The promise for artificial intelligence (AI) and machine learning in email marketing is exciting. Unfortunately, a lot of the conversations about the uses of AI are in the abstract or focused on a very few use cases. I’d like to open the aperture a little and take a look at a few more specific ways that machine learning and AI can be used to improve the efficacy of email marketing.
Content: The content of messages (including the subject line, product on offer, call to action, discount offered, and other features) is one area where new machine learning and AI techniques are making inroads. For online retailers, for example, often the most common application is serving up product recommendations based on website behavior. Many retailers are interested in current applications of AI to optimize subject lines.
How do we go beyond product recommendations and subject lines? There are many other features of an email message that can be optimized based on the past behavior of each subscriber, and automated experimentation techniques such as reinforcement learning models. What kind of discount or other offer drives a higher lifetime value for each subscriber? What call to action is likely to drive more clicks and conversions? These questions are discoverable through machine-learning techniques.
Send time: This is another area where there has been a good amount of investment by email marketing vendors. Using historical data, it is relatively easy to estimate an optimal time for 40%-60% of subscribers (those that are most active). Sending at an optimal time has been shown to drive more opens and clicks.
Unfortunately, because a large portion of email marketing lists are dormant, it’s hard to determine an optimal time for each subscriber. Some vendors are building cooperative databases using the experience of other marketers to make it easier to find optimal times for more subscribers on the list. It is also possible to capture data from other channels (e.g., app usage and website visits) to build better estimates for optimal times to send email.
Cadence: Each subscriber has an optimal frequency—particularly for campaign-based mail. Since most marketers send roughly the same amount of campaign-based mail to each subscriber, marketers are both under-mailing and over-mailing at the same time.
Many marketers find to difficult to find this optimal cadence through traditional marketing analytics techniques. One survey showed that over two thirds of marketers think they can improve frequency and have tried to develop a solution. Most of these respondents had abandoned their attempts at frequency optimization.
This is a problem that AI can solve. Experiments using sophisticated AI approaches to individualize frequency have shown significant increases in revenue and clicks (up to 40% increase in clicks for marketers that are mostly under-mailing) without driving significantly higher complaint rate or unsubscribes.
Structure of triggered messages: How many messages should I send in my abandoned cart series? What is the right content to have in each message to drive highest conversion rates and revenue? Today, these questions are answered by painstaking testing that may require many iterations. This is another problem that is amenable to solution using machine-learning techniques.
Email address capture: What content and call to action is going to optimize the capture of quality email addresses? As with the “content” section above, there are many levers that can be used to optimize email address capture. The kinds of offers used in email capture have been shown to affect the quality of the addresses (e.g., are these addresses that are active in email). AI-driven tools can optimize for both the quantity and quality of addresses going forward.
These aren’t the only applications that creative analytics and product teams will deliver of the next few years.
What about you? What ideas do you have for using this exciting new technology?