Text analytics is a powerful tool that so far has been used most extensively for analyzing customer-related information. “The tools are business-friendly and the content—once parsed—is easy to process,” says Leslie Owens, VP of research at Forrester. “The leading vendors have found the most adoption in social listening and are well established, with robust cloud-based solutions in this area.” Other applications for text analytics are also growing, and predictions for the worldwide market indicate that it will increase from under $2 billion at present to nearly $5 billion by 2019.
Developing strategies
Oberweis Dairy evolved from a family dairy business based in Illinois. Founded in the early 1900s, Oberweis still offers home delivery of milk and ice cream in the Midwest and in several other states throughout the country. In addition, the company operates about 50 retail dairy stores that sell its products, and recently opened several restaurants that are co-located with its stores.
To provide detailed insights into its operations, Oberweis decided about five years ago to deploy a business intelligence (BI) solution. Prior to that, the company had used consulting services for some of its analytics support, but wanted to bring that function completely in house. After considering a variety of analytic tools, Oberweis selected the SAS analytics platform. One of the early uses of SAS was to provide analyses that would help develop more effective marketing strategies.
In addition to carrying out numerous quantitative analyses to support that goal, Oberweis also began using SAS Text Miner to analyze notes taken by the customer service representatives in its call center. “Our home delivery business had begun to slow down,” says Bruce Bedford, VP of marketing analytics and consumer insight at Oberweis, “and we did not have a handle on the attrition rate by segment.” Oberweis decided to explore the issue by analyzing the content in its customer relationship management (CRM) system.
Using Text Miner helped identify one particular promotional strategy—six months of free delivery—that was correlated with a higher attrition rate. “When the promotional phase was over, customers called in to object to having a new charge on their bills,” recalls Bedford, “and sometimes had even forgotten that the offer was temporary.” Information from the CRM system revealed details about the customers’ dissatisfaction. As a result, Oberweis modified the offer to be in the form of a discount on delivery. “Even though this change was very slight, it was effective,” Bedford says. “The customers no longer experienced such a noticeable discontinuity in the nature of their service, and retention was improved.”
The CSR resource
The ongoing process of text analytics also provides some serendipitous insights. “We use customer complaints as an indicator for how we can improve our products and services,” Bedford says. “At one point, customers were complaining about the milk in a variety of ways, such as ‘cream on top,’ or ‘milk looks funny.’ SAS helped us group the comments despite the different terminology that the customers were using.” The graphics that present a visualization of the text data generated a new category and a spike. “We discovered that one batch of milk had been produced that was not homogenized. Once an adjustment in a machine setting was made, the problem was solved,” explains Bedford.
Over time, it became evident to Oberweis that the customer service representatives were a resource that could be used for more than customer service. “We realized that the call center could also be the primary point of contact when products were launched,” says Bedford. Oberweis identified a goal for each representative that related to cross-selling or upselling, and tied it to the employee’s bonus program. “The reps were able to use the CRM system to document the products sold and their value,” he continues. “This approach was very useful because the reps did not have to change the way they worked, yet we could capture the new information. The information is fed into the SAS repository for analysis.”
An interesting bonus was that the customer service representatives, now renamed marketing representatives, have become more engaged and are serving as the vanguard for new product introductions. “The outcome has been very positive,” Bedford says, “and we were able to create this change without having to re-engineer our proprietary database to accommodate the new information. Although we are a small company, SAS has really helped us function as a large one by allowing us to use powerful analytics to leverage our information more effectively.”
Analytics, applications and adjustments
The use of text analytics for customer-related information has been so dominant that its use for applications tends to be overshadowed, but those also are continuing to grow. Another relatively common use of text analytics is for fraud detection. “One European government was able to reduce tax-related fraud by 98 percent simply by noting the contents of certain fields on a form,” says Fiona McNeill, global product marketing manager at SAS. “Others are building knowledgebases by consolidating service notes into a searchable format and analyzing the contents.”
Root cause analysis relies on text analytics to understand scenarios in which manufactured parts are failing. “The text related to insurance claims and servicing logs associated with manufacturing can help pinpoint the underlying causes of failure in a way that quantitative data alone cannot,” McNeill adds.
It is important to consider in advance what an organization will do with the information. “One insurance company analyzed incoming queries and realized that its website was not clear enough,” McNeill says. “The company then modified its site and substantially reduced the number of queries. However, if an organization is not prepared to make adjustments, then they might want to reconsider the role of their analysis.”
Fine-tuning for customer retention
The Medicare & Retirement division of UnitedHealthcare (UHC) serves nearly one in five Medicare beneficiaries. It offers a full range of products and services, and is the country’s largest provider of Medicare Advantage plans, Medicare Part D prescription drug plans and Medicare Supplement insurance. Many of its products carry the AARP name.
