Introduction to Insurance
Insurance, Due to its business nature is very prone to Frauds. Insurance is the distribution of Risk, which requires Insurance Companies to have liquid assets in terms of reserve funds available to pay loss claims. Fraud Analytics is used in the entire lifecycle of Insurance.
What is Fraud in general?
Fraud is basically the misrepresentation of Material Facts for wrongful gain or wrongful losses for others. There mainly 3 factor which motivates the person to commit frauds –
- Rationalization – Justification of Dishonest Actions
- Opportunity – Ability to execute Plan without being caught
- Pressure – Financial or Emotional force pushing towards Frauds
Effect of Frauds on Insurance Industry
Cost of Insurance is steadily increasing due to losses that the Insurance Industry is incurring. According to the Federal Bureau of Investigation, The total cost of Insurance Frauds (Non-Health) is estimated to be around $40 Billion per year. Which means average US citizen has to pay $400 to $700 per year added to their actual cost of the premium.
Insurance companies are more likely to be dependant on Regulatory Authorities to tackle Frauds, it is mainly due to lack of expertise to carry out investigations or sometimes it is not economical to investigate. Insurance frauds to be addressed proactively Insurer needs to train their staff in identifying Red flags at the issuance stage only.
Major Types of Insurance Frauds
An agent or Salesperson collects the premium from the customer but doesn’t submit the premium amount to Insurer.
An Agent or Internal Employee can change beneficiary bank account details to someone who is connected to them so that all the claim amount can be taken by Agent/Sales Manager with the help of internal employees.
Fake Death Claims
In Life insurance, Fake death claims are reported, where the fake Death certificate is submitted to the Insurer to get the claim amount. These kinds of activities are performed by Nexus or Cartels.
Fictitious Policies or Third Party Policies
To meet Monthly target Agents/ Sales Managers submits Fake Policies or policies purchased by the Third Party. According to Insurance Regulator, it is mandatory for Agents or Sales Managers to meet the Life Assured while Insuring their lives at their residence.
Car Damage Fraud
The customer had some minor accident or damage to their car, they will present that incident as a car accident to the insurer and try to get maximum claims for the damage. In this case, the investigator might be involved to present a false investigation report.
Health Insurance Billing Fraud
Customer goes for routine health check-ups, but suddenly Medical Professional realizes that you have Health insurance under which they can charge you the hefty amount for an unnecessary surgical process that never happened. The customer wouldn’t know about these charges and becomes the victim of these frauds.
These are some common type of Fraudulent Insurance Practices performed by people, which affect the insurance industry adversely and to adjust these losses, Insurance Companies increase their product premiums.
How Data Science Can Help In Identifying Frauds in Insurance
To Large extent, Big Data Analytics is being driven by IT teams, not Core departments of the business. Data Analytics often introduced on the project basis and, if the benefit is shown then it is extended to other streams of the business. The insurer may implement these techniques and tools in Marketing, Servicing and Customer retention areas first but Fraud detection and prevention is benefitted with analytics as much.
Knowing what all data is available is most important part of the any Data Analysis process, without this step your Data Analysis might be missing something important. In today’s date Insurers are collecting all possible data points regarding the Policy Holders, but due to structure and Nature of the data point, the Data Storage is scattered over different servers and Different formats. Collating and Creating A Single Data Warehouse with the help of Big data Technology is the key to Data Management and Data Governance.
Single Data Repository With Shared Data Between Multiple Insurance Companies
Generally, Fraudsters try to attack easy to go Insurer and they try to use same formula every other insurer they can because it is effortless for them. So Insurer needs to come together to identify these coordinated frauds. There are a couple of organization ( Experian, LexisNexis etc.) trying to bring the necessary data together on the single platform so that identification is easy.
Fraud detection at Underwriting/POS stage
A Machine Learning and AI-based Risk Scoring model is implemented to Score the policies at POS level, Red–Amber–Green Codes can be given to risk scoring so that Red Category Policies can be Rejected at the POS stage only.
