Tag Archives: bart

Predictive Analytics for Self-Insureds

Predictive analytics is widely used in the insurance industry. Is it time for self-insureds to reap the benefits of predictive analytics and realize significant bottom-line improvements as well?

Self-insureds can use predictive analytics for employee cost benchmarking, early identification of late-developing claims, and budget and allocation decision-making tools. Currently, the key area of focus for self-insured risk managers is claim prevention. But even the best claim prevention methods are not enough to avoid all claims. Claims occur despite a company’s and risk manager’s best efforts. If traditional claim prevention is the only defense against losses, these companies will lag behind their contemporaries who, to keep claim costs down, are already using predictive analytics.
Primer on Predictive Analytics
Whether we know it or not, we have all encountered predictive analytics in our personal lives by simply browsing the Internet. Companies like Amazon leverage their immense amount of data to predict customers’ shopping preferences to drive additional sales. Likewise, the insurance industry, with its substantial volume of data, views predictive analytics as an essential capability. Forward-thinking self-insureds see the benefits of these tools and realize that they should be used to better understand their exposure and better control their losses.

Predictive analytics can improve both customer satisfaction and company profits. Last week, I bought a webcam from Amazon. The product page displays the specific details on the webcam and shows an assortment of additional items frequently bought by other customers who purchased this webcam. Amazon uses its customer purchase data to predict that webcam purchasers also routinely buy extension cables, microphones, and speakers along with their webcam. Thus, Amazon has effectively applied predictive analytics to identify cross-sell opportunities that benefit its customers (I was glad to be reminded that I would, in fact, need an extension cable) and increase its revenue (I spent an additional $5.99).

Amazon’s product pages are an easily visible example of predictive analytics. Similar to Amazon, the insurance industry has adapted predictive analytics to not only increase premiums but also to improve risk selection. Risk selection—being able to identify the good/profitable risks and the bad/unprofitable risks—is so important for insurance companies that the popularity of predictive analytics comes as no surprise. The range of ways insurance companies are using these tools includes evaluating the profitability of their accounts, assisting in effective underwriting and proper pricing, marketing to the appropriate client base, retaining their customers, and estimating the lifetime value of each customer.

It is important to note that predictive analytics is not just a data summary. Predictive analytics sorts through vast amounts of data to find relationships among variables to predict future outcomes. This data can often be seemingly disparate, or even appear to be unrelated; it also often combines the use of both internal company and data from outside the organization. For example, a commonly cited and successful example of such a relationship involved insurers finding a direct correlation between two variables: credit scores and auto claims. Also, the more data fed into an analysis, the more robust it will be. Self-insureds have quite a bit of employee data that greatly aids analyses’ predictive abilities. In addition to loss-specific data, payroll, human resources information, and other available third-party data should be utilized to its fullest extent for optimal results.

Applications for Self-Insureds
Self-insureds have numerous opportunities to benefit from predictive analytics. Three large workers’ compensation areas on which predictive analytics sheds light are: 1) identifying which employees cost more than industry and company averages, 2) predicting early on which claims are the most likely to have late-developing costs, and 3) constructing qualitative cost/benefit scenarios to help risk managers allocate their budgets effectively.

Self-insureds applying predictive analytics to their workers’ compensation claims, for example, have a number of employee variables to work with such as: employee age, length of employment, state of residence, employment type (full time/part time), salary type (hourly/salary, low wage/high wage), and claim type (indemnity/medical only). Predictive analytics can find relationships that will affect future claim activity on current and future employees.

The real goal of predictive analytics for self-insureds is to help guide and support risk managers’ decisions. Predictive analytics can be applied to both ‘pre-claim’ and ‘post-claim’ loss prevention methods. To aid in claim prevention, pre-claim-focused analyses are used to highlight high-risk (high-cost) employees. The loss costs of various groups are compared to each other, a company average, and to average industry loss costs provided by the National Council on Compensation Insurance (NCCI). Loss categories higher than company or NCCI averages get closer examinations for loss drivers and mitigation strategies. A higher-than-average cost for newly hired employees may signal a need for more training. A higher-than-industry cost for claims in certain states may be noted. A particular type of injury may emerge as the most costly. The potential savings can be estimated as the difference between the current loss costs and the benchmark loss costs, times the percentage of employees or expected claims involved.

