Kath Pay

Founder Holistic Email Marketing

Using Machine Learning in Email for 'Always On' Optimization

Using Machine Learning in Email for 'Always On' Optimization

See machine learning in action during "A Glimpse into the Future of Email Marketing – Reaping the Benefits of Machine Learning," featuring Kath Pay, Dela Quist, Skip Fidura and Jeremy Swift, May 19 at the Email Innovations Summit in Las Vegas.


"Machine learning" has moved out of science fiction and into real-life applications, like powering Tesla cars that run on autopilot and robots that can beat humans at the Japanese game of Go. For marketers, it gets them closer to their email nirvana: true 1:1 personalization on a mass scale.


Machine learning, at its simplest, is a method of data analysis that allows computers to learn – to analyze, predict and act – without explicit instructions or programming.

That last phrase – "without explicit instructions or programming" – highlights the difference between today's rule-based marketing automation and systems that use machine learning.

Both machine learning and today's marketing automation systems use algorithms to analyze data and set outcomes. However, even the most complex marketing automation programs rely on specific rules or conditions that users write for them. Those rules don't change unless someone goes in to alter them.

Out in the real world, though, customer behavior is changing constantly. Also, no two customers are alike. Segmenting the database and targeting messaging help marketers address these differences, but the rules that govern those activities don't recognize the subtle changes that build up over time into a mass movement. 

Systems that use machine learning to analyze data can generate insights that constantly adjust and refine the content sent to different customers based on their different characteristics and behavior.

Instead of sending messages based on one point of past behavior, these systems continually take in data, analyze it and use those insights to personalize messaging without requiring marketers or their IT cohorts to keep tinkering under the hood to keep up with changes.  

Machine learning in action

It's all around you, but it usually works so seamlessly you don't realize it's there:

Netflix: Netflix's recommendation engine analyzes data generated by three basic sources: your preferences (what you put on your list), your behavior (what you actually watch over time) and what other people are watching ("Trending Now"). Then it sifts through all that data gained by granular tagging of all scenes within the movies, to predict what you'll want to watch and presents those predictions as viewing recommendations.

As you watch (or don't watch) what Netflix suggests and change your personal viewing list, your recommendations list will change to reflect your behavior.

Twitter: Many social media listening tools use a combination of machine learning and linguistic rule creations to filter out nuggets of meaningful content from the fire hose of Tweets as they flash past, whether it's to detect customer sentiment in general or to learn what they're saying about you in particular.

Spam filtering: ISPs use a host of factors when deciding whether to route an email to the inbox, divert it to the junk folder or block it outright. User behavior such as clicking the "report spam" button, moving an email from the spam folder to the inbox and adding a sender's name to an address book or safe-senders list helps increase filtering accuracy for each account.

Airbnb: The travel service uses a dynamic pricing model that helps site hosts figure out how much to charge. It incorporates neighborhood location, amenities, time of year, fluctuating demand and other data points to help hosts get the most reservations at the best prices.

Machine learning for email marketers

Your fellow email marketers have been putting machine learning to work in their own programs. Here are a few examples:

Subject line optimization: Machine learning and marketing automation come together to help marketers choose the best subject lines with less time lost in testing. Here are two ways email marketers can use this technology:

Touchstone uses a proprietary algorithm to predict likely open, click and bounce rates using a simulation of an actual email database and comparing results to billions of other tests – using real data to power the results. Phrasee's language analysis tool predicts which emotional triggers in subject lines will drive more responses. Both services use results to refine and improve predictions over time.

Delivery time optimization: Also called "send time optimization," this service predicts and modifies email delivery times based on when recipients are most likely to see and open messages.

Copy optimization:Persado's persuasion automation platform uses algorithms to find the most persuasive language for direct-response marketing messages. Its Persado Go service analyzes message drafts (email, social posts, newsletters, ad copy, etc.) and suggests revisions using natural language processing and machine learning.

Newsletter creation:  Alchemy Worx has developed a newsletter automation service that streamlines newsletter creation and delivers 1:1 personalized content using machine learning that continually optimizes content choice based on recipient actions.

