Brands should show customers that they hear critical feedback (and then do something with it to improve their product).

For example, when a customer's dissatisfaction is detected through survey responses or contact center interactions, the company should reach out. Proactively addressing complaints head on is one of the best opportunities to convert customers to evangelists. At scale, though, responding to unstructured feedback can swamp a company’s resources.

There needs to be an automated approach to prioritize who to reach out to, and what product feature to fix, first. One approach that we’ve deployed in the past is based on Net Promoter Score (NPS) drivers. This approach uses topic modeling and linear regression to quantify an issue’s impact on a customer’s likelihood to recommend the product. We discuss it more thoroughly in this case study.

A recent paper suggests another approach. Bluemel’s and Zaki’s (University of Cambridge) primary focus is comparing classical natural language processing with deep learning approaches for complaint prioritization (spoiler alert: the juice from deep learning isn’t worth the squeeze). However, embedded in the methodology is measure of the “response worthiness” (my quotes) of a complaint.

Afterall, in order to prioritize, there must be some quantitative value that drives the rankings. Bluemel and Zaki suggest a three-part score based on:

1. Polarization or sentiment, i.e., how positive or negative a comment is. The assumption is, the more negative a comment is, the more likely a customer is to churn and the more the comment should be prioritized.

2. Frequency, i.e., how often are the key words in the comment used in other comments. The more frequent a comment, the thinking goes, the higher the issue it addresses should be prioritized.

3. Subjectivity, i.e., the extent to which a comment is an opinion vs stated facts. Here the authors cite support for subjectivity being correlated with the urgency of support.

These seem like perfectly reasonable and well-supported measures of how to efficiently address the greatest number complaints—if that’s what you want to optimize for. It’s not clear, though, whether reducing complaints is the most efficient way to grow revenue. It certainly seems likely that the two are strongly correlated. But in the absence of data showing complaint reduction is a better driver of growth than increasing NPS, why not optimize the latter?

YOU MAY ALSO LIKE