As insurers increasingly collect “big data” — think petabytes and exabytes — it’s now possible to use new data tools and technologies to mine data across three dimensions:
Large size/long duration — Traditional data mining usually was limited to three to five years of data. Now you can mine data accumulated over decades.
Real-time — With the advent of social media and the different sources, data pours in at ever-increasing speeds.
Variety of types — There’s more variety of data, both structured and unstructured, that are drastically different from each other.
The ability to master the complexities of capturing, processing and organizing big data has led to several data-centric opportunities for carriers.
Big data is playing an increasing role in sales and marketing, and personalization is the hot industry trend. Gathering more information about customers helps insurance companies provide more-personalized products and services. Innovative companies are coming up with new ways to gather more information about customers to personalize their buying experience.
One example is Progressive’s Snapshot device, which tracks how often insureds slam on the brakes and how many miles they drive. It lets insurers provide personalized products based on customers’ driving habits. A device like Snapshot captures information from the car every second, collecting data like how often drivers brake, how quickly they accelerate, driving time, average speed, etc. According to facethefactsusa.org, U.S. drivers log an average of 13,476 miles per year, or 37 miles a day. Big data systems have to process this constant stream of data, coming in every second for however long the user takes to travel 37 miles. Even if only 10% to 15% of customers use the device, it is still a huge amount of data to process. The systems have to process all this information and use predictive models to analyze risks and offer a personalized rate to the user.
People are increasingly using social media to voice their interests, opinions and frustrations, so analyzing social feeds can also help insurance companies better target new customers and respond to existing customers. Using big data, insurers can pinpoint trends, especially of complaints or dissatisfaction with current products and services. Getting ahead of the curve is crucial because bad reviews can spread like wildfire on the web.
The wealth of data now available to insurance companies — from both old and new data sources — offers ways to better predict risks and trends. Big data can be used to analyze decades of information and identify trends and newer dimensions like demographic change and behavioral evolution.
Process improvement and organizational efficiency
Another popular use is for constant improvement of organizational productivity by recording usage patterns of an organization’s internal tools and software. Better understanding of usage trends leads to:
Creation of more useful software that better fits the organization’s needs.
Avoidance of tools that do not have a good return on investment.
Identification of manual tasks that can be automated. For example, logs and usage patterns from tools at the agent’s office are important sources of information for understanding customer preferences and agency efficiency.
Automation of manual processes results in significant savings. But in huge, complex organizations, there are almost always overlapping or multiple instances of similar systems and processes that result in redundancy and increased cost of maintenance. Similarly, inadequate and inefficient systems require manual intervention, resulting in bottlenecks, inflated completion times and, most importantly, increased cost.
Using data from internal systems, systems can study critical usage information of various tools and analyze productivity, throughput and turnaround times across a variety of parameters. This can help managers understand inadequacies of existing systems and identify redundancy.
The same data sources are also used to predict higher and leaner load times, so the infrastructure teams can plan for providing appropriate computing resources during critical events. These measures add up quickly, resulting in significant cost savings and improved office efficiency.
While big data technologies now help perform regular data-mining on a much bigger scale, that’s only the beginning. Technology companies are venturing into the fuzzy world of decision-making via artificial intelligence, and a branch of AI called machine learning has greatly advanced.
Machine learning deals with making computer systems learn constantly from data to progressively make more intelligent decisions. Once a machine-learning system has been trained to use specific pattern-analyzing models, it starts to learn from the data and works to identify trends and patterns that have led to specific decisions in the past. Naturally, when more data — along all of the big data axes — is provided, the system has a much better chance to learn more, make smarter decisions and avoid the need for manual intervention.
The insurance and financial industries pioneered the commercial application of machine learning techniques by creating computational models for risk analysis and premium calculation. They can predict risks and understand the creditworthiness of a customer by analyzing their past data.
While traditional systems dealt with tens of thousands of data records and took days to crunch through a handful of parameters to analyze risks using, for example, a modified Gaussian copula, the same is now possible in a matter of hours, with two major improvements. First, all available data can be analyzed, and second, risk parameters are unlimited.
Machine language technology can use traditional and new data streams to analyze trends and help build models that predict patterns and events with increased accuracy and convert these predictions into opportunities.
Traditional systems generally helped identify reasons for consistent patterns. For example, when analysis of decades of data exposes a consistent trend like an increase in accident reporting during specific periods of the year, results indicated climatic or social causes such as holidays.
