In Insurance service various risks are covered under terms and conditions mentioned in Insurance Policies and in case of claims arising, they are managed or settled accordingly. It is a common practice to audit the procedures for underwriting, policy management, claims management etc. to ensure that healthy practices are followed by the organization. Since the number of policies and claims are usually very high, it is undesirable to audit each and every record. Sampling techniques are used to reduce the volume of audit work without compromising on quality.

Sampling techniques have been in use in various areas. Pathologists take a sample of blood from a human body and test that sample to get an idea about all of the blood. Similarly, while boiling rice, we test a few grains of rice as a sample to conclude about the entire lot of rice. We know that this technique takes us to the right conclusion. Properties of blood are uniform for pathological tests being conducted. Rice being boiled too is uniformly heated by keeping the size of the container and volume of water appropriately large. Due to this uniformity, a small sample size is sufficient for reaching to the right conclusion. If the rice is boiled with lesser water than what is required for uniform flow of heat, the rice doesn’t get cooked uniformly. In such cases, testing few grains from top and forming opinion about the entire lot may lead to wrong conclusion. Similarly, if the items being cooked together are not similar then again we need more samples to draw any acceptable conclusion.

In Insurance industry, the data related to various functions have variations because of its basic nature of dealing with uncertainties. Keeping the sample size large for auditing increases the cost of audit. On the other hand keeping it low may lead to invalid or wrong conclusion. Hence, it is important to choose correct sampling technique and sample size.

Using sampling techniques effectively is very easy. But it is not that easy that we pre-decide the sample size without knowing the nature of data. Sample size depends on following four factors:

1.       Size of the population: If we want to draw conclusion about claims management system, the population consists of all the claims; paid or denied or pending; that became part of the system. Larger the population, larger is the sample size if rest of the factors remain unchanged.

2.       Variance: This is a measure of variation in the data. If there is no variation then by checking any one item we can form accurate opinion about the entire lot of item. But, if variations are there, we need to check more. Larger the variation, larger is the sample size. Variance and standard deviation both are measures of variation in the data and represent the same property of the data.

3.       Confidence Level required: Suppose we audit 10 claims cases and find that in 3 cases a business rule was not taken care. Further, in 20 audited claims cases we find this number to be 6. We may conclude that in 30% of the cases this rule has been ignored. But, how much confident we would be on this conclusion? If we audit more cases and find similar results, the confidence level on the conclusion will be higher. So, If the confidence level required is high, the sample size required too is high.

4.       Accuracy: Higher the accuracy required, bigger the sample size should be.

As it is clear that a sampling technique and the sample size cannot be decided without proper understanding of the nature of data. An auditor cannot do anything with the population data and variance in that data because these data are generated through various operational activities performed in the organization. However, if a very high level of accuracy and confidence level is set, the sample size required becomes very high. It is recommended to keep these values realistic to keep the cost and time acceptable.

A good practice is to do a small initial study to understand the population size, variance and basic nature of the data. Then use them to compute the sample size and conduct the detailed study. While picking the records for audit, the sample should represent the population without any bias. For this, generating random numbers and associating them with actual records by using the statistical distribution of data gives the best result.