With all the enthusiasm for machine learning, which we happen to share, it is important to remember this: machine learning isn’t magic, it’s software. Like any application, machine learning can only be as good as we enable it to be.Now there is undoubtedly a lot to learn from very detailed data. At the same time, just because it is possible to feed reams of data about every single webpage into a system, doesn’t mean that that approach will actually yield valuable insight into customer journeys. When you pour detailed data into such a system, the system will necessarily react and produce results based on that data. This suggests that what you choose to put into the system matters.
Three key elements for understanding customer journeys
To begin to understand the importance of being selective about data, consider in the most basic terms the three main elements of a complete analytics program built for understanding customer journeys:
- Data preparation where companies acquire interaction data and incorporate it into a predictive model that combines segmentation and interaction data.
- The running of the analytics models, powered by machine learning, where rules are developed to prioritize customers.
- People such as business analysts who look at the insights being offered by machine learning, identify appropriate programs and offers, and ultimately improve the effectiveness of the model.
There are certainly cases where including everything and the kitchen sink is useful (consider epidemiological data). Customer journeys are not, however, one of those cases. In fact, too much content is prohibitive. That third item on the list tells us a bit about why. People not only have to interpret these results, but must contribute to future decisions about adding more data (and that person might be you).
The purpose of machine learning is to create rules based on data. If allowed, machine learning will react to every bit of minutia thrown at it. A bank’s website may present 30 distinct paths a customer could follow to pay a bill. Each journey might be slightly different and further affected by variables—e.g., a credit card ad that one customer follows before paying their bill, but another ignores. Machine learning can take all those factors into account, but do they really all matter?
Proactive noise reduction
Too much content can lead the system to build rules that the business simply can’t react to. That is why proactive noise reduction during data prep helps machine learning scale back to a meaningful and practical set of rules.
We’ve long heard about developing a “360 degree view” of the customer. This picture is built from many data sources—website data, mobile data, transactions, processes that lead to transactions, interaction with automated call systems, and even data about a service agent’s tone of voice. From a data perspective, as you get close to that complete picture, the data is so complex and voluminous that it exceeds the capacity for people to handle it in a meaningful and timely manner.
This is not a shortcoming of people. It is recognition of the vital role of people and their creative insights in the analytics of customer journeys. We'll touch on this again in an upcoming blog.
It also points to where machine learning comes into play. Previously we might have looked at how many interactions a customer has per channel. Machine learning lets us look at much more: How many interactions were expected for a given customer? Were results above or below that number? Does reacting to that difference matter? How important is each channel for a particular customer?
There are any number of questions we might ask about one customer, but what about 30 million customers, each with their own pattern of interaction with channels? It’s too much data. This is the very nature of why machine learning was introduced. Machine learning further helps us tease out which variables are important and can be used to create rules around churn and other journeys.
Collect and use the right data
Getting maximum value from the data means collecting and using the right data for the challenge. I wrote about dark data three years ago and the topic has been batted around the industry for at least a decade. There is data that is not being collected and then there is data captured and stored, but which delivers no value. Machine learning gives us the capacity to consider all the data available.
Machine learning is only as good as the data provided to the system. Products like Clickfox can help you determine the data that is most relevant to the customer journey, shape the data in ways that make it digestible to the machine learning system, and then use the results of machine learning to inform future conversations about ways to use additional data to gain an even deeper understanding of your customers and how you can best serve them.