 |
|
 |
|
|
|
| Data Warehousing | |
|
|
| Meta Data Management | |
|
|
 |
Articles & White Papers |
|
|
 |
|
 |
|
 |
|
|
|
By Michael Haisten
Insurance may well be the first data-analytic business which developed long before the age of automation. Profitability in the insurance industry requires the ability to gauge risks and rewards with a high degree of precision. The financial research of actuaries may have done more to move forward both the descriptive and predictive fields of statistics than the combined impetus of the natural sciences.
Yet this very intense dependence on aggregate statistical analysis created a data analytic blind spot. The industry became experts in normative or expected behavior but missed the fine distinctions that predict radically different outcomes. And strangely, the big picture regarding profitability by line, market, or customer was fuzzy at best.
During the last half of this decade, insurance companies began using data warehousing techniques to create new business opportunities. They did this by uncovering new perspectives at both the macro and micro level. The macro view allows them to understand a market or geography in rich detail by combining a broad array of external sources such as demographic, psychographic, public record, and locator data with the internal experience base. The micro view exploits pattern recognition techniques to reveal hidden, and sometimes counter-intuitive facts and associations that are then used to create new products, services or processes.
The rest of this article explores high-impact opportunity areas.
Risk Management Revisited
First, we offer an example of how finite analysis via one client's data warehouse caused a controversial but lucrative revision of cherished risk management assumptions. We all know that moving violations correlate with increasing risk. The exact increment of additional risk projected for each new violation per unit of time is one of the most proprietary variables in the industry. Regardless of the formulae, it is easy to assume that risk goes up with each new violation.
The startling conclusion is that some violations, when they occur alone without any others, may actually indicate reduced risk or at least no increase at all. For example, a class of drivers who receive only minor speeding tickets are a lower risk than the average population with no violations at all. Don't look for speeder discounts anytime soon, since that would be a somewhat impolitic thing to do. However, fellow leadfoots out there can now rest easy that at least one company now knows the truth and uses it creatively in their approval process.
Claims Analysis
In recent years, claims analysis has become the most prevalent, and the most successful, use of data warehousing in the insurance industry. Companies have accumulated archives of 5, 10 or even 20 years of raw claim or loss data. This base of insured party and incident data can be assembled into a rich and detailed analytic resource when combined with the right contextual information.
The process starts by categorizing claims using internal attributes known about the claim incident itself, insured party, insured property (if appropriate), claimed and total coverages, cost components and anything else recorded in the operational systems. By itself, this data can be used to calculate raw statistics and detect trends that form the baseline for further analysis.
This core-claims data is then enhanced by the addition of external data already acquired for operational uses. For instance, credit history is traditionally used as a part of the acceptance processing for new policies or when extending coverage. It can also be used to indicate suspicious claims. Conventional wisdom says that when someone is tapped out financially, they are more likely to file a fraudulent claim. But did you know that an identifiable set of individuals who have filed for bankruptcy are considerably less likely to submit a bad claim?
This counter-intuitive fact is one of the contra-indicators created via more in-depth analysis. Contra-indicators are based on a more precise understanding of the circumstances of the insured party or the claims incidence itself. These indicators are used operationally to help determine where to direct increasing expensive investigative resources.
Service Optimization
You get near term and tangible payback for your warehouse investment when analysis creates the opportunity to take cost out of the system or to better utilize existing resources. A potentially higher value outcome is to create completely new service offerings.
One medical insurance firm used a data warehouse to put a more happy face on managed care. They compiled an extensive database of procedures and client histories to create what they call "optimal patterns of care." It can be used to quickly pre-approve a sequence of care rather than the typical slow one-step-at-a-time drill that is the bane of the industry. Their objective was no less than to turn around the miserly and almost cruel image of managed care.
This process met with some resistance when it seemed to limit the prerogative of the physician. However, because it was positioned as a research tool as well as an administrative aid, it began to win over both doctors and their patients.
