AWS for Industries
Underwriting at the speed of cloud – How life insurers can optimize underwriting using AWS
Life insurance underwriting is poised for significant transformation, driven by increasing customer expectations and enabled by the growing sophistication of technologies like data analytics, wearables, artificial intelligence (AI), and Generative AI. Modern insurers are leveraging these technologies to optimize their end-to-end underwriting processes. This article discusses the opportunity space and how AWS services facilitate substantial improvements across the four core underwriting stages – data collection and organization, risk assessment, pricing, and policy generation.
For each stage, we examine the business benefits for insurers, including reduced costs, increased productivity, accelerated scalable processes, and accurate data-driven decisions. Policy holders also stand to benefit from faster decisions, expanded accessibility, personalized pricing, and superior customer experiences.
While underwriting interconnects with operations like distribution and servicing, this article solely concentrates on underwriting itself without detailing related upstream functions.
Data collection and organization
AWS offers a range of cloud services and generative AI accelerators that facilitate efficient document data extraction, secure storage, and real-time integration with core platforms. These services help insurers overcome the challenges with human intensive processing of all the required information on applicants, including personal details, medical history, family history, lifestyle, financials, and more.
Extracting Insights from documents: AWS AI and generative AI services automate data extraction and summarization. Amazon Textract uses machine learning (ML) to extract text data from scanned or photographed documents like forms, tables, medical records, and handwritten files. The extracted information can be used to automate data entry and document review, saving manual effort. Amazon Comprehend (Medical), trained on health data, identifies prescriptions, procedures, or diagnoses. Together, they enable automated extraction, classification, and identification of crucial underwriting workflow information.
The Large Language Models (LLMs) available via Amazon Bedrock can supplement traditional AI services and achieve more comprehensive intelligent document processing outcomes. Combining Amazon Textract with Amazon Bedrock’s large language models allows for document classification, entity detection, and summarisation capabilities. This enables underwriters to produce an underwriting case summary across all records, including financials, medical, and lifestyle questionnaires. Amazon Bedrock provides secure access to a range of foundation models from model providers such as Anthropic, Cohere, AI21Labs, Meta, and Amazon.
Transferring data seamlessly: To avoid data inconsistencies and inefficiencies associated with re-entering applicant data into multiple systems, AWS offers integration services such as Amazon AppFlow and Amazon EventBridge. These services facilitate secure API driven bi-directional data transfer between AWS services and external applications in real-time, eliminating the need for overnight batch processing or complex ETL (extract, transform, and load) services.
Storing and organising docs & data: Modern insurance customers on AWS are managing the increasing volume of structured and unstructured underwriting data by utilizing a scalable data lake backed by Amazon S3. Amazon S3 provides a flexible, HIPAA-eligible foundation which not only supports the underwriting workflow but also acts as a data source for broader ad hoc analytics needs and custom machine learning model development. Insurers can leverage the pre-built AWS InsuranceLake solution to get up and running quickly with an S3 data lake designed specifically for the insurance industry.
By leveraging AWS purpose-built services, insurers can optimize data collection across documents and systems, eliminate manual efforts, gain data insights, and maintain a single source of truth. The data from structured sources can be augmented by data extracted from the various documents (which typically remain in hard copy format only) and be used to better understand customer behaviour, risk profile, and buying patterns. This improves underwriting efficiency and provides a strong data foundation for the next step, risk and pricing calculations.
Risk and pricing calculations
Modern insurers are leveraging the availability of supplementary data for underwriting, in combination with machine learning technologies, to augment traditional models and core underwriting platforms. Machine learning models trained on large data sets accommodate hundreds of input variables in their predictions to improve the speed and accuracy of risk-scoring and pricing calculations. This results in improved loss ratios while freeing underwriters to focus on higher value tasks and exception handling.