With big data and machine learning, predictive analytics now helps create predictions for claims reporting volumes and trends, medical diagnosis for the health insurance industry, new business opportunities and much more.
The insurance industry has always been working to devise new ways to detect fraud. With big data technology, it is now possible to look for fraud detection patterns across multiple aspects of the business, including claims, payments and provider-shopping and detect them fairly quickly.
Machine learning systems can now identify new models and patterns of fraud that previously required manual detection. Fraud detection algorithms have improved tremendously with the power of machine learning. Consequently, near-real-time detection and alerting is now possible with big data. This trend promises to only keep getting better.
These six opportunities are just the tip of the iceberg. The entire insurance industry can achieve precise and targeted marketing of products based on history, preferences and social data from customers and competitors. No piece of data, regardless of form, source or size, is insignificant. With big data technology and machine learning tools and algorithms, combined with the limitless power of the cloud computing platform, possibilities are endless.
Data-driven analysis is a critical decision-making tool for Construction Financial Managers and other industry leaders.
Decision-making is arguably the most important responsibility of company leadership.
Companies that make better decisions make fewer mistakes, and achieve a distinct competitive advantage in the marketplace.
The underlying purpose of benchmarking is to continually improve the quality of organizational decision-making.
As construction risk management consultants, we help contractors prevent accidents, mitigate claims, and reduce the total cost of risk through a continuous improvement process.
We believe companies must instill management accountability for continuous improvement by linking performance measurement to both prevention activities (leading indicators) and operational results (lagging indicators). As the adage goes:
“What gets measured is what gets done.”
In our consulting roles, we frequently help companies establish realistic performance measures by conducting various types of claim and loss analysis.
This type of data analysis is usually the starting point in a performance improvement process — and a common practice among insurance agencies, brokerages, carriers, and risk management consulting firms.
In addition, we are often asked to conduct a benchmarking analysis that compares one company's claim and loss data against peer companies or to the construction industry as a whole.
The term “benchmarking” refers to the comparison of a company's performance results against those of similar peer companies. Benchmarking evolved out of the quality improvement movement in the late 1980s and early 1990s.
Its initial intent was to identify leading companies regardless of industry sector, and apply their best practices to improve one's own company. Over time, benchmarking has become synonymous with process improvement.
The traditional view of benchmarking required two separate disciplines focused on performance improvement: measures and methods. Identifying and capturing performance indicators (the measures) is only the first step; developing and implementing performance improvement (the methods) is the second and most important step for the benchmarking process to be truly effective.
The Health Club Analogy
There is limited value in benchmarking without applying new methods to address continuous performance improvement. Performance improvement requires more than the measurement of performance indicators; it requires the implementation of changes in management disciplines to attain improved operational results.
Using only performance indicators without implementing new methods to improve operations is akin to joining a health club and expecting the benefits without actually using the equipment or committing to an exercise program.
Merely jumping on the scale and gauging your weight relative to others doesn't help you achieve your own weight loss goals anymore than comparing your pulse and respiration rate to others helps you attain your aerobic or cardiovascular fitness goals. What matters most is that a person embarking on a weight loss or fitness program stays committed to the process and monitors his or her own progress.
Similarly, we believe the ongoing monitoring of claim and loss data specific to an individual company is even more important than the initial measurement of insurance claim and loss data relative to other companies.
Baselining As Benchmarking
The term “baselining” refers to the internal benchmarking process that occurs when a company compares its performance against its own results year after year. Ongoing, internal monitoring allows a contractor to determine if the company's claim and loss trends are improving or deteriorating, and to make the critical performance improvement decisions necessary to facilitate a change in results.
Referring back to the health club analogy, baselining does not compare an individual's weight and aerobic fitness to that of the other health club members. Instead, individual fitness goals and measures are established, monitored, and tracked to verify continuous personal improvement.
Similarly, a construction company can develop a baseline analysis of its loss cost performance by reviewing loss and claim data for a minimum of 3-5 years. Company results are compared from year to year, and ideally are broken down by operating entity, division, project, manager, or even crew levels.
Exhibit 1 provides a sample of a baseline analysis that compares one company's relative claim and loss performance within all of its operating divisions.
