The rise of social media and big data means that text mining presents an enormous opportunity for companies with the vision and resources. Insurance Networking News asked Karthik Balakrishnan, SVP of fraud solutions and analytics at Verisk Health, a division of Verisk Analytics, and Phil Hatfield, VP of operations at the ISO Innovative Analytics unit of Verisk Analytics, to explain the technology and how insurers are using it to better understand their customers, competitors and carriers.
INN: In what operational areas are insurers making best use of text mining? In what areas are they deficient?
KB: While the specific areas will be different for each insurer, the proverbial "low hanging fruit" seems to be the text mining of claims adjuster notes to identify fraud and missed subrogation opportunities. Some carriers are using text mining to identify "true" causes of loss from adjuster notes-for example, water loss caused by burst pipes.
Less common applications include text mining call center logs, competitor websites, underwriting notes, loss control notes, premium auditor notes, textual feedback from customer satisfaction surveys and transcripts from market research. If appropriately leveraged, those sources can help support decision-making in marketing, customer service, product innovation, underwriting optimization, claims and loss control.
INN: How can insurers glean more information from social media?
KB: Social media such as blogs, chats, tweets and social networks are rich sources of named entities, interrelationships and opinions. A text-mining system with the ability to extract named entities and relationships, along with sentiment analysis (automatically deciphering the writer's tone-positive, negative or neutral), is thus a foundational need. Increasingly, tools and service companies are emerging to help analyze such social information.
INN: What do burgeoning Big data architectures mean for the use of analytics?
PH: Big data is the new buzz term that denotes the convergence of several trends: the continuing decrease in the price of storage and processing hardware, the emergence of effective distributed computing algorithms and the continuing torrent of information that can now be saved and analyzed.
Hadoop, in particular, employs a high-reliability, high-access distributed file system along with a parallel data processing technique called MapReduce to deliver extremely high performance. So with Hadoop, companies can efficiently perform custom analytics on immense volumes of their data, as is the case with Web companies such as Yahoo and Facebook.
However, as more information is stored, the amount of data relevant to a particular problem will increase-but not nearly as fast as the amount of irrelevant data. That tends to make analytics and automated methods of sifting through voluminous data more important than ever.
INN: What is the link between text mining and developing new products and services?
PH: The most important step in new-product development is to first identify the new product or service idea. With the prevalence of e-mail and Web-based communications, systematically text mining those rich data sources that often contain customer feelings, perceptions, opinions and complaints, is very likely to yield different insights for business action than traditional market research.
INN: What investments should insurers make?
PH: Insurers should consider investing in three elements: text-mining capabilities, human talent and the infrastructure to support it. Text-mining systems have evolved in recent years, and good options exist both in the open-source world and among vendors. Depending on the tools chosen and the nature and extent of textual assets to be analyzed, appropriate computing and storage infrastructure will be necessary. Most text-mining solution vendors will consult with the company to provide necessary infrastructure requirements and benchmarks.
As in most areas of business, technology investments are relatively easy to make when compared to the required investments in human capital. The right technology, even if harder to find and potentially costly, can at least be depreciated. On the other hand, human capital can be expensive and difficult to evaluate before deployment. But good talent is the essential ingredient for successful analytic efforts.
KB: The right text-mining practitioners need to have a firm grounding in machine learning, statistics, operations research and related technical fields to effectively work the text-mining methodologies. They should also have a clear understanding of the business. Only that combination can result in proficient analytics to solve real business issues while providing a competitive advantage.
