Artificial intelligence is slowly being incorporated into systems thatsupport underwriting and fraud detection.Insurance companies have been wary about implementing artificialintelligence. That's not surprising, considering that carrierstraditionally have been slow to adopt cutting-edge technologies, preferringinstead to wait until other industries have proven the technologies to beeffective.
But industry observers believe that the industry is ripe forapplications that use artificial intelligence. Fraud detection,underwriting and data mining tools are among an increasing number ofsoftware applications that are incorporating some form of artificialintelligence.
"Artificial intelligence performs tasks normally done by a humanbeing," says James Bisker, a research analyst with TowerGroup, Needham,Mass.
"An insurance company would want to use artificial intelligence inunderwriting to improve consistency, to maintain loss ratios at aprofitable level, and to train novice underwriters more quickly by havingthe expert system mentor them."
Artificial intelligence attempts to mimic or replicate the humanreasoning process. There are two types of artificial intelligence, symbolicand constructionist. Symbolic manipulates pre-set rules in an effort toimitate the way people make decisions. Constructionist, a more ambitioustechnology, aims to duplicate the structure of the brain itself.
The constructionist approach replicates human neural processes. Throughthis technology, a computer is fed a course of training data then sent outto acquire information on its own.
Identifying patterns
Pattern recognition, one of the fundamental talents of the human brain,is a hallmark of constructionist artificial intelligence.
The most common use of pattern recognition is a neural network, whichtakes hundreds of pieces of information, identifies common patterns andthen fits the puzzle together in a complete picture.
A neural network, Bisker explains, can recall patterns from its memoryand store new patterns.
"The most fully developed artificial intelligence concept for theinsurance industry is fraud detection, and this is mostly done with neuralnetworks," says David Bradford, president of Heurisktics Group Inc., aBedford, N.Y.-based developer of text mining and other artificialintelligence tools used by risk managers and insurance underwriters.
Underwriting assistance
The underwriting process, and to an even greater extent, onlineunderwriting, is another area of the insurance business that artificialintelligence is making a headway.
The most effective use of underwriting programs that use artificialintelligence, Bradford says, is for underwriting personal lines policiessuch as property and casualty, auto and homeowners policies that arepurchased via the Internet.
Esurance Inc., a San Francisco-based company that sells online autoinsurance in 24 states, has a rules-based underwriting system that employsa form of artificial intelligence to offer competitively priced coverage.The system makes determinations based on its evaluation of a customer'slong-term value.
"We use decision analytics and data mining to measure the effectivenessof market response, but over time, we will be able to apply this logic toour customer base," says Glynn Evans, Esurance's CIO. "To the customer, itwill mean faster service and a very competitive quote."
Furthermore, artificial intelligence frees up the senior underwritersso they can work on the really complex problems. "An expert system willgive you the same advice every time," says TowerGroup's Bisker. "Itstreamlines the whole underwriting process."
One emerging application of artificial intelligence in the insuranceindustry is for claims valuation, a process similar to fraud detection. Thesystem determines what the value of a claim should be based on itsevaluation of comparable claims in the database. While the results are notalways precise, they tend to be more accurate than those obtained by claimsadjusters who often arrive at decisions through subjective analysis andguesswork.
Rules-based artificial intelligence, Bisker explains, takes the humanfactor out of the equation and generally produces more consistent results.However, the industry currently does not have front-end artificialintelligence tools that can help uncover fraud when a claim is entered intoa claims administration system and before it is first handled by a claimsadjuster, says Gary Craft, managing director of e-finance research for DBAlex.Brown, San Francisco.
"Insurers first need to improve their workflow systems, and then front-end artificial intelligence tools can help determine if the costly humancapital needs to review a claim or pass it through," he explains. "Onceworkflow systems, data management systems and fraud screening beginsaturating the claims industry, it is likely that artificial intelligencewill become an important tool used by the claims professional."
Mining for clues
One application using artificial intelligence that carriers arebecoming more familiar with is data mining. Data mining software usesneural network technology because of its ability to recognize patterns,classify information and predict outcomes.
Other methods of data mining include rule induction and classificationalgorithms, which are often employed with neural networks to increaseefficiency and workability.
"Artificial intelligence is a very powerful tool for data mining,"Bisker says. "It's kind of like reverse engineering. You might, forexample, look at a person's book of business to establish what rules theywork with."
Another approach is to look at a set of data and determine what can belearned from it.
"When the process works, you may end up uncovering rules that youdidn't know you had, and that maybe you shouldn't have," Bisker explains."There may be procedures in effect that are needlessly costly, totallyinefficient or downright illegal. As a company gets bigger the rules getmore complex, and harder to track. It also becomes a lot harder tounderstand the implications of a subtle change in the rules set."
Data mining frequently is used by insurers to expose liabilities inunderwriting rules. It also helps insurers identify their most profitablecustomers and provide better customer service.
"Data mining helps find claims that slip through the cracks," saysJames McCormack, CEO of Reclaim Technologies and Services Ltd., Newark,Ohio. "Whatever we find, our clients collect from reinsures."
For catastrophic claims, the company defines each type of catastropheelectronically and designates each with a distinctive signature. "In theend we know where it happened, when, and what lines of business wereaffected," McCormack explains. "Once we have that signature, we use thesame parameters to develop a signature for every comparable claim in thedatabase. Then we compare those claims to the catastrophe signature, andthat's how we find the claims our clients miss."
As artificial intelligence technology improves, experts say insurerswill begin to incorporate more sophisticated applications.
Underwriting support systems, for example, will be designed for muchmore difficult types of risk. Underwriting systems that currently useartificial intelligence are mostly rules-based, but future systems willemploy more complex neural networks, which will empower them with a higherdegree of intuition and allow them to make judgments and inferences.
Many of these technologies are already used in medical diagnosis andpharmaceutical research, but so far they haven't advanced too far beyondthose areas.
Text mining is a new artificial intelligence technology that someindustry observers believe will be widely adopted in the near future. Thiscutting-edge application makes it possible to identify individual conceptscontained in text documents and determine how all concepts within thedocument are related.
In the late 1990s, MetLife began developing a system that could analyzetext for underwriting relevance. MetLife's Intelligent Text Analyzer wasdesigned to analyze open-ended questions pertaining to life insuranceapplications, according to Barry Glasgow, a MetLife information systemsspecialist, who described the system in an article written for The AmericanAssociation for Artificial Intelligence.
By mining consumers' responses, the system could begin to makeassumptions that could not be gleaned from replies to yes or no questions,or inquires that were answered with a single word.
"Text mining will allow underwriters and other insurance professionalsto take advantage of the treasure trove of data currently online," Bradfordsays. "Right now most of the stuff is often inaccessible because theinformation you need is usually buried in a mountain of words. With textmining you'll get the bottom line and nothing else."
D. Douglas Graham is a freelance writer based in Columbia, Mo.
