Machine Learning Reshapes the Marketing Landscape

Article Featured Image

Marketing has been evolving from a discipline based on gut instincts to one relying on empirical data. In fact, marketing today is almost completely driven by data, and companies are collecting oceans of it, literally. By 2020, the world will have accumulated 44 zettabytes of information, up from 10 zettabytes in 2015, according to market research firm International Data Corp. (IDC). To put those numbers into context, one zettabyte of apples would fill the Pacific Ocean.

But collecting data is only the first step in the process. By itself, data does nothing. Its value only comes once corporations use it to garner insights that can then be used to improve the business. Given the sheer volumes of data available, that is no easy task. Using information to gain business insights is labor-intensive, requiring marketers to sift through reams of reports, correlate items in an ad hoc manner, make business decisions, and monitor their results.

Luckily, machine learning has finally reached a level of sophistication where it can help.

“Machine learning has become mainstream in marketing,” says Colin Priest, director of product marketing at DataRobot. “This is mainly due to increased competition, which forces innovative marketing teams to adopt new technologies that give them insights from data coming in from their multichannel touch points.”

Machine learning, the next rung on the artificial intelligence ladder (see sidebar), turns information into action. It relies on algorithms to illustrate data point connections and generate reports. Since machine learning is a horizontal technology, marketers can apply it to any part of a customer engagement.

Companies can, for example, use the technology to measure marketing campaigns more granularly than in the past. With traditional newspapers, radio, and television, businesses blasted out advertisements and had no idea how they impacted customers. With search and social media enhanced by machine learning, they can now monitor where users click and gain more insight into the sales process.

Advertising is certainly one area where machine learning is widely used. “The most mature of the machine learning tools in marketing is programmatic advertising,” says Andrew Frank, a vice president and distinguished analyst at Gartner. “Firms have been using sophisticated software for years now and know how to bid on real-time opportunities based on whether or not users are clicking on their ads.”

Social media also generates oodles of information. Each day, more than 3.5 billion Snapchat videos are generated, and machine learning can also help marketers filter and use that content more effectively. “Machine learning tools enable companies to capture social sentiment and identify which customers are brand champions,” says Gerry Murray, research director at IDC.

Customer service is another area of interest for machine learning, with 57 percent of executives believing its most significant benefit will be improving the customer experience, according to Forrester Research.

Rather than simply responding to random inquiries, companies can build models that anticipate what the consumer wants and respond proactively. For instance, if a customer picks up the phone after clicking through a website and ending up with an empty shopping cart, the representative who answers the call could deduce the customer was unable to find the specific product she wanted.

In addition, companies are relying on machine learning for segmentation analysis. They are trying to identify the high-value customers and determine how to serve them better, while at the same time trying to identify the low-value customers and limit the resources they consume.


Personalization is one more area of emphasis. Marketers’ ultimate goal is real-time personalized advertising that spans across all platforms and delivers optimized messaging to each customer.

One way to achieve that is with recommendation engines, another place in the marketing stack where machine learning has really taken root. Once customers purchase a product, companies can present them with similar items that might interest them as well.

Companies like Amazon and Netflix have invested a lot of time and money into this sort of personalization and had a lot of success with it They are, of course, not alone.

Invaluable has evolved into an e-commerce site and online marketplace for fine art, antiques, and collectibles. The company enables auction houses, galleries, and dealers to provide clients in 160 countries with desired items. The firm, which has 125 employees, generates about $250 million in revenue annually.

Like many e-commerce sites, Invaluable was searching for a way to improve communications with its customers, who contact it across a variety of channels. Near the end of 2015, the company heard about Evergage’s personalization platform, which combines behavioral analytics, a data platform, and machine learning to help businesses better understand what drives customer interactions. Evergage’s cloud platform tracks each visitor’s shopping behavior on websites and mobile applications and presents them with relevant offers, messages, and incentives designed to improve demand generation and optimize conversion rates.