Customer Share
The insurance industry has long known what other businesses are only just learning: It is more efficient to sell more to an existing customer than it is to find a new one. This customer share strategy is a guiding principle of agent behavior at the grass root level but historically little corporate support existed. Generally, only your agent has any chance of knowing what policies you own and what coverage you have across the board. They are the only one likely to know your needs and whether other firms are helping you meet them. This is sometimes exacerbated by stove-piped line of business organizations that make it difficult to sell the full suite of services to a customer.
A data warehouse can be used to link customer information across lines of business to produce a composite customer view. Comparative analysis and profiling can create representative portfolios of insurance coverage. Gap analysis can identify customers who have potential needs not being served today. This can feed marketing campaigns by outbound telemarketing or traditional field agent outreach to sell more business into the existing customer base.
A more hardy solution couples a data warehouse with a closed-loop customer relationship management application. This offers a continuous and more integrated flow of information between either field or remote agents and other field or corporate functions. This allows you to collect and maintain competitive coverage data. It supports synchronous and coordinated update of insured party data regardless of whether it is received by an agent, a claims processor, a customer service representative, or various connected third parties. In turn, this allows the data warehouse to present a current, consistent, and complete view of customer holdings and activity.
Agent Deployment and Location Analysis
Insurance companies are getting more sophisticated in how they deploy both sales and claims agents in the field. Using precise geographic analysis of the location of both current and potential customers, the right number of agents can be placed in better proximity to their customers.
One of the most vexing problems is to know with certainty where the insurer party or the insured property really is located. Real estate policies are likely to require a specific and geographically accurate address. This is less true of auto, life and medical insurance. A data warehouse can be used to collect and sieve through all of the available address and locator data to help determine which is current and the most likely to represent a complete and accurate residence address.
For some business questions, it is becoming important to know such things as how often an insured party moves as well as where they move from and to. Are you losing customers because they leave your coverage area? Do they switch when they move even when you offer service at their new location?
A more ambitious goal is to build "insured life histories." This means to compile a timeline of historical information about the customer including all past locations and changes in coverage. This can be a minimum effort based on backtracking through old information for extant policies. It could extend to linking back to expired policies the customer may have held with you in the past. This often requires complex processing to correctly link these prior instances.
The most audacious approach uses external public records to fill out the picture of where the customer lived and with whom they worked over time. This sounds a little like Big Brother tactics but the intent is to use the information in aggregate to better match needs to services.
Tracking the geographic location of accidents is one of the things that has revolutionized automotive claims processing. Prior analysis clearly demonstrated that, on average, claims costs go down the closer you bring resolution of the claim to incident. One response has been more geographically dispersed drive-in claims centers. The most radical response is the introduction of "crash vans" that process claims at the scene or soon after at customers' homes.
Geographic spread analysis is performed to determine where to place fixed assets or where to stage mobile assets. Claims centers need to be built in areas of high incident within reasonable proximity of where the customers live. Mobile assets can be dispersed more broadly and even be dispatched from employees' homes. The vans need to be placed efficiently based on where the accidents are more likely to occur.
Market Coverage and Niche Development
Markets are being defined by factors more complex than basic demographics. Customers are being targeted based on a more sophisticated view of behavior. This is very different from the traditional methods of actuarial analysis or underwriting. Psychographic analysis, long a tool of consumer packaged goods companies, has come to the insurance industry. Knowledge of what you like, what you do, and what you buy can be used to better predict your needs for insurance. The fact that you are a young, upwardly mobile Internet addict that loves movies but shuns day trading might indicate a strong propensity to buy a variable life policy backed by a self-managed mutual fund portfolio.
A data warehouse can be used in all phases of market identification and penetration. Companies begin with installed base analysis using directed surveys and externally purchased data to identify characteristics of current customers by policy and coverage type. Segmentation analysis defines sets of characteristics that exhibit strong potential to buy a certain policy type. Targeting involves sieving through prospect records that have also been augmented with the characteristics of interest to create a campaign list. A data warehouse can be used to track and evaluate the success of a new market campaign.
You can go further by using a data warehouse to define new product offerings. Some of the notable successes in the insurance industry in recent years are the aggressive niche players. One example is to carve out a profitable space mid-way between premium customers and the assigned risk pool. It is easy to define the low risk cream of the crop. It is much harder to define a safe middle ground. The successful companies analyzed the factors of risk and return more broadly and more deeply to create finely tuned indicators used to pick their customers. Even though their prices may be higher than the premium-only vendors, they found a ready market in customers that had previously been frozen out.
