Why Should You Replace Customer Segmentation with Machine Learning?

Machine learning algorithms are far more effective than segmenting customers to personalize the customer experience.

By Eva Michel, Oracle Insight

April 2018

Personalized marketing tops the list of technology-driven actions that executives worldwide are taking to meet their customers’ needs. In the US, 70% of retailers see the personalization of the customer experience as their top priority, according to eMarketer. One key prerequisite for personalization is to know your customer, and for that, companies typically turn to customer segmentation. 40% of global marketers say they practice customer segmentation to some extent, while 48% believe they are sophisticated practitioners.

Eva Michel

Eva Michel, Oracle Insight

So, how does it usually work? First, segments are defined. Second, customers are allocated to a segment for targeted inbound campaigns, such as cross-selling on incoming calls, or outbound campaigns, such as email blasts. Sound simple? The truth is that allocating customers to segments is hard work. Asked about the leading barriers to achieving their company’s personalization goals, the most frequently cited reason (42%) by UK and US marketers was the lack of resources such as time, people, and money.

But, to put it bluntly, customer segmentation is a waste of resources. Here’s why:

Segments are too broad or quickly get too complicated to handle. Gone are the days when customers were segmented solely on the basis of demographic data or prior product purchases. These segments are simply too broad and deliver only suboptimal conversion rates. Nowadays, companies are looking to include additional segmentation criteria captured along the customer journey, such as “opened holiday campaign email,” “clicked through to website,” “attended event,” and “looked at product x but did not buy.” The list of desired criteria is ever-growing and, if it is handled by humans, quickly becomes unmanageable.

Segments become quickly outdated. The drive to expand segmentation criteria to include interests and preferences results in another challenge. After you’ve bought your house and gotten your mortgage, you probably will no longer be interested in a mortgage for a very long time. Reallocating customers from one segment to another is often done manually or by batch, and it requires constant monitoring and shifting.

Segments cannot take the moment into account. Let’s say you are calling your bank to complain about the charges on your bank statement. On their screen, call center agents can see what customer segment you are in, and they can see the next, best cross-sell offer. But is now the right moment to sell you a new credit card? Maybe not. But what would be the best offer? Your predefined customer segment will be of no help to the agent.

So if customer segmentation doesn’t really help, what should you do?

The Capabilities of Machine Learning

Let’s take a step back. What are you trying to achieve? You are looking to provide a personalized experience so customers will come back repeatedly. You start off with defining personas representing a given customer segment. You brainstorm their expected challenges and then map out the perfect customer journey. So far, so good.

With machine learning, each customer is their own segment, defined by as many criteria as you like.”

However, you might quickly realize that creating the perfect customer journey may require substantial investments. Further, it may require a radical change in an area that might not be under your remit. What can you do if you do not have the mandate to radically transform the way your company does business?

The answer: start with what you have. Aim to maximize the impact of your current capabilities and resources—your available channels, agents, content, and offers. Do not spend ages defining segments and allocating your customers to them. Spend time understanding who will be best served through which channel with what message at what point in time. This is easy, because machine learning will do the hard work for you.

To be fair, you will still need to provide a couple of ground rules. Certain products will have legal requirements such as a minimum age or residency requirements. In addition, you might decide that you want to reserve costly channels for high-value customers only.

Other than that, the machine will learn with each customer interaction if the channel, the content, the offer, and the point in time were “right” for a given customer. Any data associated with that customer interaction will also be fed back for learning purposes. This data could include prior click behavior, age, gender, marital status, and income. Based on the feedback and data from a multitude of customer interactions, the machine can automatically build statistical models. These models will then predict what engagement mode is most likely to delight each individual customer.

The more learning data there is, the more accurate the models will be. You might have to be a bit patient for the models to produce predictions with high accuracy. To speed things up, you might consider using historical data to train the models. Alternatively, you might want to fall back on pretuned models from cloud providers. But in any case, there is no longer any need to allocate customers to segments. With machine learning, each customer is their own segment, defined by as many criteria as you like.

Customer segmentation has had its day. Machine learning algorithms are far more effective to personalize the customer experience.

Action Items

How Does Your Customer Experience Stack Up?
Get a copy of the 2018 Smarter CX Insights Report.

Special Report: Customer Experience

Oracle Insight

Photography by Graphic Compressor/