With it, Invaluable has gained better visibility into its customers’ desires and can present consistent messaging across all of its sales channels, according to Neal Glazier, vice president of marketing at Invaluable.

Sentiment analysis is another emerging area for machine learning. Using this technology, companies try to identify the emotions that their customers are feeling so they can respond in a more empathetic manner. Mattersight is just one vendor that provides sentiment analysis. Its platform builds up personality profiles for each customer over time and can route callers to the appropriate agents who are trained to engage them in their preferred communication styles. In Australia, Touchpoint Group developed Ipiphany, a customer service solution designed to help banks identify brand behaviors that irritate customers so that they can train their reps not to engage in them.

Customer churn is another area where machine learning is being applied. Churn is very common among telecommunications providers, where companies are constantly updating their devices, form factors, services, and pricing plans. Here, statistical analysis enhanced with machine learning is helping companies identify which customers are likely to switch and then deliver enticements to convince them to stay.


While marketers have increased their interest in machine learning, they face a number of serious challenges in adopting the technology. Among them, organizational issues are among the most pervasive. For one, many companies still cling to department silos that are not built to take advantage of machine learning. Others do a poor job of aligning their investments in machine learning with business opportunities, according to Capgemini, which found that the typical deployment practice is to place the tools in the hands of technologists who prioritize very challenging projects and miss the low-hanging fruit. As a result, these companies overreach on their initial machine learning deployments and have a hard time building sound business cases for further investments.

An equally large problem is that few companies have the proper skills to deploy these solutions. To put a system in place, companies need to pick their machine learning platforms and then customize them to match their needs. That’s not easy given the large number of vendors, the enormous variety in system capabilities, and the differences stemming from the design of the intelligence engines and their algorithms. Each machine learning solution has been tailored to its own set of special use cases for various industries.

And then, even when the right solutions are selected, few companies have staff adequately trained to use the tools. “The biggest challenge is that marketing teams usually don’t have any data scientists on their staff,” says DataRobot’s Priest.

This shortcoming creates a significant barrier because these solutions are quite complex to install and run. Deployment means building custom data models, something only individuals with a data science background typically do. These individuals are rare, the competition to hire them is intense, and their salary requirements are high: While salaries for machine learning specialists and data scientists start on the very low end at around $100,000 per year, they often surpass the $250,000 mark.

Not surprisingly, then, costs for machine learning can add up quickly. Building a machine learning marketing application is a complex, expensive proposition. While vendors have tried to build tool sets that can be plugged into different applications, each company usually has its own unique mix of front-end and back-end systems that need to be integrated. “Initially, we underestimated the work needed to get our system running,” admits Invaluable’s Glazer.

The integration work takes time and costs money. “One good predictive model is worth millions of dollars,” says Mike Gualtieri, a vice president and principal analyst at Forrester Research covering artificial intelligence and advanced analytics.

After trying it on their own unsuccessfully, companies often find the integration work so daunting that they wind up outsourcing development to consulting specialists like Accenture, Capgemini, and Infosys.

Another option for companies is to find a marketing solution that has the type of machine learning that they need built in right from the start. Some vendors, like IBM with its Watson cognitive computing engine and Salesforce.com with Einstein, trumpet their machine learning capabilities. “Other marketing solutions may not tout their machine learning capabilities, but they are bundled in these solutions,” Forrester’s Gualtieri says.

With these prebuilt solutions, companies only need to tune the system and analytical reports without having to build their own custom engines from scratch.


Companies also need to keep the solutions and their expectations around them in perspective. Though it’s called artificial intelligence, the technology is by no means omniscient. Systems can transform gobs of data into tidy reports, but ultimately, humans are still needed to interpret the information. The evaluation process involves a high degree of subjectivity. “A lot of factors are involved in a purchase,” Gartner’s Frank says. “How much credit does a company give to a marketing campaign, or does a loyal customer just decide that they need the product at that moment? The difficulty is accessing the value that the machine learning provides.”