Another form of niche market is low volume, high-risk groups that have traditionally been self-insured. One company created a new policy for snowmobile tour operators. Snowmobiles have the same bad reputation for accidents as all terrain vehicles. Insuring tour operators that must handle first time and careless riders is pushing the envelope for many traditional firms. This company did extensive analysis of the profitability and operation of this small but growing market. They found a set of coverages that provided more than adequate protection for the operator at a price that was fair to both parties. They did require strict adherence to customer safety guidelines and to validated service procedures as a condition of coverage.
The data warehouse is the source of cross-functional and industry data used to identify a new niche market. It is also the means for monitoring the success of the new offering.
Product Profitability
More insurance companies are expanding into multiple lines of business across a wider variety of specialized customer categories. Products are becoming more narrowly defined and are being deployed against more focused target customer groups. It is no longer acceptable to measure profitability at just the business unit level or only for a few large customer categories.
A multi-line company may have life, auto, home, and business products. Profitability traditionally is measured at the business unit or major line level (e.g. auto versus home) and at the minor line level (whole vs. variable vs. term life). The general ledger processing is setup this way and so are most of the available financial reporting options.
Now companies are moving to a product level focus with visibility to geographic and customer segment variability. Before you may be able to track how the Variable Life product line is doing by state. Now you can compare results of the Aggressive Plan B Variable against Conservation Plan C Variable by major metropolitan area across customer segment and income variables. Product differentiation is starting to play a greater role in both customer retention and market growth strategies. A comprehensive and standardized dimensional model is an essential tool in this new endeavor.
Customer Profitability
How do you define profitability in general when the timing and pattern of revenues (regular and predictable) are so radically different from the timing and pattern of direct costs (haphazard and unpredictable)? Insurance deals in much longer time horizons than most others businesses while still having to report quarterly results. There are generally accepted accounting principles for recognizing revenue but this only applies to aggregate totals at the business unit level. If product level profitability is difficult, customer level profitability is a nightmare.
Future success in the insurance industry will require fundamentally changes from past practices. The business will be managed by new rules that target customers more finitely and then measure the results with greater precision.
For most of the modern history of the insurance industry, it was good enough to know how well you were doing by major industrial categories (manufacturing, services, education, government) by region. For consumer lines the level of visibility was basic demographics (age, sex, marital status) by state. To justify product positioning or to target new niche products, profitability is now measured at sector or segment level.
Sectors, a business concept, may be as broad as the travel industry or as narrow as snowmobile tour operators. Segments, a consumer concept, may be mixed demographic (middle aged, high income women) or psychographic (moderate risk-taking active investors that are heavy consumer spenders but are also Internet-challenged) definitions. These concepts are used to define markets or deploy new products.
The challenge is to calculate profit margin with a high degree of precision at the sector or segment level to see if our bets are paying off. It is not good enough to track only premium revenue. You must carefully correlate all direct loss activity and find the must effective allocation scheme for shared variable and fixed costs. The magic is often in how to apportion these non-direct costs in a manner that does not obscure the real differences that exist. The data warehouse is both a laboratory for sampling, testing, and fine-tuning as well as the definite source for composite results.
The ultimate extension of this concept is to calculate profitability customer-by-customer. Many other industries are well down this path. Banks rank discrete business and consumer portfolios by profitability. Build-to-inventory manufacturers are now evaluating the profitability of their channel partners discretely. The success of several major retailers is based heavily on the inverse of this scenario. They know with precision the profit margin for each and every supplier and often on a deal-by-deal basis.
This is the next frontier for the insurance industry. We need to build a base of data bottom-up from our most detailed policy, premium, claims, and expense data sources to compute customer-level profitability. We need to have a dimensional model that reaches down to the atomic detail to represent a variety of groupings accounts and activities that define each customer. We also need new predictive methods that are effective with discrete data rather than being extrapolations of population statistics. What we need, in the end, is the next generation of data warehousing.