For example, a self-insured entity may know that its newly hired employees experience a larger proportion of losses than employees with longer tenure. If it conducts an analysis and discovers that low-wage employees working in Illinois with less than six months of experience have substantially higher costs than the average employee, claim prevention resources could be specifically aimed at that employee demographic to control costs.

Savings Opportunities
A notable benefit of predictive analytics is that it provides quantitative cost-saving information to risk managers. Continuing with the prior example, assume 2,500 employees are newly hired, low-wage employees in Illinois and their average costs have been shown to be three times higher than the company average of USD $1.50/$100 of payroll. We can estimate that a reduction from $4.50 to $1.50 could create $2.25 million in savings. Asking senior management for $100,000 for more new hire training in Illinois facilities will be much easier with the quantitative support provided by predictive analytics.

(2,500 employees with an average payroll of $30,000 save $3 = 2,500 x 30,000/100 x 3 = $2.25M)

Not only can predictive analytics assist with reducing cost ‘pre-claim’ by focusing on exposure, it can also reduce costs once a claim has occurred. Knowing the easy-to-identify large claims will be second nature to risk managers, however, ‘post-claim’ predictive analytics can look into claim development details to find characteristics that late-developing, problematic claims (and often not the obvious large ones) have in common. After a loss has occurred, one of the most effective ways to manage costs is to involve a very experienced claims handler as soon as possible. The results of effective ‘post-claim’ predictive analyses will assist in implementing cost-saving claims triage. Because the best resource post-claim is good claim management, predictive analytics can get late-developing, problematic claims the timely attention they need to contain the ultimate costs or even settle the claim.

Loss savings based on predictive analyses extend beyond claim cost reduction. Being able to quantitatively show potential savings and concrete mitigation plans will make a positive impression on senior company management and excess insurance carriers. Demonstrating shrewd knowledge of the loss drivers and material plans to reduce the losses can aid in premium negotiations with excess carriers for all future policy years. And if the insurer or state is holding any collateral, the predictive analytics’ results can be used by the self-insured in negotiating.

The key to unlocking further potential cost savings in your self-insured plan is readily available in your own data. Predictive analytics is the tool that will help risk managers make better claim reduction decisions and produce actionable items with real cost savings now and in the future. Risk managers and self-insured companies can look forward to possible benefits such as loss cost reductions along with reductions in excess premium and collateral, and quantitative information to help them with budgeting and allocation. As more self-insureds begin applying predictive analytics to control costs, companies that are not using these tools will be at a competitive disadvantage.

For more information on predictive analytics, watch this video of a Google hangout with Michael Paczolt and Terry Wade of Milliman, Inc.:

Ever-Increasing Unpaid Claim Liabilities: When Does The Growth Stop?

Large deductible and self-insured workers' compensation programs often face year-over-year increases in their unpaid claim liabilities, even though they are not suffering from adverse claim development, unexpected large losses, or materially changing their risk profile. Why does this keep happening? The increase is the natural result of several external factors, with the biggest drivers being rising indemnity costs and rising medical costs. Why should program managers expect these increases? And how can the inevitability of these rising costs be conveyed to executive management?

The solution starts with recognizing the nature of liability growth and focusing not solely on liabilities but rather on ultimate losses.

How Actuaries Determine Unpaid Claim Liabilities
Actuaries estimate a program's ultimate losses — the amount that will eventually be paid to close all claims that have occurred under the program. For a review as of December 31, 2012, the ultimate losses include all claim payments, current case reserves, and an incurred but not reported (IBNR) provision for any future case reserve development (including reopened claims) and for claims that have occurred but have not been reported as of December 31.

There is one important (non-numerical!) formula that relates to ultimate losses and the resulting reserves called unpaid claim liabilities. The ultimate losses as described above can be shown as follows:

Ultimate Losses

The actuarial report will estimate the total amount to be paid (i.e., ultimate loss) and we know the total amount paid to date from the claim data. The program's remaining obligation is what is left to be paid: the known case reserves and the additional estimated IBNR provision — that is, the remainder of the ultimate loss amount.

