There are many business use cases for artificial intelligence that a customer may consider as augmented processes, but that have little to do with machine learning. Chatbots or conversational AI that use text analysis to quickly answer a question. Facial recognition software for hassle-free access or authentication. Patient triage or prescription auditing systems for the healthcare consumer. Across all industries and departments, there are interesting, diverse and growing sets of use cases for enhancing certain tasks in an automated fashion using intelligent systems. But for business users trying to understand, map and improve customer journeys in ways that they could otherwise not accomplish, the primary focus of using AI tooling is to build and optimize machine learning models.
An interesting irony here is that using automated machine learning (AML) to enhance customer experience (CX) will usually never overtly reveal itself to the end consumer. Unlike a chatbot that customers likely recognize as using AI, but that does not materially improve their end-to-end experience, a customer on the receiving end of a consistently relevant, hyper-personalized CX throughout an entirety of a customer journey is usually unaware that machine learning is largely responsible for the behind-the-scenes magic. Yet even if the accolades go to AI, machine learning does the hard work of truly understanding a customer. It is why marketers turn to machine learning modeling to analyze customer cohorts, single customer view, behavioral and transactional patterns and other numerical data sets to provide a better understanding of who a customer is and what they want.
Arriving at a better understanding of a customer journey is something marketers have been doing for decades. Compiling and analyzing average monthly spend, lifetime value, average transaction value, time on a web page and other statistics about a customer are not new, of course. They have and will continue to have tremendous value to a marketer. But those statistical calculations do not, for the most part, require a machine learning model.
Where AML separates itself in shedding light on the intricacies of a customer journey is the ability to take large amounts of data that have no obvious, simple correlation between the question being asked and the information in the data and figuring out an answer. In other words, if you cannot simply calculate it, aggregate it or average it, you likely have a very good business use case for AML, especially if it needs to be at scale.
Two Types of Machine Learning Models
A key benefit of using AML to help guide customer journeys to the desired conclusion is that marketers do not have to get their hands dirty, if you will, by becoming involved with the sequence of steps needed to build models, from data prep through continual measurement of the model. Nor do they need to interact with data scientists or data engineers to develop an understanding of the sequence interplay. But to be able to develop applicable use cases for what they’re trying to accomplish, they should at the very least have a rudimentary understanding of how machine learning works, and how automation reduces the need for human involvement.
There are effectively two different classes of models that marketers will build – unsupervised and supervised. The former entails a marketer providing data, and asking the model to figure out interesting patterns in the data and what the data might say about a customer or segment of customers. The latter – a supervised model – is when there is a predetermined attribute a marketer wants to find, and where the marketer provides historical data to build a model that will sort through the variables to find what may be a good predictor of the attribute – whether it’s customer lifetime value (CLV), propensity to churn or another characteristic that says something about the customer. Data on who increased basket size in a cross-sell/upsell campaign, for instance, is fed into a model to predict who may buy more of a product or a more expensive product in the future.
The reason it is important for marketers to have a base level understanding of the distinction between supervised and unsupervised models is that ultimately they are the ones asking questions of the data, and they have a sense of what data they have on-hand that might contribute to building a successful model. They may know they have to turn over a certain number of rocks, but they lean on machine learning to determine which ones.
Uncovering an Audience with an Unsupervised Model
An example of each model type may better frame the picture in terms of how a marketer poses questions of data. A simple unsupervised model might be where a marketer has an audience, and knows they want to put the audience through a multi-step, multi-channel campaign – but first they need to understand how the audience divides out. Breaking down the audience into subsets – on classifications determined by the machine learning model – will help the marketer decide who receives a certain offer on a certain channel. Here, the model discovers data correlations that reveal something interesting about the audience. In these types of scenarios, unsupervised models are the perfect use case for audience segmentation. Rather than an artificial, manual segmentation based on intuition or arbitrary cut-offs (men 18-34, Massachusetts residents, income over $75,000, etc.) the model will segment an audience based on what – according only to the data – is important, interesting or unique about a particular audience.
Embedded automated machine learning that is part of the Redpoint rg1 platform goes beyond segmenting an audience to let marketers know why it arrived at a conclusion. With an understanding of the decision points – why an audience member was put in one segment vs. another – marketers not only have a better sense of the data they already have on-hand, but they are then able to develop and test new hypotheses. In other words, the answers provided by the machine learning model lead to new and different questions.
Finding Out a Value with a Supervised Model
An example of a supervised model is when a marketer has a value or attribute they’re trying to find, and they provide a model with historical data that has a broad range of calculations for the value. To try to find out a customer’s lifetime value, for example, the model will start with the known value for a big enough sample size of existing customers to predict lifetime value for a new customer or prospect.
Building the model is easy enough; provide historical data, point out the variable of interest, and let the model work its magic to make a prediction. In executing the model – putting it into production – a few things have to happen. First, testing will determine the model’s accuracy. By feeding the model historical data that has been held in reserve, a comparison can be made between the predicted CLV and what a marketer already knows about lifetime value – the better the correlation, the better the model.
Another more obvious kind of model execution is using the model to control rules about what to do with a group of customers, much like any other set of campaign-based rules. With CLV, for instance, a marketer may want to run campaigns for high, medium and low lifetime values. Setting that up as part of an AML algorithm is another task taken off the plate of marketers.
In this type of situation, the integrations made available in rg1 mean that models automatically set up and managed inside rg1 are also available for other applications that may care about the model. A call center, for instance, could make a call through the web server and add a pop-up window showing agents the CLV of a caller. Or it could route calls to different agents based on a customer’s CLV.
AML Makes Marketers Smarter
Both unsupervised and supervised models shorten a marketer’s to-do list for turning raw data into customer insight, but that is not to say that automated machine learning replaces or minimizes the value that marketers bring to the table. In fact, AML can augment a marketer’s capabilities just as a chatbot can act as the first-line to allow call center agents to focus on the more complex questions. Automating time-consuming processes with AML frees up time for marketer that is better spent focusing on their expertise – the customer. Also, by unlocking previously unattainable insights into customer data, AML allows marketers to ask better questions and make better decisions. The more they know about a customer or segment, the more they can design customer experiences that are consistently relevant throughout a customer journey.