10 mins, 12 secs read

The Basics of Machine Learning: Marketers and Magic Numbers

2 years ago

If you still believe that your daily activities are the result of your own choices, you are mistaken. Few of us understand that our purchasing decisions, and increasingly our life decisions, are heavily influenced by advanced algorithms.

In this piece, you’ll learn about:

  • What machine learning is and why, along with data science, it has become so popular
  • How to apply lessons from machines to marketing and sales
  • What results machine learning can bring
  • What solutions are worth investing in

What is machine learning?

Machine learning lets us take relevant information and from a mass of data and identify the key behaviors that have significance for business, marketing, sales and other fields, including those associated with the quality of daily life and thus customer experience management. Machine learning is a field of research that involves artificial intelligence, drawing conclusions, making decisions and therefore understanding context through modern programming.

By organizing huge amounts of data about customers, you can gain information that helps to predict their future behavior and reinforce it (if it’s positive) or counter-act if needed (if, for example, users cease to be interested in an offer). This is not a vision of the future — such mechanisms can be used now, with good examples being text autocorrect, virtual assistants, marking paths in navigation, etc.

Machine learning also has an influence on the development of contemporary marketing, changing the responsibilities and skills demanded of marketers.

The possibilities of machine learning in sales and marketing

  • recommendations — the system of recommending products and services
  • personalization — adapting time and contact channels to users
  • prediction — analysis of the probability of certain events
  • segmentation — advanced segmentation of users
  • correlation — putting together different data sets

Intelligent systems

Reading an article with an enigmatic title on the subject of machine learning, you surely consider how much has changed in the way we absorb information. These days, subjects and ideas once confined to institutions of higher learning have become accessible to everyone and are part of our everyday lives. Ten-year-olds today know more about programming than the average thirty-year-old, who in turn spends their free time planning with a virtual secretary.

Furthermore, trust in new technologies is growing along with a belief that algorithms that know the tastes in music and film of others with similar interests will be able to alert us to upcoming performances and products we could be interested in. Customers are surrounded by such algorithms and their lives are made easier by them. How can we put their marketing and sales potential to work?

The meaning of machine learning in algorithm construction

The origins of artificial intelligence are learning systems that operate on large data sets, constantly analyzing them and looking for the occurrence of certain events. As the database grows and becomes more organized, the conclusions reached from analysis become more reliable and precise. Here’s an example to illustrate the verification of marketing activities with the use of algorithms from machine learning.

Let’s say you’re building an algorithm that defines the possibilities for the introduction of a product on a foreign market. The decision-making process can present a barrier to entry for certain businesses, depending on the field. After submitting more information for analysis like the size of the company, its turnover and position in the market you can predict the effects of entry into a foreign market based on data collected from other markets. The collection of information related to companies and the economic, political, legal and social conditions prevailing in a given market allows for the informed recommendations of which markets to give priority to and what to be aware of.

The more involved and extensive the decision-making process is, the more precise the analysis becomes. It’s easier to predict the probabilities of certain future events. The key to success does not lie in programming results, but in the proper and accurate preparation of data that allows certain theories to be tested.

Customer data can be collected in two ways:

  • Some is shared openly and can be gathered from completed surveys and contact forms, email databases, customer loyalty cards, online and offline transactional data as well as declarative data.
  • Other data can be gathered through the use of tracking codes that follow customer actions on websites and allow for the construction of behavioralschematics. If a new user journey fits into existing workflows of other similar customers, it is added automatically to a predefined segment. Tracking codes make this collection of data possible.

Reinforcement learning

Algorithm reinforcement learning is a procedure that operates on the basis of large numbers of attempts and mistakes. It’s very effective when it analyzes databases in incremental blasts. The work of learning algorithms is a bit like being “warm” or “cold” in a search for something. When you are near, you get a prize and when you’re far away you turn around and start looking somewhere else. Each step in a particular direction can bring you closer to a more accurate analysis and solution to a problem than a step away and the current status is constantly verified with new data. Warm and cold are your prizes or punishments. This is how you are gradually lead in the right direction — the closer you get, the “warmer” you are.

The goal of the system is to increase conversions and revenue. To do this, it analyzes hundreds of thousands of transactions in order to understand what factors influence purchases. The system obtains information that lets it predict which segments of users are more likely to buy, which need more nurturing and which are unlikely to buy. The system must analyze this data continuously since factors that influence purchases are constantly changing.

Using data to make informed predictions

The ability to estimate with a high degree of probability the percentage of customers that will buy a product in a given time period can be a new era for global e-commerce. How can you use machine learning algorithms in marketing and sales? Here are a few ways.

