No matter how the metric is defined, the pressure on marketers to meet key goals and prove ROI is increasingly mission critical. Whether acquisition, retention, higher sales, increasing net promotor scores, lifetime value or converting more loyalty members, marketers tend to be more keyed in on the result than the process.
In a previous blog, we examined how machine learning helps marketers achieve their goals by analyzing data at scale to provide the needed context that will ultimately produce a next-best action. We examined machine learning in the Redpoint rg1 platform as a key component of the actionable insights layer that bridges the gap between collecting/perfecting data and orchestrating a consistently relevant omnichannel customer experience.
With actionable insights, marketers are then ready to intelligently orchestrate next-best actions at scale to create a personalized omnichannel experiences for thousands (or millions) of customers. If you’ll permit me an analogy, the term “orchestration” is accurate. Just as a concertgoer may appreciate the active harmonies, crescendos and tonal melodies, her main appreciation is for how the sound brilliantly comes together, even if she remains blissfully unaware of the multi-layered complexity that brought it all together.
The concept applies to marketers working with machine learning. A marketer using machine learning output to select an audience may, for example, may want to know the propensity of a customer to buy a certain product. Perhaps that information is presented in a range from least to most likely to purchase. We explored how the audience insights dashboard presents decision tree logic to show why a customer is placed at one end of the range vs. the other, but ultimately a marketer who trusts that the information is accurate cares primarily about the fact that it is data that can be used to achieve a result. The marketer should not have to care, as it were, whether the information is generated from machine learning or not, only that it is available quickly and provides trustworthy insights. Instead, it’s “here is my data, here is my goal – how do I get there?”
Pre-Built Machine Learning Models
Continuing with the theme of not having to concern themselves with the details, marketers do not actively call for a propensity model, for instance. Rather, they’re defining the goal, with the model created through machine learning having been defined in that goal.
Marketers, in other words, are not required to create their own models. Because in many instances goals are shared across verticals, rg1 makes it easy to solve business problems by applying and building models as part of the CDP, allowing them to be delivered in different ways according to an industry use case.
A healthcare marketing organization, for example, might want to reach out to men over 50 to schedule a preventive screening. Rather than just send an email to everyone over 50, it would be more efficient and produce better outcomes if marketers sent a communication in the patient’s preferred channel, at their preferred time, and using content they are more likely to engage with.
A model using the organization’s own data might, for example, suppress people not likely to respond to an email. They’re using an attribute derived from a machine learning algorithm, but presented in language the marketer understands – least likely to most likely being one of many ways to intelligently orchestrate a next-best action. Through learning, the model knows why a customer is more or less likely to respond to SMS vs. email, or it knows the endless permutations that lead to a purchase vs. an abandoned shopping cart. The pre-built models produce the intended results because the algorithms have been trained on countless inputs that account for variables in the data. When a model that is initially fit-for-purpose is used to pursue a goal, the ultimate result can be brought back into the machine learning algorithm in a continual re-training loop for learning and adapting.
Another rg1 feature is its compatibility with what’s sometimes referred to as the bring-your-own-model (BYOM) approach to machine learning. During ingestion, a model result is provided as a data source – or the model is called within the rg1 real-time pipeline – to provide insights at the point of requiring them. As an example, if the real-time engine in rg1 is being used to drive a next-best action in a mobile app and the current geolocation is used to drive the model, the result can be gathered immediately to present the result in-line
In preparing data for orchestration, the bring-your-own-model approach recognizes that external investments in analytics should not have to be scrapped; data should be able to be taken from anywhere. rg1 can work with an external modeling system, with pre-built models making it easy to prepare data for use downstream, i.e. using it to achieve business goals and leaving all the heavy lifting to machine learning.