Under Your Nose

Karen Pauli, research director for TowerGroup, estimates at least one third of the data collected by insurers is unstructured, and that number is growing exponentially, particularly in the claims area, where insurers often receive data through handheld devices directly from accident scenes. "Someone will pull out a phone and start taking pictures and record text messages or e-mail on the scene," she says. "Because people use mobile devices for just about everything, [unstructured data] is growing and growing."

Tony Smarrelli, vice president of national operations in claims at Infinity Property & Casualty, estimates at least 80 percent of Infinity's claims files are made up of unstructured data. "Depending on how complex the claim is and how long it remains open, it can be pages and pages of notes," he says.

RECENT PHENOMENON

As recently as four or five years ago, Pauli maintains, unstructured data usually referred to the notes claims adjusters took or the notes an underwriter made in the underwriting section of a carrier's admin system. "Generally speaking, it was notes or, in the cases of agents, sending in data in the comments sections," she says.

Most of those sections were limited in terms of how many characters could be entered on the form, so the volume wasn't great, continues Pauli. "Now, with modern claims and underwriting systems and advances in agency management systems, those text fields are unlimited," she says. "So, people can keep adding and adding and adding data."

New to the world of unstructured data are text messages, e-mail via mobile devices, and videos. "The number of videos coming in from handheld devices for claims is starting to become huge," says Pauli.

Unstructured data is core to what carriers have been doing for a while, Goldberg believes. "I don't think of it as a major initiative right now because it's been part and parcel to the insurance technology landscape for many years," he says. "When carriers started moving toward an electronic world, they began looking at how to cut down on storing all the paper."

USAGE

Goldberg indicates most carriers have some kind of content management plan in place, but what is lacking is the ability to make the document-input factor smarter and faster. "Getting better scanning and indexing technology will not only help get the data imaged but automatically make decisions about indexing," he says. "Carriers have some systems in place, but they need to make sure [data] is being archived correctly, so it is easier to retrieve."

Goldberg doesn't believe the lackluster economy has slowed progress in this area because most data projects are viewed as a source of major benefit. One such benefit is better sales information. "If I can look at everything about clients, I'll be able to sell to them better or make better decisions about whether they are valuable clients," says Goldberg. "That value proposition has less pressure during an economic crisis when you are trying to keep the business going.

"These are the kinds of projects that have other values, as well, such as increases in efficiency. The ability to reduce staff involved in manually indexing and scanning every document is important. If you can simplify that–let go of people or put them on higher-value tasks–it's valuable."

However, according to Pauli, analyzing unstructured data is not at the level it needs to be, and in contrast to Goldberg, she blames the economic problems insurers have faced for the last year as the principle reason. "Before the [financial] crisis, there were a lot more people talking about this," she says. "Now that the crisis has hit, people are focused on bottom-line issues. The scale of projects has come back a bit in terms of size and in delivery value. Everyone is trying to deliver projects with value rather than big-bang projects. No one wants to do big-bang projects right now."

Insurers definitely will get back to unstructured data, though, she predicts. "The more carriers start bringing analytics into claims operations, the more they are going to need to be involved with text analytics and text mining," says Pauli. "It's definitely a growth area."

STORAGE ISSUES

To move forward on data, anyone thinking of putting unstructured data to use, whether now or in the future, needs to have enough storage area. "You have to plan for storage; that's probably the biggest thing," says Pauli. "You need a plan to extend data marts and make certain whatever kind of architectural infrastructure is in place can scale. You don't want a really tight infrastructure that won't be able to accommodate the expansion of data capacity."

Some claims administration systems can deal with unstructured notes, points out Pauli. "They can turn them into data sets that can be analyzed," she says. "It's the really fancy analytics and mining around text messaging and such that probably remain in the early stages."

Pauli recommends turning to vendor solutions, though. "I would never recommend anybody do the analytics themselves," she says. "There are vendors that can do that very well. Consumer industries have been doing this for a lot longer, so insurance has become the next target area. I definitely would stick with the vendors' capabilities in that area."

