Jul 29, 2019

How to use Natural Language Processing and Machine Learning in your Email Program

How to use Natural Language Processing and Machine Learning in your Email Program

How to use Natural Language Processing and Machine Learning in your Email Program

Do we really need to remind anybody about the fact that email (and email marketing) isn’t going anywhere anytime soon? If we did, we’d just flash this study by The Radicati Group on them, containing such zingers as…

By the end of 2019, the number of worldwide email users will increase to over 2.9 billion. Over one-third of the worldwide population will be using email by year-end 2019.

For those of us who work in email? Stats like that are pretty temptacious, as the kids say. But today’s email isn’t your mom or dad’s email. The continuing success of email lies, in large part, to how its ability to evolve. Going mobile put email into a lot more pockets, for instance. 

Now, with the arrival of AI-related technologies, your email campaigns can become even more precise, engaging, and effective than they’ve ever been.

The arrival of email AI? That’s so 2018 

At the end of 2018, PwC said it had surveyed U.S. execs, and found 27% of them claiming to be already implementing AI in multiple areas. 

On the global front, 30% of companies worldwide will be using AI in at least one of their sales processes by 2020. But only 17% of email marketers considering automation tools gave any thought to incorporating AI.

The laggards might not realize the impact AI has already had on the email ecosystem. One very visible example was how Gmail handles email classification using Natural Language Processing (NLP) to filter incoming emails as Primary, Social, or Promotions messages. 

Here’s a pretty good explanation of how NLP does its job, presented as a primer for coders who want to hack up a spam filter. But if you aren’t interested in all the plumbing, that’s cool. One thing worth remembering, though? NLP and machine learning are only branches of the bigger, broader category “AI” and have specific goals.  

  • NLP is intended to read, decipher, understand, and make sense of human language in a manner that’s useful in machine-human interaction. 

  • Machine learning involves the application of algorithms and statistical models so computers can make decisions and perform tasks without explicit instructions by recognizing patterns in data and drawing inferences.

Right now, there are multiple tools and tactics where NLP and machine learning are being put to use to enhance email programs. Let’s look at some of the places where you could integrate them into your campaigns, shall we..?


With machine learning, you can now execute multi-armed bandit testing. If you’re used to split testing, brace yourself: Now you’ll be able to run tests continuously and put your findings to work immediately. Over time, you’ll gradually optimize your results, and simultaneously be able to test content and messaging while also sending your best-performing variant out to prospects or customers.

How’s it done? You set up a campaign and a few email variations, and machine learning does the rest, running tests throughout your campaign and fine-tuning it on the basis of test data. What can you test? Pretty much anything you’re already testing, from copy to design to images to timing. 


Machine learning and NLP – and its cousin, Natural Language Genration (NLG) – are being leveraged by multiple providers to deliver solutions that can actually generate subject lines and other copy.

Take a company like Persado, for instance: Its “message machine” applies its grasp of natural language to create copy that speaks in the marketer’s “brand voice,” leveraging a huge database of tagged and scored works in 25 languages, a database that evolves over time as machine learning delivers insights (and makes judgments) about which messages hold the most appeal for your target audience.

Touchstone, as another example, compares your subject line against a database of 21 billion emails, as well as industry trends, to predict its likely impression, click and conversion rates. automated the newsletter creation process, and uses machine learning to optimize content based on each recipient’s behaviors to provide 1:1 personalization that’s “tailored to your subscribers’ unique interests and personalities, without the time it takes to manually curate your emails.”


Want to pull off a little real-time content optimization to drive engagement? Cordial says it can “ingest and process customer event, behavior, and purchase data from virtually any source,” so messages can be dispatched across multiple channels, based on up-to-right-this-instant behavioral data. So onboarding, re-engagement campaigns, and other triggered emails can be aligned with what they’re interested in this very moment.

Another way to engage? Add a personal touch. Well, a virtual personal touch: Conversica proudly claims to deliver “personalized human touch at scale” through AI sales assistants that reach out to a user within minutes of him or her showing interest in your brand or inventory via email or SMS. 

If you’re worried the “conversation” reads like robo-copy, they claim the AI “empathizes” effectively by analyzing replies to tailor the right responses.  Moreover, the platform isn’t intended simply for initial engagement or onboarding but can handle routine dialogues throughout the entire customer journey.


For companies investing in customer data management platforms, being able to milk the greatest possible insight and benefits from big data to deliver highly personalized user experiences, especially in email, is an obvious concern. 

A machine learning solution that’s connected to these potentially enormous pools of data can do insightful segmentation in ways no human being – or boiler room full of human beings – ever could, making continual adjustments and uncovering new associations, even generating product new segments where none were visible before.  SimMachines is one of these providers, calling their particular flavor “dynamic predictive segmentation.” 

Predictive delivery

If you haven’t heard of it before, that’s because it’s a new wrinkle in applying machine learning to email. By analyzing the behavior of trillions of emails, predictive analytics and machine learning are able to optimize delivery and the overall health of an email program.

This means real-time insights are available about deliverability and performance issues, problems can be identified before they happen, and data-driven recommendations can be made about how to optimize engagement and performance.  Outages can be avoided – while ROI is maximized.

And if you’ll allow just one self-plug? It’s new to the game because this platform, SparkPost Signals, is the first and only email intelligence platform of its kind in the industry, and we’re proud to be offering it.

It’s an AI-for-email explosion

These are just a few of the areas where AI, NLP, and machine learning are making a present-day impact on email marketing. If you think it’s the tip of the iceberg – or the first trickle through the floodgates – you’d be right.

One way to see how feverish a new technology segment is getting is to see how many companies and startups have hung out a shingle, using investor or job sites like AngelList. Right now, a search for “email AI” there shows over 600 firms in the space, and there’ll be more to come.

In other words, there’ll eventually be an AI add-on for every facet of your email program.  In the meantime? Putting today’s existing AI tools to work already offers plenty of potential for discovering how NLP and machine learning can improve the way you’re using a veteran marketing channel that’s just as leading-edge as ever.