Some senior insurance executives have a hard time understanding how a seemingly esoteric technology such as artificial intelligence (A.I.) could possibly be used by the insurance industry.This mental juxtaposition is ironic because the potential applications are manifold. And more importantly, the insurance industry was one of the first to widely adopt artificial intelligence technology in the form of expert underwriting systems.
Insurers are becoming more aggressive in their adoption of technology to control costs, increase overall efficiency and to focus on the customer. A.I. and its offspring-automated decisioning, business intelligence and data mining-have proved to be useful tools in a wide variety of industries as the technology has matured over the past several decades.
As the volume, complexity and availability of information continues to expand in a more connected world, the financial services industry will need to rely more on the unique capabilities of this technology. This will be especially true of the insurance industry as it seeks to successfully compete in the world of converged financial services.
A Brief History
A.I. has been defined many ways in the 46 years since the term was coined, but can be characterized as a branch of computer science that works to emulate (and eventually automate) processes and tasks normally requiring human intelligence.
When these programs and algorithms are applied to business processes, they are sometimes referred to as intelligent or smart systems.
The "holy grail" of A.I. is not to take over the world or to replace humans with robots. A.I. is not a "computer brain." Some A.I. programs can see and recognize objects, but not very well (yet). A.I. programs can read text, but can't understand it in context the way humans do. Various aspects of A.I. are in use today in applications that range from fraud detection in financial services to controlling shift points for automatic transmissions of automobiles.
This list can easily be extended to broadly define what this technology can and cannot do through various examples. A.I. works and is improving all the time due to diligent research.
Has it ever been over-hyped? Certainly. Will it ever reach the highs (and lows) that are predicted in science fiction? Perhaps not in our lifetimes. But A.I. should be viewed as an important tool for any business and one that should not be ignored.
There are many subdisciplines of A.I. and some of those have produced commercial products that are being used by insurance companies. Most are related to the capture and use of knowledge or have to do with pattern recognition.
The most familiar form of A.I. used by business is an "expert system," also known as a rule-based system. Both are technically called knowledge-based systems (KBS).
The goal of a knowledge-based system is to enable the use of knowledge of a human expert in meaningful and consistent ways.
The insurance and banking industries understand that their underwriting process is the application of a set of rules used to either accept or reject an application. Most carriers have adopted this technology and use some form of KBS in their underwriting processes.
The chief benefits of using an expert system are: it enforces consistency in decisions it handles, it provides the ability to review exactly how a decision was made, and it allows for modifications to rules without reprogramming the base application.
The concepts found in KBS are also used to process "business rules" within software applications to facilitate flexibility and consistency. Business rules are a convenient way to express the actual programming logic that is the core of any software application.
A "rules engine" may drive the actions of software applications that process policies, claims or calculates new rates.
Just in Case
Case-based reasoning is another type of knowledge-based system. This technology is based on research into representations of knowledge.
One of the most common ways to represent knowledge is to describe the features of a concept in a consistent manner and compare them to previous "cases" of similar construction. This A.I. technique can be used to detect fraud, recognize opportunities for cross-selling or be used to match customers to financial plans that provide the best result for their financial situation.
One of the unique advantages to using this form of KBS is that it can "learn" as it acquires new cases.
Some of the most useful but hard to explain A.I. technologies are used for pattern recognition. The ability to detect patterns in collections of data is key to intelligent behavior and is very useful in the insurance industry.
Examples of important patterns to catch include excessive underwriting risks, fraudulent claims, marketing opportunities and pricing trends.
Data mining uses A.I. techniques such as classification, knowledge discovery and rule induction to uncover information and knowledge potentially hidden in existing data. A life carrier may use automated underwriting systems to assist inexperienced workers and control the consistency of their books of business.
As the underwriting guidelines are being entered into the expert system, it may be useful to examine existing policies to determine exactly what rules were used when the business was accepted.
Neural networks
One of the most powerful A.I. tools available today is an artificial neural network. This is a simulation of the biological neurons found in brain tissue that have been coded as computer algorithms.
Mathematically, they are nonlinear curve-fitting programs that can be linked together to analyze data. Although hundreds of different types of artificial neural nets exist, some of the most useful are referred to as multilayered, back-propagation neural nets.
This type of neural net needs to be "trained" to recognize patterns. For instance, if the neural net is designed to recognize a fraudulent automobile claim, it will be shown thousands of examples of normal claims and a statistically accurate number of claims that turned out to be frauds.
The neural net reviews each data field of the input record and compares it to the actual disposition of the claim to learn the difference between good claims and bad ones.
Once the neural network is properly trained, it can be used to analyze claims it has never seen and very accurately detect the fraudulent ones. These programs also can be used to classify new business for more accurate pricing or develop segmentations of policyholders for marketing purposes.
The A.I. Winter
One of the reasons that A.I. draws mixed reviews from business people and IT professionals is that it had its own boom and bust cycle in the mid-1980s.
Although there were successes, many projects failed to produce the advertised results and A.I. research funding dried up. However, some A.I. projects were commercialized and were over-hyped by eager marketing and sales departments.
When both the military and commercial aspects of A.I. met with reality, it led to an "A.I. winter" of reduced funding and reduced research, which lasted for almost a decade. Research did continue and many positive results were achieved during this time, but A.I. was tarnished.
The cause of the perceived failure of this technology was rampant expectations and greed. However, expert systems continued to work in their domains of expertise, and neural nets caught fraud and recognized market segments they were trained to find.
Today, the technology is even more mature and is used in an ever increasing number of applications. For example, almost 90% of all U.S. credit-card transactions are checked for fraud by neural networks in real time during the point of sale swipe.
Whenever a form of judgment or expertise is used in an insurance workflow, A.I. techniques should be reviewed to see if they can automate the task to provide consistency and reduced costs.
Besides their obvious use in underwriting functions, intelligent systems could be used in fraud detection and prevention, marketing, pricing, and reinsurance. Even a carrier's response to the regulatory requirements, such as those of the USA PATRIOT Act, could be assisted by this technology.
Although it is not a panacea to all the problems of the insurance industry, artificial intelligence is one of the few technologies that should be taken to business as a solution looking for a problem.
James Bisker is a senior analyst at TowerGroup, Needham, Mass.
