Last updated on June 9th, 2023 at 10:49 am
A company called ActiveCampaign is trying to adopt machine learning in e-mail marketing in order to help marketers frustrated by A/B tests that fail to improve the results of e-mail campaigns with a machine learning-based application designed to predict what’s most likely to trigger an opening or other form of engagement.
Based in Chicago, ActiveCampaign had already launched a tool called Predictive Sending this past May. However Predictive Content, which will be available for the Professional and Enterprise versions of its platform, focuses more on what’s being sent in order to help personal messages to drive better experiences for recipients.
Those who use ActiveCampaign’s Email Designer tool, for instance, will be able to create five different versions of a single message. These will be associated with preference profiles based on who will receive it.
According to ActiveCampaign’s vice-president of product, Jen Busenbark, the dataset of nearly two million words spanning 300 unique dimensions was built over the course of several months by the firm’s data science team using data from millions of anonymized historical campaign emails. The launch of Predictive Sending, meanwhile, offered a glimpse into how marketers needed to be more strategic about the content of a message.
“We knew that we needed to make these fairly complex and robust features extremely easy to use,” Busenbark told B2B News Network. “In the case of Predictive Sending, users simply need to click a toggle to use it. We carried that focus on simplicity and usability into the design of Predictive Content, where users simply write their emails and the feature does the rest.”
Its machine learning technology utilizes a dataset of nearly 2 million words spanning 300 unique dimensions, or ways to look at those words. The model is able to identify and adjust over time whether or not an individual likes long or concise emails, prefers a serious or fun tone, the topics they engage with most, and much more without requiring input from ActiveCampaign customers.
ActiveCampaign customers simply launch Predictive Content from within Email Designer, where they can create up to five different versions of an email. Once triggered to send, Predictive Content automatically builds a preference profile of each person receiving the email and matches one email version to each person, ensuring optimal engagement.
Marketers can write as few as two or as many as five versions of an e-mail message and Predictive Content matches each person being emailed with the version they are most likely to engage with, Busenbark said. There are no inputs they need to enter to build the preference profile of each person. Users don’t need to do anything differently whether they are sending to a group of 10 people or 10,000.
The preference profiles, meanwhile, consist of complex vector data that Predictive Content builds to match the optimal email with each recipient.
“This is not something that’s digestible for users today, although we see some very interesting and valuable opportunities here,” she said. “We plan to provide reporting that customers can use to draw conclusions about their e-mail audiences by seeing the email versions sent to each recipient, as well as engagement data such as open rates and click-through rates.”
Machine learning in e-mail marketing would help B2B firms in areas like account-based marketing (ABM), meanwhile, given that it could ease the process of creating granular messaging at the account level. Using Predictive Content, an ABM campaign could be based on several different versions of one marketing update, which is then matched against the most engaging and relevant version for each person receiving the message at that one account.
“For example, you could imagine selling a product to an HR department and want to engage both HR as well as leadership across departments,” Busenbark said.
Although Predictive Content was designed to measure probability of engagement against any conversion ActiveCampaign’s customers, Busenbark said metrics like clicks could eventually give way to purchase data or other types of conversions.