This analysis reviews the historical loss cost data for the entire company and breaks it down into meaningful data relative to each operating division. The total workers' comp, Comprehensive General Liability, and auto liability incurred claim costs (sum of paid and reserves) for each company division over a five-year period were compared to the total man-hours for each division, producing a cost per man-hour figure.
The results illustrate dramatic differences in total claim costs per man-hour for each division. This baseline analysis was the first step in raising awareness of the predominant loss leaders within the company. This increased awareness led to a detailed analysis that established plans of action and realistic cost targets by company division for the upcoming year.
We acknowledge that there are numerous benefits to measuring the frequency, type, and cost of insurance claims compared to peer groups and/or the entire construction industry. Such analyses provide the ability to:
Identify leading types and sources of claims
Establish strategic objectives to prevent the occurrence of common industry claims
Create awareness among managers and employees about the costs of claims and the impact on profitability
Post positive results on company websites and for use in other marketing materials
The Bureau of Labor Statistics provides safety-related data so that companies can externally benchmark injury and illness data against specific industry groups. (Check out the Web Resources section at the end of this article for more information.)
In addition, Bureau of Labor Statistics data is used to calculate and compare OSHA Recordable Incident Rates and Lost Workday Incident Rates, both of which are common construction industry benchmarks. This data is useful when making high-level comparisons within construction industry segments relative to injury and illness rates.
We also use external benchmarking analyses to establish risk reduction, loss prevention, or cost containment goals. In “Risk Performance Metrics” by Calvin E. Beyer in the September/October 2007 issue of Building Profits, a sample benchmarking comparison shows a representative contractor's duration of lost workdays workers' comp cases in median number of days compared against the median duration for the industry. Results such as these can highlight the importance of an increased focus on injury management and return-to-work programs.
The benchmarking analysis in Exhibits 2A and 2B compares a contractor's workers' comp claim and loss performance to an established group of peer contractors in the same specialty trade. (These companies engaged in similar work, and performed in states with similar insurance laws and legal climates.)
The analysis was based on total incurred workers' comp costs and total number of workers' comp claims as compared to payroll for each entity. Overall, Company D had worse results than the other three companies.
This prompted an in-depth review of Company D's workers' comp losses by division and occupation. As shown in Exhibit 3, the company experienced significant claim frequency and severity issues within the first six months of employment.
These findings triggered the development and implementation of specific activities designed for Company D's new employees.
Below are some of the activities that were incorporated into the formal improvement plan:
new hire skills assessments
daily planning meetings
Other Sources Of Benchmarking Data
Professional associations and industry trade/peer groups also provide comparative data for benchmarking purposes.
The Construction Financial Management Association's Construction Industry Annual Financial Survey is an excellent source for understanding the key drivers of contractor profitability. We use the survey data to determine comparative profit margins for different types and classes of contractors when we calculate a revenue replacement analysis to show the additional sales volume needed to offset the cost of insurance claims. (This technique was highlighted in the “Risk Performance Metrics” article previously mentioned.)
Similarly, the Risk and Insurance Management Society (RIMS) conducts an annual benchmarking survey that reviews insurance rates, program coverages, and measures of total cost of risk.
An example of a peer group data source for benchmarking is the Construction Industry Institute (CII). The Construction Industry Institute is a voluntary “consortium of more than 100 leading owner, engineering-contractor, and supplier firms from both the public and private arenas” (www.construction-institute.org). It develops industry best practices and maintains a benchmarking and metrics database for its participating members.
Another peer group example involves members of captive insurance companies sharing and comparing claim and loss data for the group as a whole. There is a major advantage when a true peer group shares benchmarking data: Such data sharing often leads to peer pressure in the form of increased ownership and accountability for improvement by the companies shown to be the poorest performing members.
We continue to search for more new sources of industry best practices and comparator data. A possible emerging source for the construction industry is the National Business Group on Health. This organization has developed standardized metrics known as Employer Measures of Productivity, Absence and Quality™ (EMPAQ®).
EMPAQ® helps member companies gauge the effectiveness of their injury and absence management and return-to-work programs. The founder and principal of HDM Solutions, Maria Henderson, served as a project sponsor for EMPAQ® from 2003-2007, and co-presented with Calvin E. Beyer on “Return to Work as a Workforce Development Strategy” at CFMA's 2008 Annual Conference & Exhibition in Orlando, Florida.