Utilizing the estimated ultimate loss and the actual paid losses, we can arrive at the estimated unpaid claim liability:

Estimated Unpaid Claim Liability

Just as the unpaid claim liability is the ultimate losses minus the paid losses, the change in the unpaid claim liability from one evaluation to the next is the change in ultimate losses minus the change in paid losses. The key to understanding the change in the unpaid claim liabilities is to understand the drivers behind the change in the ultimate losses relative to the change in paid amounts. Because, in an ongoing program, both will theoretically increase at each subsequent evaluation. The amount by which the change in ultimate losses outpaces the change in paid amounts will determine the change in the unpaid claim liability.

When unpaid claim liabilities are increasing, it is really because the ultimate loss estimates for each new accident year are increasing faster than the losses are paid during the calendar year. So what causes the ultimate loss estimates to increase?

Many risk managers are concerned that executive management will perceive increasing ultimate losses and increasing unpaid claim liabilities as a reflection of an underperforming current risk management program as well as claim management practices. However, there are many factors completely unrelated to risk management that influence the unpaid claim liabilities.

Changing risk management practices such as claim reserving methods and claim payment speed, or shifting the type of work performed or injuries incurred, will influence actuarial studies, but we're going to focus on external trends that influence programs and cannot be controlled by risk management. These factors, even with keeping all aspects of the program's risk management consistent from year to year, cause the natural tendency of ultimate losses to grow each year.

An Illustrative Example — Without Trend
While actuaries and risk managers are fluent in the actuarial terms of ultimate losses and unpaid losses as described so far, when describing the cash flows to executive management at a high level it may be easier to think in terms of buying a house and making annual payments on it. If a $250,000 house is purchased, it's clear that ultimately the full $250,000 will be paid. Until then, the remaining unpaid portion is equivalent to the actuarial unpaid claim liability.

Let's say a $250,000 house is purchased and $50,000 payments are made every year for five years to pay it off. An additional house is purchased every year. The first house is purchased in 2012 and, at the end of the year, $50,000 is paid. The unpaid amount is $200,000 ($250,000 – $50,000). In 2013, another house is purchased and a $50,000 payment will be made on each house this year.

At the end of 2013, we know that ultimately $500,000 for both houses is owed and that $50,000 was paid in 2012 for house #1 and $100,000 in 2013 for both houses. The unpaid liability as of December 31, 2013, the remainder left to be paid, is $350,000 ($500,000 – $150,000) — which is an increase of 75% over the prior year end.

Unpaid Liabilities

If the same thing is done in 2013 as 2012 (buying a $250,000 house and making $50,000 payments on each), why does the liability go up? It's because the change in ultimate losses (another $250,000) is much larger than the additional $100,000 payments in 2013. In 2013, the ultimate increased by $250,000 for the new year's exposure of an additional house and the payments increased by $100,000 for the two $50,000 payments made on the two houses. The prior liability of $200,000 plus the change in ultimate losses of $250,000 minus the change in payments of $100,000 equals a new liability of $350,000 at the end of 2013.

For the next four years, the liabilities will keep increasing. The liability will go unchanged only when the payments made in the calendar year are equal to the new exposure brought on with the purchase of a new house. When five (or more) $250,000 houses are owned, $250,000 in payments will be made each year, offsetting the additional $250,000 in liability picked up with each new house purchased, as shown in Figure 2 below.

Liabilities with Multiple Houses - No Trend in Housing Cost

Many executive managers (and some risk managers) feel that their programs are in this “steady state” and do not expect to see increases in the liability. However, in the real world of large deductible and self-insured workers' compensation programs, the ultimate losses for each new exposure year are heavily influenced (and increased) by rising medical and indemnity costs. The injuries that cost $250,000 in 2012 dollars will cost more in the future.

An Annual Occurrence?
It is very important that executive management is aware that unpaid workers' compensation claim liabilities are expected to increase each year. And that the increases will be more than inflation (as represented by the consumer price index, or CPI). While other balance sheet items may be subject to inflation only, unpaid claim liabilities are subject to inflation, severity trends, and frequency trends. The 2012 State of the Line annual presentation by the National Council on Compensation Insurance (NCCI) shows a five-year indemnity trend of 2.9% from 2006 to its projection for 2011, with a medical trend of 4.4%, noting that the negative frequency trends that had been offsetting the rising indemnity and medical severity trends are beginning to flatten out.

This means that as business continues into the future, and even if all other variables remain the same, each new year of additional workers' compensation exposure is expected to cost more than the prior year by the aggregate trend. This is analogous to buying the same house every year in a housing market that's rising.

