For one InsurTech company, a new round of Series A funding will lead to a revolution in  insurance claims management. It is truly a story entwining the spirit of innovation with effecting actual change. Such further investment in the ongoing projects of this pioneering InsurTech company marks a future where insurance claim management is no more a process but a seamless experience that could overhaul the present relationship between the insurers and the policyholders.

Mission and Vision of a Trailblazing InsurTech Company

This InsurTech organization, established in 2018, set out on  a journey to revamp the way policyholders claim insurance. It utilized new applications in the area of blockchain development with other modern technologies to ensure that making an insurance claim is an efficient and free-flowing process. To this end, this pioneering InsurTech organization has reinvented its approach to making the insurance claims process more efficient, transparent, and user-friendly with the intent  to simplify what has traditionally been a difficult and time-consuming process.  

Setting New Industry Standards 

The company’s vision is to be an innovator in the  insurance technology space. Having garnered recognition through numerous awards, it’s already making significant strides as an industry innovator. This InsurTech  innovator wishes more than just pushing through technological advancement, it aims to reshape the whole insurance landscape toward more efficiency, better customer satisfaction, and higher levels of trust. 

Smoothening the Claims Process

The company imagines a future where the process of insurance claims won’t necessarily be a source of stressful feelings but would rather be a very smooth part of the insurance experience. This vision embraces the empowerment of insurance providers with tools for faster, more reliable services, reduction of overhead administration, and enhancement of efforts in customer-centric innovations. The company is introducing a new pathway in insurance claim management by committing to using blockchain development and contemporary AI technology.

Recognition through Series A Funding

In February of 2023, this groundbreaking InsurTech company secured a new round of Series A funding from varying stakeholders including HSBC, Wing Capital ventures, G&M Capital as well as many others to the tune of $6.85 million. 

This newly-acquired  funding will go toward improvement in the existing technological infrastructure, the modernization of AI algorithms, the enhancement of data-processing capabilities, and the adaptation of the system to be scalable in such a way that an increasing flow of claims will be managed properly. Part of this funding also provides for the expansion of the team. This is all part of their commitment to recruiting top talent in AI, machine learning, data science, and customer experience.  It’s not a growth number; this newfound commitment features diversity in perspectives, along with expertise that can add value to the innovative culture of the company.

The company also has an eye on geographical expansion, and this has been made possible due to successfully implementing their system in initial markets. That expansion brings with it its own sets of challenges as it involves navigating different regulatory environments and customizing solutions according to various market needs. However, with this, the company has enough ammunition in place to take on the challenges that come their way. It is a positive step towards revolutionizing the management of insurance claims the world over.

Omdena Case Study: Enhancing Fraud and Anomaly Detection Through AI

To help this pioneering InsurTech company navigate such groundbreaking digital transformation, they enlisted the expertise provided by the Omdena organization. Omdena is the world’s largest collaborative platform for grassroots AI development. This is a philosophy that incorporates training local members of the community and equipping them with the necessary technical skills to find technical solutions to the challenges presented within their own communities. In essence, this philosophy embraces the mantra of “ technology for the people by the people”. 

Omdena’s team of collaborators were able to create a two-pronged approach that leveraged both Supervised and Unsupervised machine learning. Supervised learning refers to machine learning approaches that feature labeled data. In the case of this ambitious project, it was used to quantify the number of claims requested versus the number of claims approved. In the case of Unsupervised learning, this approach seeks to find hidden patterns in unlabelled data by clustering data with similar characteristics into similar clusters. While evaluation in Unsupervised learning is more challenging, Omdena’s collaborators were successfully able to use this approach to provide performance benchmarking using popular algorithms such as K-Means clustering and DBSCAN. In this case, Omdena collaborators used both approaches to facilitate a comparative analysis of their performance metrics. 

However, the same way ideal collaboration requires the joint effort of a team of talented individuals, the same can be said for the varying algorithms that work together to create what is known as a stacked model. A stacked model, also commonly referred to as an Ensemble Method, is when more than one algorithm is used to learn from the mistakes of each other to ultimately create a ‘super model’ that provides greater accuracy and performance. 

For this, the Omdena team used the popular XGBoost algorithm. XGBoost is a scalable tree boosting system that is widely used by data scientists and provides state-of-the-art results on many problems. In simple words, XGBoost models the data with (N) number of decision tree models with each model learning from the mistake of the previous model (hence called boosting). The illustration below shows roughly how this algorithm works. 

As a result of these state-of-the-art modeling techniques, Omdena’s collaborators were able to adequately deduce that out of 846,657 claims, 276,320 claims were indeed fraudulent leaving 570,337 legitimate claims. 

Not only was XGBoost able to perform on claims data exceptionally well, Omdena’s collaborator also set their sights on approval data as well. From the approval data, Omdena’s team was successful in identifying 196,103 fraudulent approvals versus 163,548 legitimate approvals from a total number of 359,651 approvals. 

Although XGBoost performs very well, there are some inherent challenges when using this algorithm. For example, the Omdena team found the following to be of the utmost concern:

• Even though XGBoost performs the best among the supervised models, the loss incurred is very high and there is a lot of scope for improvement.
• XGBoost is very sensitive to outliers. Data scaling helps handle this to some extent which is incorporated in the data preparation step.

As mentioned above, there were two approaches taken to build the fraud detection tool. The second approach used clustering to provide a broader fraud detection strategy. To do this, the same data for claims and approvals as used in the clustering process. The reason being is that both datasets were pre-processed using feature engineering techniques that removed outlier data and imputed missing data points. 

