Making Sense of Customer Feedback
As customers demand greater transparency and data immediacy from their services providers, the ability for water utilities to collect unstructured text data is growing. With modern, digital customer engagement interfaces such as web portals, mobile applications, and social networks, utilities now have a window into more nuanced interests, demands, concerns, and satisfactions expressed by their customers. But with the evolution of systems designed to capture and convey textual information comes a significant challenge: making sense of large volumes of unstructured data.
Unstructured data includes strings of text supplied by customers through open ended self-service forms and other interfaces. Spelling, punctuation, idioms, and grammatical syntax is inconsistent and thus very difficult for computers to parse and interpret. And as the volume of such potentially insightful information grows, it quickly becomes impractical for humans to review each request, note, post, tweet, and text to identify trends as it pertains to service quality and customer satisfaction.
Natural Language Processing
Fortunately, an emerging field in computer science is gaining momentum to help address these challenges. Under the broader rubric of machine learning, Natural Language Processing (NLP) is the ability of computers to understand human language. Recent computing advances are allowing computers to approach, and in some cases exceed, human-level language comprehension accuracy. Since computers are doing the interpretation, it’s not surprising that they are much, much faster than humans, thus bringing the ability to parse unstructured customer text feedback at scale. Utilities have a growing treasure trove of information from their users, and the ability to quickly and easily understand and organize customer feedback will have a profound influence on the future of customer service organizations.
The Investigation
WaterSmart is at the forefront of digital customer engagement and through our portal interfaces we have been able to collect customer feedback from our water utility partners over several years. This gives us unprecedented access to a high volume of unstructured customer feedback, and the ideal environment to perform tests on NLP technology.
Consequently, we recently undertook a project to mine this data to see if we could identify insights into customer engagement that would prove useful to our partners and to the industry as a whole. The specific data set we used for this investigation included 6 years of data from 69 separate surveys conducted across 36 utilities partners. There are two broad areas of investigation that we looked at in order to field insights: Sentiment Analysis and Topic Modeling.
“Sentiment analysis and topic modeling allow utilities to better understand customer needs“
Sentiment Analysis is a technique to search text to identify subjective meanings such as emotions and opinions. We are looking at survey data to determine how customers feel about their utility and the services that are provided. Are they satisfied, frustrated, indifferent? Insights on these topics can help utilities redirect customer service activities to areas of most need, but also develop trend data to benchmark performance over time.
Topic Modeling seeks to uncover latent topics embedded in large volumes of text. Methods analyze word counts and their occurrence in relation to other words to capture information on relative word importance. Insights from Topic Modeling can give utilities a better understanding of interests that customers have around billing, leaks, information access, or other common areas of need.
Response Categories
After conducting the surveys and aggregating and tagging the data, we were able to break down the customer responses into nine high-level categories. The results of the feedback is listed in the chart below:
Combining the response categories with topic and sentiment analyses, we were able to derive several key insights and practical take aways for water utilities and their customer service organizations in particular.
Key Takeaways
The survey results yielded several interesting themes that may not be intuitive to most water utilities:
- Households are hungry for water information and look to their utility for expertise: Surprisingly, many water customers are eager to have access to more information on their water services and rely on their provider as a source of important and useful data. This insight provides utilities with an opportunity to proactively communicate with their customers to improve satisfaction, reduce support costs, and build political capital for future system investments.
- There is significant unmet demand for utility programs such as appliance rebates, audits, and community events: Customers are often unaware of many programs that utilities offer to help end-users better manage their water spend and protect their property from costly water damage. The results of our analysis indicate that customers are hungry for these programs and are likely to take advantage of them if awareness were higher and access easier.
- Sentiment toward water utilities, in aggregate, skews negative but is mostly neutral: While generally good news, it does indicate that there is room for improvement in terms of how customers view their utility. By addressing the first two takeaways above, water providers are likely to improve overall customer satisfaction and thus improve the skew of sentiment more positive
Conclusion
WaterSmart Software is exploring new ways to help water utilities make use of all their data to reduce costs, protect revenue, and increase customer satisfaction. Water utilities are just beginning to explore these powerful techniques that have already been adopted in other industries. To help utilities better understand and leverage these new technologies, we have recently published a white paper on how utilities can improve stakeholder engagement with Natural Language Processing. The paper goes into more detail on the techniques outlined above as well as the investigation methodology and results. We encourage you to download and share the document with your peers and consider how it might aid your future customer engagement efforts.We believe that Natural Language Processing tools, particularly the techniques explored in this article, will fundamentally change how water utilities serve and engage with their customers over the next five years. What will emerge is a more responsive industry better able to work with its customers, regulators, and other stakeholders to continue to address the significant financial and operational challenges on the horizon.
