Claims processing cost and fraudulent claims payment can increase the operating cost of insurance companies and greatly inconvenience customers. Claims management has therefore become a key focus area for insurance companies, as it impacts the bottom line and affects the customer retention strategies. Digital transformation and disruptive technology in the form of artificial intelligence (AI) can greatly enhance claims management practices and ensure greater customer satisfaction.
Adoption Of AI In Claims Process – Current Scenario
Various insurance companies are changing their claim processing lifecycle to make it increasingly automated. This has significantly reduced their turnaround time, regulation cost, and claims processing cost which has led to better customer service and business profitability.
Automation in the insurance industry has entered the following areas:
- Digitizing documents and records.
- Automating intake of FNOL via interactive technologies like ChatBots.
- Assessing the policy coverage, possibility of a fraudulent claim, loss and payouts via predictive modelling and deep learning techniques with big data.
- Using GPS sensing software to assess the loss.
According to a survey conducted by O’Reilly (2017), only 1.33% insurance companies are investing in AI. Specific challenges in the adoption of AI in existing business models is likely preventing the insurance industry from leveraging these advancements for claims processing.
What Are The Challenges In AI Adoption?
Let us look at the various challenges in the adoption of AI for claims processing.
Unbalanced Data Sets For Training
The AI system needs to be trained on a large volume of data to cover each probable scenario of a claim. This requires the AI system to deal with a variety of structured and unstructured data, which consist of historical claims, documents, transactions, investigative reports, GPS data, images and so on. Unbalanced datasets for training lead to low predictive accuracy of the AI system in fraud prevention and claim management.
Engineers create the AI algorithms and their learning is largely based on historical claims. Hence, human prejudice can enter the AI algorithm. This can cause the AI system to become biased and mishandle thousands of claims.
The model’s learning is largely based on historical claims. Post deployment, the model enters a self-learning mode. Failure in continuous monitoring can lead to a minor change in the algorithm, which can directly affect the outcomes prediction of thousands of future claims and result in improper settlement. This can directly impact the revenue of the organization.
The AI system deals with a large amount of data for training as well as for real time decision making of present claims. This data is stored on the insurer’s servers or cloud. In order to process a claim, various applications that are used for filing a claim, evaluating the loss, and calculating the payout, access this data. In addition, this data is also accessed to enable continuous learning of the AI system. Considering the size and nature of the data and the connectivity among applications, there arises a risk of data leak and security breach. This makes insurers reluctant to adopt the AI system.
Regulatory Compliance and People
Different states have different regulatory frameworks. This results in complexity in compliance and restrictions in the data collection for building AI models. The use of data is the key but the insurer must ensure appropriate permissions for its use the data. This restricts the insurers to go global with the AI system.
Using the data or model with present potential ethical and public interest issues may lead to ambiguity among the various stakeholders involved. Further, explaining the AI model to backend business operators like auditors, and regulators is difficult. All this results in increased risk to the business.
Possible solutions to the challenges encountered in adopting AI for claims processing include:
- Using of unbiased and balanced datasets to train the deep learning AI system for higher predictive accuracy. This requires pre-processing of data, which involves normalizing, image processing and deep learning algorithms to generate model for the AI system.
- Performing extensive testing of the AI model and error analysis in the form of precision, accuracy, and overfitting, along with validation from an independent non-competing agency for unbiased results.
- By evaluating the performance of the AI model by on-going monitoring, and applying stringent standards to data assessment.
- Performing vulnerability assessments of organization’s information security function in order to address cybersecurity / data security risks.
- Educating internal and external stakeholders about the data access restrictions and state specific regulations ensures compliance of the AI model with regulatory norms of the country.
- Providing complete disclosure to policyholders about data collection and its use.
- Building a three-layered defense structure comprising of – business, compliance and internal audit. This ensures that no regulatory or compliance issue is overlooked. Complete participation from all three units from the beginning would allow them to understand some of the critical technological aspects and customize the AI system according to their needs.
Adoption of AI in claims process results in consistency in handling of claims, reduction of turnaround time, and accuracy in claim settlement for both the insurer and the policyholder. It also reduces the operating cost and resource use of the insurer. However, there are several challenges that need to be overcome before AI can be an integral part of claims processing. Strategic planning towards AI adoption with a solution based approach to minimise risks involved in handling of large data sets, extensive testing and training, and ensuring regulatory compliance, can help the insurance industry to process claims at a lower cost and achieve better business outcomes.