Limitations Of External Benchmarking
We fear that the increasing popularity of external benchmarking analyses may indicate that it has become a “quick fix” solution or a management fad. When asked to conduct an external benchmarking analysis, we always ask the following questions:
What is your purpose in seeking these comparisons with other companies?
Who are you trying to convince and what are you trying to convince them to do?
What specific peer companies should be used for comparative purposes?
Are these companies (and their operations and exposures) truly similar enough for a fair comparison?
Beware Of Pitfalls
There are many hurdles to surmount in locating suitable companies for external benchmarking comparisons. Generally, when benchmarking comparisons can be made, more often than not the greatest value lies in the workers' comp line of insurance coverage.
Here are some key factors to consider when choosing contractors for external benchmarking comparisons:
Percent of self-performed work vs. subcontracted work
Payroll class codes and hazard groupings of selfperformed work
Differential geographic labor wage rates
Payroll rate variances between union and merit shop operations
Size of insurance deductibles
Claim reporting practices
For example, claim reporting practices must be similar in order to minimize distorting the frequency or average cost of a claim. If one or more comparison companies self-administers minor claims or does not report all claims to their carrier, using carrier loss reports for the comparison is an invalid method.
We also find that comparing the frequency of claims and total loss dollars divided by thousands or millions of dollars of payroll (exposure basis) is a helpful workers' comp benchmark between companies of similar operations in similar states.
Likewise, a suitable benchmark for auto liability performance compares the frequency of claims and total loss dollars per one hundred vehicles.
When benchmarking fleet-related claims, ensure that the number and size of fleet vehicles — as well as the type of driving (urban vs. rural) and the total number of miles driven annually — are similar among the contractors whose claims are being compared.
Benchmarking comparisons of Comprehensive General Liability insurance results are especially challenging due to delays in reporting third-party bodily injury and property damage claims, in addition to the expected long tail of loss development for these claims.
All of these factors are compounded by vastly different litigation trends and liability settlements in various states and regions of the country.
Common Limitations Of Data Sources
Whether or not you intend to develop a baseline of your company's claim data or to benchmark your company's performance against a peer company, there are several issues that must be successfully resolved regarding the data's quality and integrity.
Based on our experience, we classify the key challenges associated with exposure and claim/loss data into the categories shown in Exhibit 4: availability, accuracy, accessibility, standardization, reliability, comparability, and date-related problems.
Value Of Multiple Measures
Evaluating data from various sources and different angles is also valuable. Why? Because it's possible to gain a better understanding of the whole by dissecting the parts. This practice illustrates the principle of multiple measures.
This approach is substantiated by 2006 research, which concluded that the “simultaneous consideration” of frequency and severity provides a more comprehensive result than performing analysis based solely on one factor.1
This is similar to our approach when we conduct a “Claim to Exposure Analysis” and review historical frequency and severity vs. the relative bases of exposure for each line of casualty insurance coverage.
Returning to the health club analogy, when starting a formal exercise program, you often begin with such general baseline measurements as height and weight; this is usually followed by additional measurements, such as BMI, body fat content, and the girth of arms, legs, and chest (the baseline).
As we all know, weight alone is not always the best indicator of success in fitness efforts. In fact, since muscle weighs more than fat, an increase in total body weight may actually occur after beginning and maintaining a fitness program.
Although you might not experience a dramatic weight drop, you could see a reduction in waist size and BMI — positive changes that would not be evident unless multiple measures were being used and reviewed.
Benchmarking insurance claim and loss data performance is like comparing one person's height and weight against the ideal height and weight charts based on the entire population.
Wouldn't it be more effective to establish your baseline weight and other multiple measures initially so you can see the progress you are making?
This is similar to the baseline measurements that a company should take (as well as the multiple measures) that are necessary to meet your company's performance improvement goals for financial success, operational excellence, or risk reduction.
Cal Beyer collaborated with Greg Stefan in writing this article. Greg is Assistant Vice President, Construction Risk Control Solutions, at Arch Insurance Group. As a member of the Southeast Regional team in Atlanta, GA, Greg supports underwriting and claims in risk selection, claim mitigation, and risk improvement activities. He is also responsible for high-risk liability risk reduction initiatives including contractual risk transfer, construction defect prevention, and work zone liability management.
1 Baradan, Selim, and Usmen, Mumtaz A., “Comparative Injury and Fatality Risk Analysis of Building Trades,” Journal of Construction Engineering and Management, May 2006, pp. 533-539.