Making Sense of Unstructured Customer Feedback (leveraging Big Data) and Linking to Processes
In part one of this series, we discussed the complexity of customer processes and interactions that result from the complexities of internal legacy systems and processes. We discussed the importance of mapping attributes among different systems in order to develop consistent metrics across a process. In this segment, we will review the role of unstructured data and text analytics and describe how to make sense of Big Data in the context of customer processes. [ or “this” meaning customer analytics?]
Making Sense of Customer Feedback – A Big Data Approach
Beyond the challenge of integrating multiple system lies the equally challenging task of making sense of feedback sources. The problem is similar to the mapping problem on the transaction side of the equation. Customers can provide feedback through any number of avenues – from simple feedback on telephone surveys (press 1 if you were satisfied, press 2 if you were not satisfied) to online support ticket submissions through a web site, call center agents, email feedback, discussion boards and forums, social media sites. In many of these channels, the feedback is unstructured text and a process of text mining or text analytics needs to be applied to understand it.
Much of this information exhibits characteristics of Big Data – there is lots of it (many times coming from social media), it is varied in nature (different sources, formats and structures) and it changes quickly. This is the traditional volume, variety and velocity of Big Data. Text analytics approaches are used to make sense of large amounts of unstructured information. These approaches mine the information for the ideas that are important to the organization and connect those to positive and negative sentiments about various aspects of the product, services and overall customer experience. The volume is too great for humans to do more than sample it – machines are needed to classify the contents based on keywords, terms, concepts, product names, and sentiment.
However, with more detailed analysis, additional insights can be derived from the text. because italso provides hints about the context and processes involved in customer interactons. Comments and complaints can be classified according to the nature of the issue. Take the following comments from three separate customers and their experiences:
“Waited on hold too long, had to leave a message and was never called back”
“Rep was rude and curt. Still can’t track my application”
“My bill is very confusing and disputed charges were not removed”
Through text mining, these might be classified according to internal processes. The first says something about a call center, the second about the underwriting process, and the third about the billing department. By linking an internal process metric (call volume, loans processed) to customer feedback (number of complaints related to the process) the organization can identify where the problem is, perform an intervention, and track the impact of the intervention on these metrics.
Likelihood to Recommend and Net Promoter Scores
Some financial firms are looking at Likelihood to Recommend metrics for understanding customer behaviors and preferences and “netting it out” with the “Net Promoter” score which is the difference between those who would recommend and those who would not recommend (promoters and detractors). Of course, this metric can hide a number of factors and is typically part of a larger data gathering framework. Surveys with either open-ended or structured questions can further tease out the important operational issues that impact whether someone would recommend or not recommend a company to their friends and colleagues.
Various domains can be assessed, such as product or service attributes (specific features), price considerations (clear fee structures), brand (well known, trusted), or service experience (support personnel knowledge), and comments and feedback can then be classified in terms of whether factors in these areas were positive or negative contributors. These domains essentially become categories in a “domain model,” with terms comprising a taxonomy of descriptors and factors. Text analytics tools can then use that taxonomy to classify voice of the customer data sources to make sense of the customer experience.
Customer analytics programs require organizations to understand customer behaviors and link those behaviors to internal systems that are serving the customer throughout their lifecycle. Customers need to interact throughout a range of states based on the where they are in their process. Many organizations consider the “journey” as customers travel through various departments and processes. The customer journey:
- Traverses multiple channels and touch points
- Interacts with every part of the business (throughout the product of service lifecycle)
- Is supported by multiple departments
- Realized or transacted across many systems and applications
- Governed or managed through various processes and organizational structures
- Leverages models of the customer to varying degrees (attributes, characteristics, preferences)
- Is more than sales and support and depends on all the other parts of the organization
Each interaction leaves the customer with an impression, and each interaction is an opportunity to strengthen or weaken the relationship – to sell new products or services or solve problems, or fail in the attempt to do so. Harvesting data from across processes can be linked to customer feedback which can provide a score card around process health. These feedback mechanisms help organizations allocate resources and measure the impact of their customer experience programs.