Real-tme Content Optimization:Cordial uses machine learning and proprietary multi-armed bandit algorithms in coordination with real-time systems to rapidly find content which is most likely to drive a conversion for each individual subscriber.  The system allows unlimited variants to be tested and self optimized while reducing lost conversions. 

Take the next step toward 'always on' optimization

The advent of machine learning doesn't mean you have to toss out all of your rules-based marketing automation. Instead, identify areas that would benefit from continuous optimization. That's where automation based on machine learning will drive better results without a constant injection of staff time and money.

The best way to see how machine learning solves the twin problems of lack of resources for optimization and scalability for personalization is to talk with email industry people who are making it work for their brands, clients and customers.

You'll actually be able to do that in a special session during the Email Innovations Summit (May 18-19 in Las Vegas). The panel includes Dela Quist of Alchemy Worx, who will show how machine learning powers both the Touchstone subject-line optimization service and Alchemy Worx' newsletter automation system.

Check the agenda to read more about the discussion session (Track 2, 3:40pm on May 19) and then register to attend.


Title: Using Machine Learning in Email for 'Always On' Optimization
About: Machine Learning and subject line optimization
Audience: Email Marketers
Publisher: OnlyInfluencers.com
Copyright 2015, Only Influencers, LLC

Last Thursday, people sat down with family and friends to celebrate Thanksgiving and eat their hearts out. Whether the tradition is to eat turkey, tofurkey, or chinese takeout, one things is for certain, most of the people who are busily enjoying a large meal on Thursday are ready to shop on Black Friday and Cyber Monday.  In early October, National Retail Federation released their forecasted numbers for 2016. While the 2015 holiday season didn’t perform as well as estimated, increasing 3.2% over the previous year, they’re forecasting in-store sales to increase 3.6% to $655.8 billion. Moreover, NRF is forecasting non-store (online sales) to increase a whopping 7-10% to as much as $117 billion.

At this point, retailers are in full swing, running promotions, stocking up for Black Friday and Cyber Monday, and organizing their team to execute flawlessly. What retailers may be forgetting is how to handle this influx of customers after the holiday season is done and gone. With more consumers turning to online shopping as the quickest and easiest way to get their holiday shopping done, retailers can enjoy more visibility online and more opportunities to turn browsing shoppers into loyal customers. However; retailers need a game plan quarterbacked by two key strategies in order to succeed: retention and predictive marketing.

Retention marketing, also known as lifecycle marketing, helps retailers speak to consumers wherever they may be in the buying lifecycle, from an onsite visitor to a one-time customer, to a high value, loyal customer. The 2016 Retention Marketing Report states that retailers have embraced the idea of retention marketing, with a 55% increase in retailers budgeting 30% or more to marketing to existing customers. The main channel for retention marketing, is, of course, email marketing. Interestingly, we found that retailers who are winning in the retail space and seeing a competitive edge are employing predictive data on top of their retention marketing strategies.

In fact, in the 2016 Predictive Marketing Report, we found that, anecdotally, retailers who have invested in predictive marketing are seeing increases across the board from sales, to engagement, and even inventory management. These findings jive with what we’re seeing in the industry. A recent Forrester Report states “predictive marketers are 2.9x more likely to report revenue growth rates higher than the industry average.” Additionally, Salesforce Marketing Cloud found that 79% of top-performing marketing teams are using predictive intelligence to inform their marketing communication and strategy.

So how does all of this fit together? The first step to get started with retention marketing or to add predictive marketing is getting access to your data in an actionable way. Having access to product, purchase, and customer data in your email marketing platform allows you to start slicing and dicing your customer list by key features, such as, last order date, products purchased, geographic location (for in-store promotions), etc. This enables retailers to ensure they’re not sending Harry, who lives in Montana, promotions for women’s bathing suits in the middle of winter or Joan, who lives in Florida, promotions for a brand new snow shovel.  Once retailers have set up foundational retention marketing campaigns - first purchase series, abandoned shopping cart campaigns, browse abandonment campaigns, and a best customer series - it’s time to sprinkle in predictive data and create predictive campaigns.

Predictive marketing can take form in a few different ways in the world of email marketing.  If you’re just getting started, I recommend starting with the low hanging fruit. One of the easiest steps you can take is enabling dynamic product recommendations for your existing email campaigns. Dynamic product recommendations are populated based on buying trends of the individual person and the trends seen in the retailer’s aggregate customer base. This one addition creates a more personalized experience for each customer. 

