Big Data at United Healthcare
United Healthcare, like many large organizations pursuing big data, has been focused on structured data analysis for many years, and even advertises its analytical capabilities to consumers (“Health in Numbers”). Now, however, it is focusing its analytical attention on unstructured data - in particular, the data on customer attitudes that is sitting in recorded voice files from customer calls to call centers. The level of customer satisfaction is increasingly important to health insurers, because consumers increasingly have choice about what health plans they belong to. Service levels are also being monitored by state and federal government groups, and published by organizations such as Consumer Reports.
In the past, that valuable data from calls couldn’t be analyzed. Now, however, United is turning the voice data into text, and then analyzing it with “natural language processing” software. The analysis process can identify—though it’s not easy, given the vagaries of the English language - customers who use terms suggesting strong dissatisfaction. A United representative can then make some sort of intervention - perhaps a call exploring the nature of the problem. The decision being made is the same as in the past—how to identify a dissatisfied customer - but the tools are different.
To analyze the text data, United Healthcare uses a variety of tools. The data initially goes into a “data lake” using Hadoop and NoSQL storage, so the data doesn’t have to be normalized. The natural language processing - primarily a “singular value decomposition”, or modified word count - takes place on a database appliance. A variety of other technologies are being surveyed and tested to assess their fit within the “future state architecture. United also makes use of interfaces between its statistical analysis tools and Hadoop.
The work to put the customer satisfaction data, along with many other sources of customer data, into a customer data warehouse and analyze it is being led by Mark Pitts, who is based in the Finance organization. However, several other functions and units of United, including its Optum business specializing in selling data and related services to healthcare organizations, are participating. Pitt’s team includes both conventional quantitative analysts and data scientists with strong IT and data management skills.
Source: Big Data in Big Companies, Thomas H. Davenport and Jill Dyché, May 2013 (Go to Suggested Readings to view full article)