AI and Machine Learning drive Email Marketing
Marketers strive to personalize emails in the quest for relevance. This is not new to seasoned marketers, who have been using marketing automation platforms to create customer journeys or leveraging segmentation and/or personas.
This is a good usage of data and tools, but it simply can’t scale beyond a certain point. In our experience, most clients have around 20–25 data segments. For relatively large user bases, too many journeys and segmentation could potentially ruin the user experience, and complex rules make it hard to maintain.
The genius of ML and AI can save the day! AI power, combined with human intellect (IQ), is going to create a hyper-personalized experience to the user, which will compel them to remain engaged with the brand.
Broadly there are four major areas where AI and ML are helping email marketers today.
1. Content Selection: What content to send over email
2. Data Segmentation: Whom to send the email
3. Send Time Optimization: When to send the email
4. Delivery Optimization: How to deliver the email
Content is the most crucial part of an email marketing strategy controlled by the marketer. Some marketers have been optimizing content for a long time now, with varying results — based on demographic and behavioral data points available.
However, the optimizations were mostly based on manual observations and are limited by the sheer volume of opportunities to change the emails. ML and AI add value, where the content is optimized per user.
Say, for an eCommerce site, the level of customizations could be: John likes the latest mobile phone accessories — the email body and subject lines are optimized for him. Jane likes the latest fashion — the email body and subject lines are optimized for her.
The optimization would be applied anytime an email is sent to John and Jane.
Early forms of personalization and relevance are things like abandoned cart emails and other mappings between the user’s behavior on the website, bringing that intelligence back to email.
Now the opportunity is more significant is we get to a level of “hyper-personalization.” In the world of “hyper-personalization,” every user will have custom content delivered. Some solutions are available already:
Subject line optimization: Where subject lines are optimized based on previous history, an input of spam words to avoid, and can generate multiple subject lines in seconds, that used to take marketers hours to create.
Email body copy optimization: Like subject lines, body copy can be created and adjusted for the user on the fly based on behaviors from the past. For example, if Ellen has been consuming information on a news and information site about politics, her e-newsletter could begin with politics, then followed by other categories.
But if George spends all his time in sports, followed by entertainment, his e-newsletter would be very different, starting with Sports, and then entertainment. This might sound familiar to some readers, as dynamic content emails have been around for quite some time.
But most of those emails were created when users provided their preferences. AI and ML come into action by evaluating the content on the website and changing the e-newsletter from week to week, or even on the fly when a user begins to explore a new category of content — all based on the rules that are set.
Optimizing Call to Action (CTA): Any element of an email that can be generated dynamically can be switched out as required, and we’ve been doing this for many years. But what if a new variable comes in? How quickly can we handle the changes in the platform and the emails to address, without redoing them?
AI and ML will be able to not only help us find the nuggets within our data and the user’s behavior that will help us engage users faster, but it will also enable it as well. Imagine that a user has not been accessing technical information on a website, but their behavior changes, and they become very engaged in technical documentation.
This could be a signal that their next email CTA should be an offer for a whitepaper, all happening automatically through AI.
Cross channel activity feeding into email strategy: AI and ML will make our wishes come true, as it pertains to cross channel activity. How many times have you heard someone say, Right message at the right time in the right channel?
This is where it can finally happen, as AI and ML take on the task of determining those moments through on the fly data analytics, current behaviors, and channel activity. Cross channel creative and content will be customized, nearly at the moment that the customer accesses it. Along with email, channels include SMS, browser notifications, app notifications, app content, web content, phone, and even direct mail.
Some advanced marketing automation platforms are offering parts of this as a solution already, while most of the other solutions are marching towards it. Eventually, all of this should be a standard offering.
Understanding their user base is a fundamental need for a marketer. Each marketer should know as much as they can about their customers. To that end, usually, marketers segment the users based on demographics (age, location, etc.), marketing segments that they may have created in the past based on purchase behavior, or email behavior like email activity (opens, clicks, etc.). Traditional approaches are usually based on experience and limited data points.
Segment of ONE is the future state. While we will still need to understand our customers, it will be more critical than ever; when we market to them, we’ll be focusing on what individual customers need. It’s not possible to do this when using segments and after-the-fact behaviors purely, but with AI and ML, it is possible.
Many marketing automation solutions are scratching the surface of this. Realtime and accurate marketing segments based on customer insights drawn from cross-channel activities help brands communicate with users according to their preferences.
Two basic approaches can be used, which will no doubt be expanded on as we move into this “new world” of marketing. Experiences can be created individually based on each user’s behavior, or marketers can leverage modeling to create journeys based on activities of similar users.
When looking at individual users, each journey will be different. Like snowflakes, no two journeys are alike. But there are similarities in how users engage with a brand, which brings us back to personas and/or segments.
Individual Journeys or Segment of ONE: As marketers begin to think through journeys based on behaviors, rather than planned steps, new sets of requirements will start to unfold. Content will be one of the primary considerations, as customers step from channel to channel.