Supplemental data acquisition: With the growth of connected devices and sensors, insurers can tap into new data streams to improve underwriting analysis. Data from genetics and DNA testing, smartwatches, and other wearables that monitor exercise, heart rate, and sleep provides insights into an applicant’s overall wellness and lifestyle risks. IoT-enabled health screening devices collect applicant vitals during underwriting to get more objective data. In countries where regulators allow, DNA and genetic testing make it possible for family history and genetic risk factors to be assessed with accuracy. This rich supplementary data feeds machine learning models to enable greater accuracy and optimization of the risk and pricing process. Using new data streams will also support the transformation of underwriting from a one-off activity to a continuous process, where pricing is dynamic and changes in response to underlying risk data.
Beyond IoT data streams, services like AWS Data Exchange facilitate the secure sharing and integrating of large insurance and health datasets, further enhancing risk scoring accuracy.
Building ML risk models: The power in machine learning comes from its ability to identify patterns and relationships in vast amounts of data. This is useful for risk modelling. For example, a life insurer can supply Amazon SageMaker with historical underwriting data (first party and third party), along with the claims and mortality outcomes. These models synthesize all available data, including demographics, biometrics, and medical history, to accurately predict risk scores aligned with the insurer’s past underwriting decisions.
Predictive ML models support the reduction of applicant information required and bring visibility into the expected time of underwriting response. Lower risk applicants get an accelerated customer experience with fewer questions. Higher risk applicants are guided to provide additional details and, given an expected turnaround time, promoting transparency.
Pricing with advanced analytics: Insurers want to price accurate policies based on each applicant’s specific risk profile using multivariate pricing models. Until recently, they limited pricing approaches to Generalised Linear Models (GLMs), which can only handle a few rating factors. Machine learning provides an opportunity to move away from the constraints of rating tables and towards ML algorithms, such as XGBoost and random forest, that can work with hundreds of features and, subsequently, can be more accurate.
There is a wealth of data insurers can use to enhance pricing models, but deploying it can be difficult. By using AWS services like Amazon SageMaker, insurers can use diverse data sources, train, deploy, and embed dynamic pricing models into underwriting. This enables real-time, risk-based quotes adapted to each applicant’s full breadth of characteristics. The end goal is more accurate pricing and a streamlined customer experience powered by data.
Fraud detection: Fraud continues to be a significant challenge and operating cost burden for insurers to address. Detecting and investigating fraud is a human resource-intensive process, and with teams of limited size, they examine a small percentage of policies and claims. The breadth of fraud permutations covering medical and financial misrepresentation, agent, criminal, and identity make detection complex. Generative AI and machine learning technology are well-positioned to detect fraud more accurately during the underwriting process. AWS services like Amazon SageMaker and Amazon Bedrock help insurers quickly use, train, and deploy fraud detection ML models using their own historical fraud data. Using ML, insurers can apply fraud detection to all applications and claims, stack rank the model outputs, and support specialist teams to prioritise their investigations.
Using AWS machine learning services, insurers can bring market leading data science capabilities to transform the efficiency and accuracy of the underwriting risk assessment and pricing phases. ML models lead to highly accurate risk prediction and fine-tuned pricing that focus on the right risks.
Policy generation and issuance
Underwriters are required to manually create and then populate policy document templates using applicant data. This often requires the gathering and re-keying of information from multiple systems. Document templating technologies produce static documents with a small subset of dynamic fields, making it time intensive to assemble complex policies. AWS services are available to integrate with existing core systems, automate dynamic policy document generation, and enable delivery to end customers.
Workflow orchestration: AWS Step Functions is a serverless visual workflow service that lets you orchestrate API integration and data transfer across distributed systems to automate complex processes. Insurers can build a workflow that orchestrates data collection, policy generation, updates to core policy and billing systems, third party digital channels, and end customer communications. This automation reduces manual processes and operational overhead.
