The Role of Text Analytics in Managing Customer Loyalty and Churn

by   |   April 8, 2016 5:30 am   |   1 Comments

Mark Schmelzenbach, Chief Technical Officer, Attensity

Mark Schmelzenbach, Chief Technical Officer, Attensity

Businesses used to be able to spot unhappy customers by the looks on their faces and the tone of their voices. When doing business was like living in a small town, it was easy to notice quickly if they stopped coming to you – especially if you could see that they had started visiting the business across the street! You might also hear from friends or family that the reason somebody isn’t coming around any more is because they are traveling or ill.

Doing business in the 21st century rarely resembles having a shop in a small town. Today, the first obvious sign of customer unhappiness may also be the last one: when you lose their business, possibly forever – if you even notice one departure in a sea of interactions.

As customer intelligence systems have come into existence, there’s a temptation to blame the customer loss on the last negative thing that happened to them – a broken product that slipped through quality controls, a poor service interaction, perhaps even a change that you suspected would make some customers unhappy. In reality, it is more typical for customers to become unhappy and leave only after a series of negative events. Blaming the last bad event for driving away a customer is like blaming the fire alarm for starting the fire.

The flip side of churn is customer loyalty, which shows up as customers who buy more and buy more often, on their own, via up-selling, cross-selling, and other offers and incentives. Loyalty is good for any business, but the higher your customer-acquisition costs are, the more churn hurts and the more loyalty helps.

The new rules for assessing customer loyalty, value, and propensity to churn call for probabilistic models that become increasingly accurate by incorporating more data from the customer journey. Chances are, you are capturing and storing much of that data already, but not yet using it. Customer insights come in two broad categories: what they do and what they say; actions and words.

Many of the things that customers do are revealed as structured data.

  • Customers visit web sites, which record click-by-click details of their actions.
  • They view and click on advertising.
  • They put items into shopping carts and buy some of them.
  • They email, call, or otherwise engage with support resources.
  • They post and “like” messages in social media.
  • They take surveys and provide feedback scores.
  • They pay bills, on time or not – or fail to pay them.

 

Related Stories

No Excuses for Poor Customer Service in the Big Data Era.
Read the story »

Use Data to Understand Customer Attitudes and Drive Growth.
Read the story »

A Customer Data Taxonomy.
Read the story »

Overcoming Three Challenges to Onboarding Customer Data.
Read the story »

Although some of these actions may be ambiguous, they produce structured data the biggest challenge of which is the volume that is produced – billions of web server clicks, thousands or even millions of potentially relevant posts in social media.

Some of those actions include words that customers say when they talk about your company, brands, products, and services, and competitors. Although actions really do “speak louder than words,” it’s also clear that customers’ words matter.

A customer who posts, “If they keep me on hold for five more minutes, I’m switching banks,” hasn’t acted yet, but those words clearly are important. There’s no assurance that the words mean that she really will churn, but this is one more piece of evidence from the customer journey, a chain of events that can lead to loyalty or leaving.

Words can also answer the “why” behind negative feedback scores. If a restaurant’s customers rank breakfast poorly, the only way to identify the underlying problem is to read what they wrote. Language as simple as “the food was cold” tells you exactly what needs to be fixed. Open-ended survey questions, unsolicited reviews and comments, call-center summaries – the words in these customer interactions are the doorway to discovering problems and praises that you might never have put on the structured part of a survey.

All of this structured and unstructured data about the customer experience live in a wide variety of information systems on both sides of the corporate firewall – web servers, CRM applications, billing and finance, support communities, social media, blogs, and so forth. Measuring loyalty and predicting churn become increasingly accurate with each data source that is added to the mix. For example, a customer who appears to be dependably loyal because she pays her bills on time and never visits the support web site might also be complaining loudly in social media. Conversely, there are plenty of people who never post in public, whose unhappiness might surface in data such as repeated call center interactions that take a long time to resolve.

The point is that if you are blind to part of the customer experience, your ability to predict churn and manage loyalty will suffer. Today’s customer intelligence systems need flexibility above all – the power to rapidly and inexpensively add new data sources and incorporate them into the kind of mathematical models that big data analytics enable.

Unless you care only about the top line – revenue, there is one more important data point that needs to be integrated when managing loyalty and churn: customer lifetime value (CLTV). Without a measure – ideally, a prediction – of CLTV, there is no way to prioritize which customers you should work hardest to keep or which ones are worth dropping because they are costing you more than the revenue they produce. Once again, flexibility – a system that can incorporate data from all of the customer actions and words along with financial data – is critical.

Flexibility doesn’t apply just to getting data into an integrated application. You also need to operationalize the results of your analytics. Customer support, marketing, sales, and finance all need immediate access to information about individual and aggregated customers so your business can become highly responsive to individuals, segments, and the whole market.

Today’s business calls for continuous optimization for loyalty, churn, and profits. You need to know what your customers are doing, what they are saying and what they value. Flexibility is the key to rapidly and inexpensively achieving the accuracy that comes from incorporating data from the entire customer journey into loyalty and churn modeling. Flexibility with the results of analytics allows you to operationalize the results, pushing the information your team needs into the systems that they use to interact with customers, make strategy, and measure success.

Mark Schmelzenbach is Chief Technical Officer at Attensity. He brings to this role over a decade and a half of experience in Natural Language Processing in the commercial arena. Mr. Schmelzenbach is a key contributor to Attensity’s semantic extraction technologies and leads the Attensity engineering teams in the development of innovative research and applications that solve real world problems for both enterprise and government clients. He holds a Bachelor of Science in Computer Science from Northwest Nazarene University and a Master of Engineering in Computer Science from the University of Utah.

Subscribe to Data Informed for the latest information and news on big data and analytics for the enterprise.




eBook: crack the unstructured data code with deep learning




Tags: , , , , , , ,

One Comment

  1. Posted June 12, 2016 at 12:52 pm | Permalink

    Dude works for inContact and did at the time you published this article. InContact bought Attensity’s text analytics in February, a few months before you published.

Post a Comment

Your email is never published nor shared. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>