Pauli also endorses the need for a plan. "The first step is to make certain your structured data is in good order," she says. "You don't want to go running willy-nilly for unstructured data. But once you have a modern claims administration system, you can start planning for the capabilities of bringing in unstructured data and utilizing it."

WHAT TO DO

The number-one thing for insurers to do with unstructured data is index it, advises Goldberg. "It's not necessarily about taking all electronic text and figuring out some way to search it or store it," he says. "It's about indexing things correctly–picking out keywords that are critical so you can pull them up later."

There are several ways to achieve this, explains Goldberg. One is to have some sort of software archiving system. "You need a document management system that allows you to store and retrieve things," he says.

The problem insurers face is getting the indices onto the document. One way is the old-fashioned, manual method. "Any time a scrap of paper comes in the door, someone manually scans it and determines what the indices should be," says Goldberg. "If an e-mail comes in, you are responsible for storing it and entering indices."

This requires some diligence on the part of the company, notes Goldberg, as multiple systems are integrated to the document management system.

There are tools that help companies make what Goldberg calls "intelligent guesses" about how to index a new piece of information, a document, an e-mail, or a scanned application form. There also are tools where a manager can build rules that help define how the automatic indices
are created.

Going through this effort has its rewards. "It's a huge benefit that didn't always exist," says Goldberg. "Insurance carriers can see a history of interactions–documents, e-mail, application forms–whether they are paper or electronic. It has helped carriers deal with audits and be better at sales. It enables them to understand the experience of clients so they can make better decisions in the future."

TEXT ANALYTICS

Pauli describes text analytics as a "fledgling area" for insurers dealing with unstructured data. "It is definitely needed," she says. "Anybody who has both eyeballs open in the claims, underwriting, or marketing areas knows this is going to happen."

The claims reps with Infinity gather and document thousands of files each month, which Smarrelli points out go untapped in terms of review and being able to identify important pieces of information.

"As our claims investigators investigate the files, they are required to document each incident, each claim number, each file with the information that's gathered," says Smarrelli. "They are able to search for things that might lead us to better investigate the claim or [find] if there might be fraud."

Infinity has been working with software provider SPSS for two years on text analytics, and the work has enabled the carrier to identify claims that might have the potential for fraud. "That sets off a set of business rules and some predictive analytics," says Smarrelli. "With those claims we see text analytics being able to elevate our investigations to a different level where [an investigator] can read the notes our claims adjusters are putting in the files as they document the actions taken in the claims. [The adjusters] will be able to drive different types of rules off those notes that will help us raise additional red flags so we can get them in the hands of our SIU."

Infinity is focusing on claims for its text analytics program. One area is helping to identify fraud, and the other is what Smarrelli calls "the right tracking of claims."

The goal in tracking claims is to get the claims assigned to the correct adjuster as quickly as possible. The software identifies which claims are going to be more complex right at the time of the loss report. "As we enter that information, we are able to obtain the scores the text analytics helps us identify. [The scores indicate] which level of adjuster that claim should be assigned to," he says.

KEYWORDS

Infinity will be able to utilize this system to pick out keywords and phrases that will alert business users to situations in the file that may not have caught the adjusters' attention. "We are moving from a manual process to a systems-generated process, which helps us with the management of our claims," says Smarrelli. "We should be able to drill into almost all of that unstructured data and focus on keywords and phrases."

In addition, the search could involve specific pieces of information to see whether there is any correlation or link analysis with a fraud case, indicates Mike Jorgensen, director of the claims call center group for Infinity. "Now, we have the ability, without human intervention, to go through those files and be able to conduct a deeper investigation into any current fraud rings," he says.

The key for insurers in dealing with data is to develop some form of common terminology–a data dictionary–so members of the organization can speak and understand the same language, emphasizes Smarrelli.

It is important to have the carrier's independent adjusters use some of the same type of terminology, adds Jorgensen, pointing out the SPSS system has the dictionary in it and it will pick up different abbreviations to make sure those match or be changed to fit the carrier's needs.

