Many of the best technology tools that have improved performance among insurance carriers have multiple layers of productivity and nowhere is that more evident than in the use of predictive modeling and analytics. Today, not only are models assisting insurers in their ability to predict outcomes, the analytical tools are also helping carriers prevent greater losses in areas ranging from workers’ compensation to catastrophes.
As one analyst sees it, the use of predictive modeling and analytics allows carriers to “up their game,” according to Bill Jenkins, a consultant with Agile Insurance Analytics. Modeling enables insurers to not just compete with others in their market, but to grow their business as well, adds Jenkins. The bad news, though, is most insurance carriers still have a long way to go to attain the real value that this technology presents to the industry.
“[Analytics] has made insurance a much more competitive market,” says Jenkins. “Some of what I call Stage 5 carriers—Progressive, Allstate, and other big carriers—are jumping into analytics with both feet and are using the tools to get more focus around the long-term value of the customer. They are using modeling to further differentiate themselves.”
Prevention the Key
Insurance companies traditionally focused their predictive modeling on the underwriting field because it was easier to develop models that predicted potential losses of policyholders. It was a straight-forward process, explains Marina Ashiotou, director of predictive modeling, Accident Fund Holdings, Inc.
Carriers used that information to determine a fair premium to charge a customer that also would enable the insurer to be profitable. The insurer would look at a renewal or whether to write a new piece of business and base their decisions on the models that had been developed.
Carriers have begun to look at prevention of incidents as yet another way to develop the models and improve profitability.
“When you look at prevention, the models are different,” says Ashiotou. “It’s not just about the policyholder. The information that you need to use to do your modeling becomes more granular because now [in a workers’ compensation policy] you care about the employees, how skilled they are, their age, characteristics, and the location the business is operating in.”
The information is different from what has been used for underwriting a policy, explains Ashiotou.
One area many insurers have been using data characteristics is with fraud. Success there can make a positive impact on expense ratios, according to Jenkins.
“Carriers are using characteristics or variables around projecting expenses as they relate to claims as well as claims severity and frequency,” says Jenkins.
Loss expenses are being analyzed and insurers are seeking independent variables around the account information the carrier has on their policyholders.
“Carriers are getting a better feel around what kind of investigation needs to take place,” says Jenkins. “Analytics have been used for years around reasonable and customary costs for claims adjudication. Carriers are looking to refine that even further and push through models with more granular information on the policyholder, claims history, and policy history.”
Karen Pauli, research director at CEB TowerGroup, also believes prevention is changing the way carriers conduct predictive modeling. Prevention requires an expanded use of external data or data from non-traditional sources.
“There are some great third-party data providers—LexisNexis, Axion, TransUnion,” says Pauli. “These are companies with great data. Tools such as telematics also give insurers great data. There are things insurance underwriters and modelers and actuaries wouldn’t know without that data.”
Another advantage modelers have is what Pauli calls the relational social types of data. For instance, SAS Institute has data on fraud where they add the relational data so insurers can determine the fraud range, adds Pauli.
“You couldn’t do that as an underwriter or a claims person,” she says. “It’s great data and it is changing outcomes, particularly cutting off a claim before you pay [too much]. External data definitely is helping.”
One challenging piece of information that companies are looking for in claims predictive modeling, and the prevention of incidents is text analytics, which refers to the process of deriving high-quality information from text. This information is usually derived through devising patterns and trends.
“Carriers have discovered in the notes turned in by their claims adjusters that there is so much relevant information that has not been captured in any other way,” says Ashiotou. “There is a move to try and systemize that information so it can be used in modeling.”
Carriers have been unable to perform this type of modeling in the past because the data wasn’t ready to perform this kind of work, according to Ashiotou.
“With models for underwriting, nearly anyone can pull the data from fancy policy systems,” she says. “But when you are using models for prevention companies don’t necessarily have the data available to do any sort of analytics or modeling.”
Ashiotou points out that in other lines of insurance, particularly life insurance, there are similar problems involving prevention.