Amber category policies can go for underwriting and then Investigation and then the decision can be taken after investigations. Green Category can be directly issued.
Agent Profiling / Branch Profiling for Finding Patterns for Claims
Agent profile can be identified with the help of Segmentation methods. So insurer can identify which are the common characteristics of Agents to make frauds.
Later a Classification Machine Learning Model can be implemented at Recruitment Stage where Insurer gets the flags which agents are like to be involved in Fraudulent activities and be cautious about them.
Geography Level Analysis for Claims
There could be any particular region or State from where Insurer is receiving the high number of Claims and Most of them are fraudulent. So Basic study can be conducted every quarter and these locations can be identified for Investigating claims or policies.
Fraud Detection at Claims Stage
According to SAS research, 1 out of 10 Insurance claims is Fraud Claim. How do Insurer spot suspicious claims before huge claims are paid to fraudsters? In the current scenario, Insurers are solely relying on traditional Rule Based system, but frauds are dynamics in nature. Here to control the Fraudulent Claims, Data Science can come into the picture.
Predictive Models can help the Insurer to identify claims which are likely to be fraudulent. A Scoring Model can be used in reverse was as well to Streamline Claims, low-risk claims can be directly processed without much investigation and assessment; this will help insurer save time and focus on High-Risk claims.
Combining Insurers’ Internal data with External Data (Social Media footprints and Demographics Information) to create the customer profile for assessment.
Image Processing and Machine Learning techniques together can do wonders in reading a claims documents and segregate Red, Amber and Green type of Documents. With this techniques, Document Forgery can be reduced significantly.
Now let’s understand how to understand the early High-Risk signs at the policy underwriting stage; Download the dataset used in this case study
To understand the Early Signs of Claim, let’s study the claims received by an insurance company(Specimen Data). We will study the characteristics of the customers who are likely to be fraudulent.
Let’s understand the claims received by Each Variable and let’s try to identify common factor between Frauds.
V1. Product Distribution
So below graphs show that 52.7% Claims had purchased Products P1 and P2. These products must be Traditional products which has multiplier attached to the Sum Assured paid to Customer.
V2. Premium Cycle
Premium Cycle of the Policy, determines how frequently customer and Insurer will interact with each other. Below graphs shows 46.7% of Policies had Semi-Annual as Premium Cycle. The logic behind the Semi-Annual is Fraudsters doesn’t have to pay the Complete annual premium as well as interaction with Insurer is also less. Customer- Insurer interaction reveals Customer trustworthiness.
V3. Policy Tenure
Policy Tenure is one of the factors for the Policy Premium Amount. Higher the term Lower the Premium Amount, as Insurer is going to get premium for longer durations. Our Study shows that Fraudsters are using this feature for their profitability. 79% Policies were having 30 Years and above tenure.
V4. Customer Education
Customer Education shows the how literate customer is which can tie to Customers’ awareness and decision capacity. Our study shows that 65.4% policies are having Customer literacy as Grade 1 and 2. Low Literacy can be tied to Frauds.
V5. Occupation of the Customer
Occupation can be described as the Financial Stability. As Mentioned above in the article, Financial Stability can be a Motivational Factor to pursue a fraud. As per our Study, 27.3% policies have O4 as an Occupation. This O4 occupation could labor work/ Agriculture work or Simply Business.
So Crux of the story, There is a chance that all the Fraudulent claim are generated by Group of Fraudsters and not an Individual Person so, Points which can raise Early signs are
- Product P1/P2 are chosen, as these products have multiplier attached to Sum Assured
- Half Yearly Premium Cycle to maintain low contact with the insurer
- Maximum Policy term is taken to reduce Premium Amount (Investment for a Fraudster)
- Low literacy population is targeted, as no person is literate to ask Fraudster a question (Share of Profit might go to Customer whose name is used)
- An occupation which doesn’t generate any trace of income are used, so fraudsters can easily manipulate customers income just to get a policy.
Watch out for these early sign Mr. Insurer.