STRUCTURING YOUR DATA

Kurt Bonigut, CIO with Frontline Home-owners Insurance, sees two primary components within the carrier's foray into unstructured data. The first part is the unstructured data it actually produces on a cyclical basis, such as company documents for agents and insureds, which include declaration pages, invoices, renewal notices, and cancellation notices. "That's a massive source of unstructured data we ended up producing," says Bonigut.

The second component is the unstructured data Frontline receives from third-party sources, which includes claims-related information for claims processing, such as adjuster reports.

For the purpose of categorizing, the unstructured data Frontline produces always is treated in a consistent manner, explains Bonigut. "We were able to work with ImageNow and take the output stream, slice it and dice it, put it in the appropriate document categories, hold it for internal review purposes, and share it electronically with agents in the field through our Web site," he says.

As part of a substantial rebranding effort one year ago, Frontline implemented a number of technologies to take its existing print output, reformat it, and create a more polished finish.

"We essentially created the indexing parameters for each of these new outbound documents and had ImageNow look at the spool file, output the values, and categorize that information as appropriate within the system's structure," says Bonigut.

As of January 2010, Frontline is going completely paper-free with its agents, with all of the documents the carrier produces available on the carrier's agency portal for the agents to print or download.

"It's a huge benefit for us and for our agents to move away from a paper-based world and live in the paper-free environment we do," says Bonigut. "The integration to get that information out to our agents was seamless, and the benefits were immediate. Our agents are big fans of that."

CHANNEL GUIDE

With the workflow components that are available, Frontline is able to take documents through a number of channels–fax, e-mail, or hard-copy mail–and get those documents split into the appropriate beginning phases of a workflow.

"If our claims people are reviewing claims information they have received electronically, they quickly can link documents to the appropriate claims and get them in the appropriate bucket," says Bonigut. "Any unstructured data that comes in has to receive a certain amount of review to determine how it should be attached, but [the system] is fast and easy."

If a supporting document comes to Frontline for a given claim, it will be scanned, identified as a claims document, and go into the first step of a claims workflow. A claims specialist will look at the document, identify the claim and policy it relates to, link the document to the policy and claim, and route it to the next logical step in the business process function, according to Bonigut.

"If it is a document that needs to be reviewed by a supervisor to be sure it meets the criteria we have in place for a given process function, it will be escalated accordingly," he says. "There is a whole series of business logic we have behind the process whereby the document is moved through the workflow system, but in terms of taking the document and getting structure out of the unstructured document, that is very simple."

The IT department worked closely with each of Frontline's business units as they were brought online, reports Bonigut. "We had a good feel about what our indexing requirements would look like and how indexing would affect our world," he says. "We found with an imaging and workflow system this powerful, you really can rethink a lot of your business processes. A lot of things you are doing in a paper-based world you don't necessarily want to replicate. You want to be able to squeeze efficiencies out of the system. It was critical for everyone to understand the power of the tool prior to implementation so design doesn't change implementation."

NEW MEDIA

Carriers need to be aware of new media when considering unstructured data policies. Goldberg notes many insurers offer live chat for customers. "These are interactions that can and should be stored, just like the unstructured text of any mail," he says. "Most carriers are approaching this as a whole chat log being one document that can be stored or retrieved later."

Goldberg has seen some experimentation with social networking, but nothing too drastic. "What I think you'll see more of is business networking as opposed to social networking, where your independent agents can share knowledge and experience with each other," he says. "It's more user-generated content."

There still is much work to be done putting structure into the video carriers receive, although Pauli remarks there is little doubt video has value in determining who was at fault or not at fault in an accident and whether to adjudicate or not adjudicate.

"Frequently, the video comes in with someone narrating it, but it is unstructured information the carrier has to deal with somehow," says Pauli. "It has to be stored somewhere, so there are security and privacy issues."

From a marketing perspective, she anticipates carriers that own their own distribution channel or direct writers will be using text analytics to figure out what people are saying about them on the social networks. "That will be the wave of the future," she concludes.