Once dynamic product recommendations are added, it’s time to move on to using predictive data - predicted replenishment date (for consumable products), predicted gender, predicted order date for a predictive win-back campaign, etc  - to trigger and build out campaigns.. Depending on your industry, certain campaigns and promotions may work better than others.


  1. com, an IR1000 retailer (#617), a web-only retailer of coffee, tea, and related products, created a replenishment campaign based on predicted replenishment date. This campaign, pictured on the right, triggers based on each customer’s buying cadence. This campaign’s revenue per email (RPE) is $0.73. Note: The industry average is $0.11.

Similarly, Artbeads.com, also an IR1000 retailer (#550), a bead and jewelry-supply online retailer, created a replenishment campaign based on predicted replenishment date and has seen a 161% increase in opens from this particular campaign.

In both instances, the retailers are leveraging dynamic product recommendations in their emails, as well as relying on the predicted replenishment date to trigger the campaign. 


For US based stores, we’re able to predict with a 99% confidence rate the gender of each customer in your database. This data can be extremely useful when you’re creating your promotional calendar. While the promotion might be the same, the copy and products your pushing may vary based on gender. Additionally, you can employ this information to create suppression lists, so Harry will no longer get promotional campaigns for women’s swimsuits. This not only helps create a more curated and personalized experience for your customers, but cut down on unsubscribes and spam reports.


Predicted order date can be used in a number of ways, but primarily we tend to focus on creating a predicted win-back campaign. A static win-back campaign uses a set number of days past the purchase date as their trigger date, generally retailers will look at their average latency period to help determine the timeframe; however, this approach leaves room for error. For any customer who has purchased 3 or more times, retailers can employ predicted order date to create a predictive win-back campaign. This campaign is triggered based on the predicted order date for each customer.

SurfStitch, Australia’s number one surf and lifestyle brand, created a predicted win-back campaign. Running both a static for 1-2 time buyers, and predictive campaign for 3+ buyers, SurfStitch has seen a 72% decrease in churning customers.  To top it off, SurfStitch capitalized on their existing copy and creative when creating

Pennington & Bailes, Windsor Circle’s green pants provider, also created a predictive win-back series. They have seen a 62% lift in opens and a 137% lift in clicks from their predictive win-back series when compared to their static win-back campaign.

Retailers are just beginning to scratch the surface of what predictive marketing can do. As consumers demand a seamless experience across devices, regardless of whether they shop in-stores or online, it’s imperative that retailers create the curated, personalized experience consumers are looking for. To learn more about predictive marketing, download the 2016 Predictive Marketing Report and make sure to keep your customers after the holiday rush.

The just-concluded Presidential election was – in a word – interesting.

As a marketer, it was fascinating watching two diametrically opposed candidates take each other on. This was not a test where one candidate was just slightly different than the other – it was a case where the candidates could not have been more opposite.

 While you may not be thrilled with the results, there were three key things that all marketers - and especially email marketers - can take away from the drama to make each and every one of their programs be more successful.

Digital marketing leaders continue to promise consumers a true 1:1 personalized interaction. We have all said it, preached it, and many of us have it tattooed somewhere. Right Message. Right Time. Oh, Right Channel. Recently. Right Person.

Simple: load up the data, drag and drop, press magic button, personalization.

We already know that email can form the hub of your digital marketing program, with the email address housing all of your information about each customer in your database. Now, I want you to think how email testing can also drive your multichannel testing program to gain insights across your entire customer database.

“Why is my mail being blocked if I still get spam?”

It’s almost an inevitable question when handling delivery issues. I understand why I get it so often. People look in their inbox and see this mail is clearly spam and it’s in the inbox. But they look at the mail they send that they know isn’t spam and it ends up in the bulk folder. It’s logical to ask why legitimate marketers have to follow all these complicated and arbitrary rules to reach the inbox when spammers reach the inbox and they don’t follow any of the rules. 

Cron Job Starts
Subscribe to the Only Influencers Newsletter
  • Email Marketing News
  • Latest Email Tool Reviews
  • Email Marketing Jobs
  • Top Email Thought Leaders