Two distinct K-Means models were employed for the analysis: one configured with three clusters and the other with two clusters. Each model was carefully fitted to the claims data, taking into account the intricacies and patterns present in the dataset. The effectiveness of the K-Means models was assessed by determining the percentage of accurately identified fraudulent claims among the outliers.Both the three-cluster and two-cluster models showed similar capabilities in detecting fraudulent claims. However, the detection covered only a portion of the total fraudulent claims present in the data.

Additional insights were drawn by incorporating a unique feature in the claims dataset related to the quantity requested and approved. This feature was instrumental in defining an accurate label for actual fraudulent activity. The three-cluster model’s success in identifying fraudulent claims was quantified, revealing it identified approximately 6.52% of the actual frauds in the dataset.The overall predictive accuracy of the model in the context of the entire dataset stood at around 66.62%.

One of the most fundamental insights when using clustering is the financial disparity based on the model’s predictive accuracy.These disparities could be seen in both claims and approvals depending on the number of clusters used. After adjusting the data against both models, the team came to the following conclusions based on the two datasets (claims and approvals):

Approval Dataset

• Perfect Model: Demonstrated an incurred loss of $61,300, serving as a benchmark for comparison.
• Outlier Models Using K-Means Clustering: Both the 3-cluster and 2-cluster models resulted in significantly higher losses, each approximating $3,541,286.91. This substantial increase in loss compared to the perfect model underscores the challenges and potential financial risks associated with employing these particular clustering approaches in fraud detection.

Claims Dataset

• Perfect Model: In the context of the claims data, the perfect model incurred a loss of $98,840.
• Outlier Model with 3-Cluster K-Means Prediction: This model led to a loss of $637,891.82. The total amount involved in the claims dataset was $2,182,071.34. The loss observed with the 3-cluster K-Means model, while lower than that of the approval dataset’s K-Means models, still represented a significant increase compared to the perfect model. This indicates a notable gap in the effectiveness of the K-Means clustering approach in accurately identifying fraudulent claims.

As shown by these metrics, there are challenges with the clustering approach. However, used together with the Perfect Model (XGBoost), a more holistic view of the claims and approvals and their predicted accuracy become more clear. That being said, the challenges presented by this second approach can be summarized into the following points: 

• Sensitivity to Outliers: K-Means is sensitive to outliers. Fraudulent transactions are often outliers and can skew the centroid calculations, leading to less accurate cluster representations.
• Assumptions made by Spherical Clusters: K-Means clustering assumes that clusters are spherical and evenly sized. This could prove different with real-world data, especially in fraud detection where fraudulent transactions might not form a coherent cluster.
• The Effects of Choosing ‘K’ Values: Determining the optimal number of clusters (K) can be challenging without domain expertise or additional methods like the elbow method. This means that sometimes clustering is not clear-cut for complex and high-dimensional data.
• Random Impact of Initialization: The results of K-Means clustering can be extremely influenced by the placement of centroids. Different initializations can lead to different clusters, which can affect the model’s performance.

Lessons Learned 

Unsupervised Models:

• The K-Means clustering models still incur significant losses. This indicates that while useful for identifying patterns, K-Means may not be the best suited for the task of detecting fraudulent claims. As mentioned previously, K-Means relies on seeing fraudulent claims as only outliers, which is not the best method.

Supervised Models:

• The XGBoost algorithm was the strongest model in lowering losses. It has the ability to handle a variety of features effectively making it a good choice for fraud detection. Random Forest models demonstrated promising results in the broader dataset. This suggests their overall effectiveness in accurately deducing fraud.

Performance Evaluation:

• Evaluating models on the full dataset could possibly skew the perception of a model’s real-world effectiveness. However, caution should be taken when looking at these results. They might not fully represent the model’s performance on new data. On the other hand, it is a helpful supplement to claims assessment. 

Ensemble Methods:

• The exploration of ensemble methods, particularly from outlier detection models like K-Means and DBSCAN show some promise. However, these approaches have not yet surpassed the effectiveness of the more traditional Random Forest and other Tree-based models in minimizing losses.

Bridging Gaps: Elevated Customer Experience with Operational Efficiency 

The benefits that are gained by insurance providers who are making a digital transformation are manifold: This system, being AI-driven, reduces manual work, minimizes the chance of errors, and also reduces the time for processing claims. This leads to a very cost-effective solution for the insurers to better serve the customer but also to spend less time on administration. The system built in collaboration with Omdena  is also fortified with advanced fraud detection, which acts as an added security measure to protect against financial losses, fraud, and reputational damage.

This means that the complete customer experience is optimized resulting in greater customer satisfaction. Further proof that collaboration with Omdena can translate into success for any organization looking to transform their digital business and enhance their operations with AI technology. 

Future Perspective: Sustainable Growth and Industry Leadership 

Taking this into perspective, this InsurTech company outlines a future vision of sustainable growth. From solid technological underpinnings to an ever-growing team of insurance experts with a crystal clear vision, this InsurTech provider is well on its way to becoming a true powerhouse in the field. But it’s not entirely keen on growing the customer base only; there’s a focus on relentless innovation. To be ahead of technology, exploring new innovations and approaches becomes necessary to inject new lifeblood into the landscape of insurance claim management. With the help of Omdena, the world’s largest collaborative grassroots AI platform, this collaboration allowed this InsurTech company to take a step forward into the world of better claims and approval management. 

These InsurTech innovators work with the diverse set of the insurance industry’s biggest players: from insurance providers to policyholders; from regulators to technology partners who will enable the company to understand the pain points affecting their market. It will thereby make sure that deep and systemic improvements are built to cater to all corners of the insurance industry. 

All in all, it’s the story of a game-changing InsurTech company that starts with a vision, has a tidal wave of Series A-led talent flowing through it backed by a global AI platform, and focuses every single aspect of its execution on a client-centric approach quickly changing the playing field of insurance claims management. It’s a story of what technology, when steeped in vision and executed with perfection, can truly do in an industry.