In this next example, the houses bought are 3% more expensive every year. All the houses have similar characteristics from year to year — square footage, tax rates, school districts, etc. — just as from year to year, many companies' workers' compensation exposure stays similar, as far as the number of employees, the mix of NCCI class codes, the location of employees, and the retention limit of the insurance program.

Adding Trend
Continuing from Figures 1 and 2, if a 3% housing increase trend is used, the change in liability will now increase each year and will continue to increase by the 3% trend. In our first example, a constant liability was achieved because the total calendar year payments ($50,000 on each of the five houses with a mortgage at any one time) lowered the liability the same amount as the new house raised the liability each year.

In the 3% trend example below, payments are made on less expensive houses while the incoming houses are the most expensive. In 2018, the first trended house is purchased for $257,500 ($250,000 × 1.03). The first of five payments is made on this house for $51,500. Each of the other four outstanding mortgages is still being paid at $50,000 each, so in 2018 a total of $251,500 is paid. But this is $6,000 short of the $257,500 change made to the ultimates when the new house was purchased.

Recognizing that the majority of payments are being made on mortgages (or claims) in earlier, “cheaper” years, while more expensive, trend-influenced mortgages/claims are being brought onto the program's liabilities, is the key to accepting the natural rise in liabilities over time.

Long-Term Liabilities with Multiple Houses - 3% Annual Trend in Housing Cost

Large deductible and self-insured workers' compensation programs face liability increases that are due to cost trends every year. In order for a program's liabilities to stay at the same level or decrease, the program would have to overcome these trends. But inflation, severity, and frequency trends aren't the only variables out there.

What To Watch For In The Future
As the economy recovers and workforces grow, this will increase the future additional costs applied to the program's liabilities. As a workforce increases by 1%, 3%, or 5%, so too do the anticipated ultimate losses for the next year. It would be comparable to buying a house with 1%, 3%, or 5% more square footage at the same cost per square foot.

Additionally, as the insurance market slowly hardens, programs may begin raising their retention levels. As the retention levels rise, so do the oncoming liabilities. Retention level increases will have a much larger impact than trend increases, as retention levels can increase by substantial amounts. In this example, a workers' compensation retention increase from $250,000 to $500,000 is similar to suddenly buying more expensive houses. While a 3% trend will raise the house price from $250,000 to almost $300,000 over six years (and raise liabilities by 3% each year), a jump to buying $500,000 houses will noticeably increase the liability with the addition of $500,000 mortgages, while most payments on the houses are in the range of $250,000 to $300,000.

In these examples, a very even payment pattern was used, but it is also important to note that the timing of payments influences the unpaid claim liabilities. In real life, workers' compensation payments may not be this regular, and large claims can have sudden payouts due to judgments or settlements.

As shown in the earlier formula, unpaid claim liabilities are equal to the ultimate losses minus the paid amount to date. Let's say that a decision is made to fully pay one of the mortgages on these houses. An additional payment of $250,000 would decrease the liabilities by $250,000. And while this large payment will decrease the liability, it does not have an effect on what is paid in total — the ultimate payments will remain the same — they just have to be paid sooner. And the decrease in the liabilities came from an equal decrease in the assets, which will offset on the financial statements.

Your actuary can explain and quantify the trend effects on your insurance program in detail and how any other changes in your program (to retention levels, employment levels, etc.) will affect the liabilities going forward. Internally, the actuarial report should be reviewed for its information on the change in the program's ultimate losses, which is a better reflection of the program's performance over time than the change in liabilities over time that is due to their sensitivity to payment timing.

Risk managers who need to manage executive management's expectations on unpaid claim liability growth can explain the external factors that play a large role in changing liabilities over time. If liabilities are increasing with the increasing trends, that's to be expected. If liabilities are increasing less than trend, staying flat, or decreasing, that's great. If liabilities are increasing by more than trend, ask your actuary for some suggestions.

A safety analysis can help identify which kinds of injuries make up the majority of the loss dollars, the program's retentions and allocation methodology can be reviewed, or for a thorough review of claims, predictive analysis can zero in on loss drivers.

An actuary can't make external trends disappear but can help explain them to executive management and make other suggestions on finding solutions in your data.