I recently read an article on how a major ESP was going to bring Artificial Intelligence (AI) smarts to email marketing campaigns. As I read through the piece, I was both concerned and frustrated with what little it said about what this new technology actually meant for customers. The increased use of words like AI, Machine Intelligence, Machine Learning, Deep Learning and Cognitive Computing to drive buzz around various products does little to clarify how these technologies might actually benefit marketers.  

One problem is that as the marketplace throws around these words interchangeably, offering little differentiation among the terms themselves. Believe it or not, there is a difference. So, before we dig into the practical implications of these technologies, let’s dive into the terms. Once we understand the differences, we can discuss how these technologies are or are not connected to email marketing.

Artificial Intelligence (AI):

In 1950, Alan Turing forever shaped the definition of AI by proposing an imitation game,” in which a computer had to behave in such a manner that a human was unable to distinguish it from a person. This view of general AI shaped research for decades and eventually led us to the weak (narrow) vs. general distinction we see today.

Narrow AI is a machine that emulates some facet of human intelligence and does it well but is lacking in many other areas.  For example, a machine that is great at recognizing sound or imagery — but nothing else — would be a great example of Narrow AI.

General AI is all about having the characteristics of the common definition of AI. General AI essentially mimics the characteristics of human intelligence (i.e., think Rosie from The Jetsons).

Mentioning AI raises people’s eyebrows because it’s new and exciting. While recent advances in AI technology certainly are exciting, the concept itself is almost 62 years old. That said, the rise of applications based on AI and the expansion of the Internet of Things (IoT) gets us at Trendline excited because it will ultimately change the way we experience the world and is already beginning to change how we leverage technology to create more personalized experiences for email subscribers. However, any company that says it is applying AI to its product without further clarification as to what that means is simply looking to capitalize on current hype rather than the practical application of how it’s going to improve your email program. The first question is, “Will the technology apply Machine Intelligence, Machine Learning or Deep Learning?” The second question is, “What problem does the product they are pitching solve?” Lastly, companies that are pitching AI as a solution often gloss over one important aspect of what makes these technologies worthwhile: lots and lots of actionable and clean data. This is a HUGE challenge for many organizations today, so any technological claim needs to be clear about what data inputs will be required in order to make the solution effective.

Machine Intelligence (MI):

Machine Intelligence and Artificial Intelligence are interchangeable terms.  The use of the term MI has been popular in Europe, while AI has been more popular in the US.

Machine Learning (ML):

Machine Learning is a subset of Narrow AI, which is broadly defined as enabling machines to improve on tasks with time and experience. In essence, it is a learning algorithm that generates more algorithms from the data that is fed into the system. Then, based on certain instructions and rules, the machine creates a new set of rules from said data.

There are three primary methods for how machines learn:

  1. Unsupervised learning: The data given to the learning algorithm is unlabeled, and the algorithm is then asked to find patterns within the input data. Good examples of this in email are browsed/abandon emails or transactional emails containing product recommendations based on similar items viewed or bought. Yes, folks… that is what Amazon does and has been doing for a long long time.
  2. Supervised learning: The data is given to a learning algorithm and is told what the desired output should be. A good example of this in email is spam filtering/detection in Gmail. In reality, Gmail is learning to determine whether or not the email you send as a marketer falls into Spam/Not Spam folder. That said, the next time you are battling a deliverability issue with Gmail, know that you are up against a very smart machine that is learning about predicting an outcome.
  3. Reinforcement learning: In this type of ML, the machine is not given examples of correct input-output pairs, but the machine has a method to quantify its performance in the form of a “reward signal.” Reinforcement learning is supposed to emulate how humans and animals learn, which is to say it tries a bunch of things and is then rewarded when it does something well. One example is the work being done by IBM Watson to automatically tag images. In this scenario, Watson can learn to recognize objects such as trees — or even specific species of trees — and associate meta tags with the image. People then have the ability to correct Watson if the wrong tag is inserted. These corrections help optimize performance so that Watson becomes that much better over time at recognizing different people, places, or objects. The IBM Watson team and Saks Fifth Avenue also recently published an academic article on the use of reinforcement learning in cross-channel optimization for DM.  If you care to get into the nitty-gritty, you can read more about this case study here.

ML is also not new. It’s been around since the 1980s, but current growth in the ML world has been due to data availability, computing power, and the innovation around algorithms in layered neural networks (AKA Deep Learning).

In the email marketing world, ML examples can be found in applications like Send Time Optimization, as well as the use of natural language processing (NLP) and engagement history to optimize subject lines, body copy, CTAs, and so on.

Deep Learning (DL):

Deep Learning is a subset of machine learning, and its only been around for about seven years.  Deep learning is cool stuff because it’s a special type of ML algorithm that uses large data sets with a multi-layered neural network (a computer system modeled after the human brain), all with the goal of using new data to make predictions. There are two phases in DL: training and inference.  The training phase is massive because a tremendous amount of computational power is needed to get to the inference phase.

Deep learning in email is years away for most marketers because of the sheer amount of centrally located data required to apply these technologies. In this case, technical capabilities are outpacing most organizations’ ability to leverage them, given that data is often widely dispersed and data teams are often overextended trying to centralize their data.  

Still, we believe deep learning has practical applications that can be leveraged as data becomes centrally available. For example, imagine the ability to assemble tens of thousands of data points for both offline and online behavior (in random things like TV shows watched, browse history, opening of emails across the entire internet, etc), add in demographic and psychographic data, location and life stage history, and then layering that into a learning algorithm in a multi-layered neural network to predict the consumer’s next search, site, purchase, interest, etc. And then imagine sending subscribers a message at the moment they are thinking about those things to nudge them into action. Scary and totally exciting stuff ahead.

Looking Ahead

We at Trendline are excited about what AI and ML can and will bring to the email marketers of the future, but we strive to balance this enthusiasm with understanding the practical applications of this technology in any email program. As the AI/ML hype cycle grows into a fever pitch, these technologies still need to have clear objectives and realistic goals. Only then can we determine if the return on these investments is going to be positive.

To that end, Trendline is actively involved in the implementation of AI- and ML-based technologies across a few of our clients. As with any other new and exciting development, we first identified the problem that needed to be solved, found the appropriate technologies to solve those problems, developed KPIs of success for each client, and are tracking performance diligently via our analytics practice. AI in email marketing is here to stay, but the temptation to buy into the hype must be tempered with the sound application of tried-and-true marketing and business principles.

And of course, if you need help in doing so, Trendline would be happy to partner with you to maximize your investment and reap the benefits of these technologies as they evolve. Contact us today.

 

Part 2 and beyond in this email series will focus on how Trendline is working with partners around the practical implementations in AI and how they are using the various types of ML to make email more advanced, efficient, and profitable.

 

 

author image

Andrew Kordek