To determine whether UnitedHealthcare is meeting the needs of its customers, the company has used a variety of software tools to measure its success. “We use a mix of quantitative and qualitative data,” says Tom Allenburg, VP of data analytics and marketing solutions for UnitedHealthcare Medicare & Retirement. “However, we were underutilizing the information that is captured in free form text during call center interactions.” The information was being stored, but not analyzed.
UHC selected Mu Sigma to support a number of analytics initiatives. “Leveraging text was a new venture for us,” Allenburg says, “and we wanted to partner with an organization that had both the professional expertise and a set of powerful analytic tools.” Mu Sigma’s text mining engine, muText, analyzes text in structured and semi-structured content to evaluate customer comments.
One of UHC’s priorities was to identify customers who were likely to let their policies lapse. “The more we understand customers’ feelings and experiences, the better we can develop ways of retaining them as subscribers,” Allenburg adds. “If someone felt negatively about an interaction with our customer service area, we wanted to be able to find out why.” UHC had used both quantitative and qualitative data to derive a predictive model about policy lapses, but wanted to incorporate the call center notes to improve its accuracy.
In fact, UHC was able to show that bringing in the call center log notes provided lift, and improved its ability to identify those likely to lapse. “A model won’t lower the lapse rate, but since we are now able to better target the customers who are at risk, we can develop plans to reach out to those customers,” Allenburg explains. In addition, the analysis allows training activities with the call center representatives so they can be alert for certain key phrases. “If we know that certain phrases are associated with lapse, we can improve the customers’ interactions upfront and improve retention,” he adds.
Better communication
Over time, additional applications for text analytics have been developed. UHC is using voice recognition to convert audio to text and then analyze it. “This process allows us to ensure that certain statements required for compliance purposes are being made during the calls,” Allenburg says. In addition, UHC is using text analytics to improve interactions with customer service representatives during open selling season. “Some of the newer reps were not clearly conveying a sense of urgency during these limited time periods, to ensure that a prospective customer would make a decision by the deadline,” he notes. “We were able to analyze the conversations and make positive suggestions for better communication.”
The use of text analytics allows a much greater degree of personalization, which is a strong trend in marketing. “Traditional demographics and transactional data only get you so far,” Allenburg says. “Two people who are demographically similar might have very different reactions to a product, a service or an interaction. It all comes down to the individual, and personalizing our response to meet their needs.”
The desire to know what customers are saying in near real time is one of the most frequently mentioned drivers for implementing text analytics, whether the comments are coming from a CRM system or on the Internet through various social media channels. “Many large companies have years of focus group information that has not been analyzed,” says Mike Feldner, regional head of client services at Mu Sigma. “More broadly, everyone is contending with the explosion of data, and trying to decide what the right tools are. Since companies can now answer questions that they could not in the past, they are having to change their mindset about what is possible.”
Mu Sigma’s text mining application is integrated into its Decision Sciences workbench, muRx, which provides a guided method for data exploration, modeling, analysis and reporting. “We offer both a platform and professional services,” Feldner says, “but our services and products are transparent, so the clients know exactly how their data is being transformed and modeled.” The outcome of the decision process is a set of conclusions and recommendations on which the company can act.
What’s next?
Several trends can be anticipated for text analytics over the next few years. “Specific text analytics capabilities, such as sentiment analysis, will mature,” says Leslie Owens of Forrester. “Right now, the analytical models are at varying levels of maturity, depending on the uniqueness of the topics or the source material itself.” In addition, better integration with traditional business information systems will bring a more comprehensive understanding of the context of the comment, such as the store the customer visited or what they bought.
“All the characteristics of an interaction between a customer and a company need to be put together to form a complete picture that supports the right action to be taken,” Owens adds. Finally, big data (see sidebar following or on page 13, KMWorld, Vol 23, Issue 7) will take on an increasingly important role as the burgeoning volume of data puts greater pressure on creating effective automatesystems for interpreting unstructured information.
BIG DATA and TEXT ANALYTICS
Big data is unquestionably a driver for the increased use of text analytics. “In particular, big customer data is more available than ever before,” says Brian Koma, VP of research, enterprise feedback management practice leader at Verint Systems. Verint provides solutions for capturing and analyzing customer data, security intelligence and fraud/risk/compliance information. The volume of unstructured information in social media customer records, voice recordings, chat sessions, e-mail interactions and open-ended comments in surveys is growing rapidly. “Most of this information is underutilized or in some cases, not analyzed at all,” Koma says.
Yet most organizations are still listening to customer data in silos, because different channels evolved at different times, and fully integrating the data poses many problems across an enterprise. Being able to answer the “why” behind the “what” is where text analytics shines. “Highly sophisticated companies are putting together the structured and unstructured information to track, profile and segment customers,” Koma says, “but the integration process remains a challenge, both organizationally and technically.” However, by overcoming those barriers, companies have the best shot at proactively detecting patterns and fully understanding their customers.