“There is no data available there to do real predictive modeling,” she says. “What companies find themselves doing in that space is spending five years or so to get their data in shape to do the analytical work. Not a lot of [life insurance] companies have done text analytics because it is challenging and not the push-of-a-button type of solution where someone can supply you with the software and say now you can do text analytics. This is something you need to work on quite a bit before it can become useful.”
Jenkins maintains the business rules used in modeling should be the province of the user community.
“[Business users] have to define what earned premium is and what claims count means,” he says. “IT can’t do that.”
Jenkins recently met with a group of insurers and the discussion led to who had the responsibility of maintaining the business rules. Jenkins was disappointed when he learned one carrier didn’t have a data governance policy in place.
“They planned to sit down as a group to do that, but I cautioned that they were going to have problems,” he says. “It really gets down to a data governance process that is required. The business rules generally are put into place by the users, but they can be refined as things go forward and as companies utilize predictive analytics and see the outcomes, things like live data and third-party data, and the rules around them get adjusted.”
Pauli maintains that throwing more information at already complex claims or underwriting processes is not always helpful, which is where predictive analytics and models come in. Carriers need to have the analytics available to pick out the predictive data, put it into a model, model the predictive outcome, and then to be able to automate that model to scores, straight-through processing, decisioning or decision support.
“You can model the decision and the underwriter or claims person looks at the model and then uses their experience to decide what that information suggests,” says Pauli.
While much of the industry has been focusing on refreshing legacy systems, Jenkins believes these carriers have been ignoring the data piece of the business.
“That’s a shame,” says Jenkins. “Data is a differentiator with a company. [Insurers] have their own history with customers and have experience from a service standpoint and an account standpoint, but they are not using that data as well as they should.”
One area of uncertainty concerning modeling deals with which side of the operation should handle implementation of the business rules surrounding the models. Pauli concedes having the business side perform the work would take the burden off IT, but she points out that it would be a change in culture.
“IT is used to doing certain things while actuaries are used to working with models and handing an outcome to underwriting and claims folks,” she says. “You have to look at the scarcity of resources that are competent and proficient in things like analytics and models. The number of business analysts that have those capabilities are needed to adopt certain technologies to enable the business side to interact with the models.”
Pauli believes there is a great advantage in letting a business person interact with the models and let the business person perform “what-if” scenarios. She adds that the ultimate goal for an actuary, a modeler or a mathematician would be to understand nuances, so they can interact with the model.
“Maybe they even break the model—push the data to the point where the model is not working anymore,” she says. “That way they are finding out what they shouldn’t be doing. That’s valuable to a business person. It wouldn’t be valuable to a modeler or an actuary because they are trying to hit a goal. We said all last year and now in 2013: The business side needs to interact with the models.”
Business rules are extremely important in the modeling process, agrees Ashiotou.
“Traditionally, whenever models were introduced, the modelers were hesitant to say the model was the ‘be-all and end-all’ and [business users] should follow it blindly,” she says. “It was always accompanied by a set of business rules that could override the model or would work in conjunction with the model to recognize that just using something as automated as a model—essentially a formula—is not enough for a company to be operating.”
Still, as the models are being used, the underwriters discover that sometimes things need to be changed with the business rules because conditions have changed or something else changed that necessitated abandoning the rules.
“One example in that area is fraud,” she says. “You always have to be one step ahead of people who commit fraud. When you develop a fraud model you build it based on prior information, but if you are talking about claims fraud there might be a new type o0f fraud that the insurer has never experienced before and the model isn’t capturing that information. That is when you need to change business rules until you can incorporate the changes into your model in a more systematic way.”
Jenkins has seen large carriers dabbling in the customer service area with predictive modeling, pointing to a Gartner report that stated just one-third of property & casualty carriers know their customer.