Dynamic document generation: After insurers make underwriting decisions, they need to generate customized insurance policies. Multiple disparate IT systems and complex documents make assembling and issuing policies human resource intensive. Generative AI is well suited to addressing the problem of complex document generation. Amazon Bedrock provides customers with access to foundational generative AI models, which can create a personalized policy document containing all information extracted from various systems. Models prompts can include sample templates and instructions to perform quality control checks to ensure all the required information and terms have been included in the policy.
Instant policy delivery: Regular and timely communications are critical to customer satisfaction and retention. Amazon Pinpoint enables high-volume delivery of multi-channel customer communications. Supported channels include push, email, SMS, and voice. Email delivery of policy documents avoids lags in postage of hard copy documents while providing full traceability of reads and bounces. Push and SMS channels support real-time updates to customers and agents as applications and claims progress through processing.
Automation use case examples
Here are some real-world examples of how insurers are using AWS automation capabilities across underwriting and other related functions:
- Canara HSBC Life Insurance applied Amazon Textract and Amazon Rekognition to build a solution to extract information from government IDs, financial documents, application forms, and medical reports. This reduced the data entry by 70%.
- Elevance Health used Amazon Textract to digitize and automate their processing of medical records by extracting information and indexing and classifying the documents. They have been able to automate 90% of the document processing for claims.
- Root Insurance Co. leveraged Amazon SageMaker to apply machine learning to its business functions including pricing risks using telemetry from mobile phones and other devices. This approach led to customers saving up to 52% in car insurance premiums based on driving behaviour.
- Sumitomo Life built a solution leveraging Amazon S3 and Amazon SageMaker to analyse terabytes of health data collected from their health promoting product “Vitality” and offered personalized health services to their customers.
End to end underwriting optimisation with AWS
Cloud services, including generative AI and machine learning, provide a significant opportunity for insurers to augment core insurance systems. Insurers can use AWS to streamline and optimise the end-to-end life insurance underwriting workflow, including initial data collection, risk assessment, pricing, and ultimately policy issuing. AWS provides a full breadth of services to support each step with the flexibility to use them independently or in combination. Provided below is a reference architecture that summarises the complete opportunity space and the key AWS services and their interactions.
Getting Started with AWS
Insurers are discovering the tangible benefits of leveraging AWS to modernize underwriting. Below are the best practices to get started with underwriting automation and establish an optimisation roadmap:
Start Small, Then Scale: Focus initial optimisation projects on high friction areas like document data extraction or improving risk-based pricing to show value. Gather lessons before expanding to full end-to-end workflows. Leverage agile principles with iterative delivery.
Choose Serverless First: Serverless services like Amazon Textract, Amazon Bedrock, and Amazon EventBridge shift undifferentiated infrastructure management tasks to AWS. They seamlessly scale and integrate with data lakes and applications. Starting with serverless speeds up development and keeps costs low.
Take Advantage of Partners and Pre-Built Solutions: Consider engaging the services and experience of specialist partners. AWS has AI/ML competency partners across geographies who help you get started with AWS solutions and implementation expertise. AWS offers out-of-the-box solutions tailored for insurance, including underwriting data processing, risk analysis, and document generation. Leverage proven architectures and speed up your time-to-value.
Focus on Business Metrics: Define success criteria and measure the true business outcomes from increased underwriting optimisation and automation, such as improvements in productivity, underwriting expense ratio, loss ratio, and customer retention. The ability to measure outcomes and return on investment (ROI) will help fund ongoing iterative improvements.
Conclusion
In this post, we have covered how life insurers can optimise the end-to-end underwriting process by leveraging key technologies across data, machine learning, and generative AI. For each of the four stages of underwriting, we unpacked the most compelling optimisation opportunities and the AWS services that make it possible.
Carriers that are successful in the modernisation of their underwriting process will reduce per submitted policy underwriting costs and increase profitability through faster, more accurate, and repeatable decision making. Importantly, insurers can transform the customer experience with the ability to offer a greater range of products with personalised pricing and an accelerated application process.
AWS has the services, solutions, and partner network to support your underwriting modernisation journey. For further support in getting started, contact your Account Manager or AWS Sales Support.