News & Insights
EN ▾

Call the right people

Traditional quality survey of health insurance is not very effective for detecting and proving fraud

What is fraud and how to deal with it
Detection of "added" services using big data

The article continues the series of posts about fraud in health insurance:

All health insurance companies conduct surveys of insureds on the quality of provided medicl services, but such surveys in their current form are completely unsuitable for the purpose of fraud detection, despite the fact that feedback is almost guaranteed evidence.
publication date:
November 6, 2023
Health Insurance
Health Insurance
Examples of reactions to «added» services and visits are usually very vivid, especially in personal oral communication:

We would like to know how the visits to the therapist on the 10th, 17th, and 24th of the month went at the X clinic?

This is strange. I only saw the therapist once. He examined me, gave recommendations, and that's it. There were no follow-up visits! He didn't even prescribe any tests for me, although I was expecting it, since I had a respiratory infection, and I usually get blood tests. In general, everything was formal, and I didn't like it.

Such examples highlight problematic cases very well and are convenient to use for dialogue with the clinic. The question remains: how to get this feedback? The existing survey system is not the best practice because:

- It is expensive (involves trained staff, processing a huge amount of unstructured data, etc.).
- It annoys insureds (fatigue from incoming calls, more and more people don't answer unexpected calls).
- It is not informative (often the insureds does not remember what services were provided to them, how many times they visited the clinic).

Moreover, let's be honest, from the insured's point of view, such a call is no different from spam. And since the overwhelming majority of cases do not involve fraud, the majority of responses will be useless from this point of view.

One of the solutions is feedback through apps for assureds. This is an effective tool for the insured to control the services provided, as after visiting a doctor, they receive a breakdown of medical services and can identify services that they did not receive. Most commonly, despite the obvious convenience, personal accounts are used by no more than 1% of insureds, and there is no trend towards an increase in usage.

Another effective and accessible option is to identify suspicious cases using machine learning models and big data analysis to select only those insureds whose cases have the highest probability of fraud, and thus reduce the number of calls by an order of magnitude.

Additionally, it is advisable to mark those insureds whose cases have a significant probability of fraud, but no feedback has been received. In this case, feedback can be obtained when the insured contacts the insurance company on their own, for example, to make an appointment or receive a consultation. Then, after successfully resolving the issue, the satisfied insured can be transferred to a special operator, assuming that a level of trust and goodwill will allow them to obtain the maximum amount of reliable and necessary data.

In general, feedback for fraud detection can be implemented as follows:

collecting feedback from insureds from the previous sample.

collecting feedback when the insured contacts the insurance company on their own

step 1

step 2

step 3

Sampling of insureds:

selection the cases with a high probability of fraud.

(for example, to make an appointment or receive documents).

Active contact:

Response contact:

To collect evidence of fraud, it is sufficient to survey only about 1% of all insured persons:
  • only the most suspicious cases are selected;
  • fraud is a recurring pattern of behavior, that’s why confirmation of individual examples of a pattern is sufficient to consider it fraudulent in other cases.
Below is a diagram illustrating the positive effect of using ML fraud detection tools to obtain feedback.
Figure 1. Randomly selecting cases (only ~1-2% of the investigated cases may contain fraud) can be effective in detecting fraud.
randomly selecting cases for fraud testing
case without fraud
case containing fraud
Figure 2. Fraud detection using ML-tools (~50% of the investigated cases contain fraud with high probability)
selecting cases with a high probability of containing fraud
Traditional quality survey of health insurance is not very effective for detecting and proving fraud.

MAINS LAB solutions, which use machine learning, artificial intelligence algorithms and natural language processing, identify of fraudulent cases to obtain quality feedback especifically from insured persons who have become victims of unscrupulous clinics, and whose experience is most valuable in the combating fraud.

By surveying just 1% of the insured persons, you can be confident in the extent of fraud present in your accounts and take further decisive action.