And no matter how far the technology advances, machines are not as smart as humans. “Computers do not understand really basic things because they lack world knowledge and common sense,” Frank adds.

Mostly due to overambition, not all machine learning projects have been successful. Chatbots, for example, “are not very intelligent” right now, Frank points out. “They understand language in terms of words said but are not good at interpreting the context of a conversation.”

These tools can perform simple tasks, like connecting phrases to a reference document in FAQ files, for example, but they are not yet able to complete complex correlations.

Because of these shortcomings, some executives might be hesitant to trust machine learning results. “Marketers often feel comfortable making decisions based on gut instincts,” says Goutham Belliappa, a principal data integration and reporting practice leader at Capgemini. Marketing executives, he notes, have a lot of knowledge about the subject and often prefer to base their decisions on that knowledge.

This is often, then, permeated throughout the entire corporate culture, which could result in the whole marketing team’s reluctance to adopt machine learning.

Changing corporate culture can be challenging, so in some cases, corporations are splitting the chief marketing officer’s role in two: one who follows old-school principles, and another who’s steeped in more modern technologies. In other cases, firms are taking their chief marketing officer from within the IT department because they understand how the technology works.

So while marketing has been looking at machine learning, many companies are struggling to incorporate it into their marketing processes. Many large companies are dabbling with the technology, but they are confining it to narrow areas and trying to get a better idea about its potential.

Fortunately, businesses have time to make needed changes. While there has been a lot of hype about machine learning, the market is in a nascent stage of development. More time will be needed for organizations to put all of the pieces in place to fully take advantage of the technology.

At the same time, though, there is no doubt that machine learning will eventually reshape marketing. “Machine learning frees up marketers’ thinking,” IDC’s Murray concludes. “They have more data on more people than at any other time, and the power of industrial analytics is growing. Increasingly, businesses will compete on their idea of what drives customer engagements: How can we create value-added services on top of what we try to sell? That is where the new customer loyalty will come from.”

Paul Korzeniowski is a freelance writer who specializes in technology issues. He has been covering CRM issues for more than two decades, is based in Sudbury, Mass., and can be reached at paulkorzen@aol.com or on Twitter at #PaulKorzeniowski.

Making Sense of Machine Learning

At one time, individuals punched cards that computers then read as part of their programming. Through the years, computer software has taken on more of the work that humans once completed. As technology has evolved, terms arose that described new system capabilities around intelligence. Here are just a few of them:

Automation—the simplest form of computer intelligence. Here, a computer system completes one specific task. For instance, a network monitoring program automatically sends an alert to a technician’s phone when response times go over a predefined limit.

Orchestration—the next step in the process. Here, the system completes a number of tasks autonomously. Configuring a server is one area where orchestration has recently taken hold.

Artificial intelligence—a term that emerged in the 1950s to describe actions that computers could take that possess the same characteristics as human intelligence and some degree of reasoning. Through the years, various systems have been programmed to perform a range of activities, like image classification on Pinterest.

Machine learning—a field of study that gives computers the ability to learn without being explicitly programmed. This type of programming is often referred to as unsupervised because the technology identifies patterns, builds insights, and automatically acts on those insights. Here, input shifts from people teaching computers via code to teaching with examples.

Neural networks (also known as deep learning)—this is the next step. The approach was inspired by an understanding of the biology of the human brain. Here connections are more free-flowing than traditional linear programming. In the brain, any neuron can connect to any other neuron within a certain distance. With neural networks, programs have more connections and can draw more complex conclusions. Neural networks have taken on tasks such as colorization of black-and-white images and automatic handwriting generation.

While the various terms are helpful in understanding how these solutions work, they lack precision, and overlap is common. Also, new terms will emerge as the technology takes further shape, experts agree. —P.K.

CRM Covers
for qualified subscribers
Subscribe Now Current Issue Past Issues