“[Knowing their customers] is still a struggle for the industry,” says Jenkins. “That’s a master data management issue and as you know policy systems are really policy centric. You have a policy as a homeowner, but carriers don’t know if that customer has multiple policies with other lines of their business. End-to-end becomes a focus for carriers to consolidate that information and assign a unique identifier to build a customer hub so the carrier can utilize the data to run analytics and inquiries against.”
Ashiotou points out that on the claims side, if the insurer is assisting the employer in getting their employees back on the job in a timely manner as well as provide those injured employees the best of care, there is a direct return to the customer.
“With underwriting, it’s all about trying to charge the right premium,” she says.
Models generate straight-through processing and Pauli believes an average consumer with an average risk profile demands a quick answer to questions of changing limits, or adding a vehicle or driver.
“The models allow that so most of us don’t have to go through some torturous underwriting process,” she says. The models deal with our data, provide the answer, and you are done.”
Pauli puts the terms predictive analytics and modeling together because she feels carriers can’t do one without the other. Models are changing the insurance experience, she adds, but there remains a long way to go to put models in small commercial lines.
“For most insurers, modeling improves their lives,” she says. “It lets underwriters and claims people deal with the hard situations and takes the routine work off their desks.”
In building models for claims in workers’ compensation, the problem is basic: there is a claimant and they are injured at work. The claimant makes a statement and the insurer collects all the information, explains Ashiotou.
“At the end, you want to know how much this claim is going to cost and if there is anything you can do to make it less expensive,” she says. “A lot of times you don’t have any information about pre-existing conditions and [the claimant’s] personal information. It becomes challenging to do modeling at the claimant level because you don’t have all the data available. Some companies are creating a system where adjusters can enter that type of data, but you need a few years of experience to build a model.”
Gaining full advantage of models and analytics is helped greatly by having modern core systems in place, points out Jenkins,
“Part of the problem with legacy systems is trying to get granular-level data out of the systems,” he says. “It’s difficult because when those systems were put in, some of the data went away—it was summarized. Legacy systems limit the ability to extract data unless someone customizes the systems to the degree that they have changed the documentation. [Business users] don’t know whether that information was consolidated, summarized, or dropped altogether. It’s difficult to get data out of those older systems.”
Even if carriers purchase a suite of new systems from one vendor, Jenkins points out the data still has to be synchronized, which he feels is a huge endeavor.
Without that synchronization, Jenkins says, “It will keep carriers from getting granular data to look at patterns of data at a detailed level so they can compete better and know their customers better. There is a progression. Where carriers seem to be headed today is to segment their business—class underwriting. Most carriers write that way.”
The movement today is toward relationship-level pairing of data—risk underwriting—points out Jenkins.
“The old underwriting adage is there is a right price for every risk, but carriers are mainly doing class underwriting,” he says. “The idea in the classes is there will be profitable risks and unprofitable risks, but you need more granular information about the risks so you can write the risks properly.”
When Jenkins served as CIO for Penn National Insurance, the carrier started off with the three basic tiers of rating: standard, sub-standard and preferred.
“We went from three risk categories up to 29,” he says. “If you look at larger companies, they now have hundreds of categories.”
The next frontier is telematics where an unprecedented amount of data will be available to analyze the driving capabilities of a customer or potential customer, according to Pauli. Modeling with telematics data is in its infancy stage because of the volume of data.
For 2013, CBE TowerGroup predicts that because of things like high-speed memory, analytics can deal with all of that telematics data in an effective way that will start changing as carriers look to get new information sources to create competitive advantage.
“It’s a real change for IT departments to deal with that much data so you have to look at new sources for handling the data—probably in the cloud—because there is so much data,” says Pauli. “The technology is right there to support the carriers in using it, though.”
In the coming year, Pauli predicts the industry will see more modeling activity conducted in the specialty and commercial lines.
“There’s a real need in that area from a competitive perspective,” she says. “We are going to see that move up the food chain in terms of complexity. It will be a very different world when you get the telematics and social media data for insight. Things are going to get more complex and we need more people with skills.”