Limburg Water Supply Company

How WML gets more insight into repeat customer contact

"It started with our Willy Wortel council," begins Myrko Webers, head of customer service at WML. "That's an internal innovation platform of WML to develop new initiatives in the form of pilots. We have a huge amount of data and wanted to get more value out of it." The company engaged Inergy to provide insight into the potential of the data and to develop a concrete use case.

"My colleague Hedwig van Berlo is a strategic policy advisor and coordinator of the WML Innovation Lab / the Willy Wortel council," Myrko continued. "She was looking for a big data case within the customer domain because we hadn't done much with it within WML until now. We were keen to discover what the possibilities were and also learn how to approach something like this."

Advanced analytics

"WML supplies water throughout the province of Limburg," explains Melvin Agten, Data Scientist at Inergy. "The continuity of supply and the quality of the tap water are paramount. In addition, the company wants to keep its customer satisfaction as high as possible. During this project we applied advanced analytics to all customer contact data. WML asked for our expertise because it did not have any knowledge and experience of Data Science projects itself."

"What we liked was that Melvin was in our customer service department for whole days in the beginning. He literally sat between the customer service employees and constantly asked all kinds of questions. In this way he immediately understood our primary process, created support and at the same time we could learn a lot from him," says Myrko.

(Un)structured data

WML has several channels for customer contact: telephone, myWML, email, letter and social media. Repeat traffic occurs when a customer regularly contacts WML about the same issue. "To identify which customers cause repeat traffic, I developed a model. WML's CRM system stores all customer data from past years. This includes the date, time and subject of when someone called or emailed, for example," Melvin explains. "We call that the structured data." "Per customer contact, we also make manual notes in the system," Myrko explains. "A few sentences containing, for example, the customer's reason for calling. We call that the unstructured data." 

Connections in the data

Melvin: "In the analysis I made a distinction between these two types of data. The unstructured data are a nice addition to the structured data. They often confirm the findings from the structured data and supplement them with new insights."

"Melvin looked for us at the connections in the data that say something about where the largest amount of repeat traffic is and why," adds Myrko. "The results of that were delivered in a Power BI dashboard. Very valuable insight for us, because with the help of the interactive dashboard and the insights from the unstructured data we now have a good idea of where (unwanted) repeat traffic is and we can work on reducing it. Moreover, we can continue to update the dashboard with the structured data so that we can monitor how things develop over time and whether we will see the effect of improvement actions."

Sentiment Analysis

Of interest is the sentiment analysis applied to the unstructured data. This was used to test, using models in Python, where negative and positive sentiment is in the processes. This revolves around the meaning and context of words and word combinations. "In total, I applied sentiment analysis to more than one million texts.

So those are the texts that were manually entered into the CRM system based on contact moments with the customer," Melvin continued. "To test reliability, I shared a few hundred classified texts with Myrko. The sentiment analysis algorithm did very well, but the results of many texts were classified as neutral. That has to do with the fact that they were written from employees' perspective and not from the customer's."

Text mining

In the case of complaints, text mining proved to be a valuable tool. This method looks for example for the originality of words and word combinations. "If there are complaints, they are registered with us in the CRM system," says Myrko. "We had a suspicion that not everyone was registering complaints in the same way. That turned out to be the case. The insights from various text mining methods showed us that a large part of the customer contacts were not registered as complaints, while the results from the unstructured data told us otherwise.

Conversely, we also found that some customer contacts were registered as complaints where there was no word or word combination with complaint. Of course, we'll be delving deeper into this to see where things go wrong and why. In the future, this may lead to other agreements."

Future

"It is great to see how we as Inergy have been able to help WML in creating insight. Of course, we are happy to facilitate this company in the future on how it can use the insights from the data," Melvin concludes with satisfaction.

"And certainly not unimportant is that successfully completing this project also had to do with the excellent distribution of work. The beautiful end result is because we really worked together as a team. For example, as a Data Scientist I really needed the knowledge of WML. As a result, I knew exactly what the definitions were of the data in the CRM system. Smiling: "We both now look back on a successful project. A great start for perhaps other great projects that will help WML get value out of its data in the future."