be translated,” according to Skymind officials

Machine Learning Powers Specific Marketing Tasks

Beyond the terms they need to know, marketers may be excited for the prospects of AI enhancing their marketing program. They may expect to apply machine learning to marketing to boost results across their entire marketing program. 

However, machine learning applications in marketing are designed to accomplish truly specific tasks. Not to be an ace throughout the entire marketing stack. “It's not a platform at all," Sterne said. "I can set up the machine learning system to build a model to improve email open rates. But then I have to build a completely separate model to improve conversions from that email, because that's a different problem to solve. These are very specific tools to do specific things so you're going to be starting at something specific.”

Related Article: Why Marketers Have AI on the Brain

Where to Start: Look at What You’ve Got

From an "AI in Marketing 101" perspective, discover whether you’re already leveraging it with existing tools, according to Sands. There may be features or functions that may have been added and upgrades that flew under your radar, she added. 

“That's the best place to start,” Sands said. “And then really kind of peel back the onion and better understand use cases for your role within a marketing team. People often jump toward the, 'How would I do this? Do I need to hire someone? Do I need to buy new tech?' And I think that just kind of continues to lead to more confusion.”