The big fear as marketers in the past was always that they would have to try and understand each moment and create all this content in advance. AI and ML are changing that concept. It will come down to the rules that are set and the related content for those behaviors; the rest will be up to technology. While pieces of this are available today, more work will need to be done for true cross channel journeys on an individual level.
As an alternative to creating individual journeys, marketers will have the option of using all their behavioral data in aggregate, or by a level of segment. This will feel a little more comfortable, potentially, for marketers who are accustomed to marketing segments and persona-driven marketing. Using combined customer behaviors to make decisions on triggering an email or SMS and continuing to learn from those behaviors is still an advanced state of marketing over what many companies do today.
Of course, no matter which path is taken, there is homework to be done to understand how customers engage with a brand. The first step is a map of the customer journey, looking across all touchpoints used when accessing a brand and its products.
This can be an extremely intense undertaking, so be prepared, especially if the brand has brick and mortar locations. Marketers can begin the process by talking to customers and asking them how they choose services or buy products. Once they have a good understanding, they will be equipped to leverage the power of AI to get the right content to the right user, at the right time and in the right channel.
Send Time Optimization
Send time optimization is not new to email marketers. Many of them are doing it in at least one way — based on the time zone of the user, or the best time to generate responses. There is no point delivering email at midnight for users in Cleveland and the same message sent to users at 8:00 am in Dubai, in a single send with the same message.
If a brand has users from multiple time zones, the delivery times must be optimized per time zone, allowing the marketer to adjust their content accordingly. In the past, marketers may have tried to get at the top of the inbox by sending late night, or early morning hours. Many emails are still being sent in this period, but with the volume of email that is being sent out and received by users today, it would be better to try and get into the inbox when they are actively looking at their emails.
Manual techniques limit the targeting to a bucket of users based on some primitive heuristic method, but with machine learning, the optimization is 1–1. The time to deliver a particular message is determined based on when the user is actively checking email, based on the previous behavior of that user.
With Send Time Optimization, marketers can begin to get serious about Flash sales or limited time offers, knowing they are more likely to be sent when the user is active.
Delivery optimization is usually invisible to the marketer, but if the email service provider (ESP) is using some advanced machine learning and AI techniques to optimize delivery for better inboxing and reputation building, this should provide significant stability in overall reputation and a jump in ROI of email campaigns. There are a few ESPs out there today who power delivery with AI.
Two ways ESPs can optimize delivery for better inboxing are Domain/IP Warmup and Email Delivery Optimization:
The goal of a warmup process is to build a stable and good email reputation for sending domain and IPs at various Internet Service Providers (ISPs) like Gmail, Yahoo, etc. Every domain must go through a warmup process when it starts sending emails.
The premise of the warmup plan is to send emails starting 100 or so on day 1 and gradually increase the volume with positive consistency and build a reputation over a period of 3–6 weeks. During the warmup process, the number of emails a marketer should send each day is based on the performance of the emails previously sent, plus the engagement those emails generated.
Depending on the success of the previous emails, the strategy may need to be adjusted. Machine learning can help determine the exact strategy that’s required in order to continue and improve warmups.
Email delivery is a challenging task by itself. There are different types of emails — Time-sensitive (Ex: sign-up confirmation emails, purchase confirmations, password resets, etc.) and Delivery sensitive emails (Ex: promotional newsletters).
Since their needs are quite different, the ESP should treat both differently with adjusted delivery throughputs accordingly. Further, there are cases where the delivery layer of an ESP must look for minute signals from ISPs like throttling (where the ISP is slowing down the delivery of the emails), identify the cause of throttling and fix it to prevent further damage if needed.
For example, if a sender (or marketer) hits more than a certain number of invalid recipients in a minute, ISPs like Gmail would grow suspicious and slow down the acceptance rate of emails from that marketer. The AI layer looks for and understands these minute signals, fixes the root cause, and feeds this as a pattern to not occur again, safeguarding the email infrastructure. While AI and ML in this situation is not something that a marketer can see outwardly, they will benefit considerably.
When these situations occur today, it is reliant on people to see the issue, which usually means the problem isn’t found until it’s too late. Add this element of AI and ML to the requirements for the next ESP list to consider when contracts are drawing near.
The industry has been using words like “Right time, right message, right channel” for many years. Before the Internet, with fewer options, data and predictive analytics could do a fairly good job of finding the right targets to market to, and then the message convinced that user they needed something. Fast forward to now, when users have so many options and new behaviors that were never a consideration before.
Research that used to happen at the dinner table or water cooler at work is now a major step in the journey process. Consideration with review sites, comparison sites, all getting closer to decisions. Search and digital advertising’s role in all of this and the signals that occur each step in the process that users take.
The customer journey is complex. The good news is that AI and ML can help marketers move to the next level of marketing.