Recommendation systems

Consumers now expect to see recommended products based on their search history, featuring a variety of options with similar parameters like brand, color, etc. Recommendations are fully personalized through consideration of the individual’s browsing and purchase history. This creates opportunities to promote conversions through relevant suggestions.

Example: To improve the personalization of their offer, Netflix hired a team to assign films to hundreds of genres and categories. They used tags to divide the film database into more than seventy thousand micro-categories. This allowed them not only to correlate films watched by a user but also to deploy algorithms to analyze even the most seemingly insignificant user actions like pausing the film at a particular time, stopping watching altogether, etc. They also integrated information about what user friends liked and even ages to better personalize the user experience. Using these techniques, Netflix offers content adapted to the tastes and preferences of each customer. They emphasize that three-quarters of films watched on the service were suggested by their algorithms.

Personalized time and channel contact

Modern systems analyze customer behavior like which channels they use most often and at what times they typically answer incoming messages (email, SMS, etc.). On this basis, businesses can select the appropriate channels for reaching consumers and the best time to do it in order to maximize the chances of conversion. All of this is done automatically, without the need for direct participation from marketers.

Example: Let’s say someone signs up for a newsletter from an electronics retailer and they are sent once a week, on Tuesdays. Now imagine a particular subscriber always get around to opening those emails on Friday afternoons, when he has a break in his schedule and looks through unopened emails. This same customer doesn’t react to SMS’s or push notifications, he prefers to look at new offers on a desktop. The system recognizes these preferences and automatically starts sending messages on Friday afternoons and stops sending SMS. In doing so it adapts to the needs of the customer and minimizes the risk that sent emails get lost in a sea of other messages.

Predictive analysis

The system analyzes customer behavior and groups them into segments. If a new online user begins to match the behavioral schematic of a certain segment, he is added to it. This makes it possible not only to group customers according to the likelihood of conversion but also to take the appropriate actions regarding groups with lower probabilities.

Example: A user visits a page on a website and purchases a product. The next two visits end with viewing the main page only. Behavioral analysis of other visitors to the site indicates that most customers that visited three times don’t buy anything else and don’t return. What do you do? The system has a preventative measure for just such a situation. Just after the second visit, the customer will see a banner with information about a promotion or discount. It will encourage him to view more products and make additional purchases.

The normal procedure works on the assumption of hand-crafted marketing rules, but the use of machine learning eliminates the necessity of adjusting automation to the decisions made by the software. The means that every event in the system — buying a ticket, uploading a picture, each click on a newsletter — can be applied to making future predictions of probability.

Segmentation by details

Segmentation of customers according to the possession of certain characteristics is accomplished by the earlier use of tags that help to create groups based on, for example, place of residence, gender, age, purchase history, etc. Data is typically gained from external CRM systems. Based on information about where the customer was gained from, you can select the most effective communication channel and the best time and frequency of communications. All of these elements have an influence on purchasing decisions. It’s not only what you sent, but through what channels and how often.

Example: Communications can be adapted to behavioral and demographic data. Demographic features allow for the automatic adaptation of content to users. For example, an email can display different products for each gender, age group, location, etc. Behavioral data lets you run the appropriate kind of automation. For example, a user that visited a certain category on a website can get an email featuring a graphic used on the same page she visited.

Correlating different sets of data

Transactional data from different sources and can be analyzed for common characteristics and correlating features.

Example: Intelligent systems can, for example, compare transactional data with weather patterns and look for correlations to buying patterns in a certain field. Of course, transactional data can be combined with any other collection of information, you are only limited by your imagination.

Indispensable marketer-analyst

Algorithms are now essential marketing tools and have redefined the role and duties of marketers, who can now save time formerly spent on repetitive actions now done by technology. Now, the focus can be placed on analysis of data delivered by the system while adding insights from their own experience.

Tips for marketers in data analysis

  • Don’t focus on analyzing individual events, look for trends in bigger samples.
  • Pay attention to related products — customers who buy a computer are more likely to be interested in related and complementary products.
  • Analyze the technology surrounding your customers — screen sizes and operating systems often say something about disposable income
  • Analyze purchasing paths, check which paths are most common before those purchases, what customers look at in physical stores and their interactions with customer service.
  • Analyze data and engage with enrichment data that may correlate with other factors in the economic, political or social landscape.
  • Monitor data related to the days, times and devices involved when purchases are made — these can be clues to effective promotional ideas.

Wrapping up

If you still believe that you are not affected by Big Data, you are mistaken. If your business is doing well without using artificial intelligence, you will soon be confronted with the need to manage with huge amounts of data coming from every direction. If you want to accept this challenge, start now and gain an advantage on your competition by getting to know more about your customers and doing more to meet their needs. Providing personalized experiences for your customers will bring them closer to you.