Leveraging health data in the intelligent automation of underwriting

Given the myriad stakeholders along the healthcare journey – from doctors and pharmacists to nurses and hospitals – integrating the almost countless, disparate data sets to get a singular view of a consumer is a complex undertaking. Being able to do so effectively creates significant opportunities for both health and life insurers to better understand and manage new customer applications, develop rich engagement strategies and gain fresh insights into existing policyholder risks.

Traditionally, the insurance underwriting process has been a complicated, time-consuming process with questionnaires and medical forms to complete. These questionnaires can result in incorrect new business pricing and decisions, and conflict at claims stage, due to potential non-disclosure that the consumer may or may not have understood when completing the original paperwork.

Often, medical tests are required to give underwriters better insight into the health, and underlying health risk, of the person they are insuring. These tests inevitably slow the underwriting process down, bringing substantial cost and decision time delay to the insurer, which ultimately creates a poor customer experience.

Technology disruption, in particular, digitalisation of health records, API-based solutions, artificial intelligence (AI) and machine learning (ML), self-administered and cost-effective testing capabilities such as remote photoplethysmography (rPPG) and genomic sequencing are changing the fundamentals of underwriting.

The shift to electronic health records is becoming a game changer. As broader sets of health data become more readily available digitally, the ability for the insurer to view historic consumer health records exponentially increases. This in turn enables them to better understand the risk they are underwriting. Data sets becoming more readily available include doctor, clinic and hospital visits, medication prescriptions and usage, pathology lab results, as well as data from diagnostic and wearable devices. Access to this data is not without challenges, due to its fragmented, unstructured and distributed nature, as well as the time and cost of setting up data partnerships with providers.

Here again, technology plays a vital role. API environments are integrated now in even basic IT systems, as are often found in medical and other health practices. Integrating these disparate data sets into a health platform, such as the solution that Alula provides, enables stakeholders to share and exchange enriched data sets which, outside of the insurance vertical, enhances care, diagnosis, and clinical decision-making. The data that can be retrieved through integration includes clinical diagnoses, known chronic conditions, medications taken, as well as historic records for blood pressure, BMI, cholesterol, glucose levels and other important blood markers. Existing consumer data, which the insurer may have, is also extracted to be combined with the new, external data sets.

A key element that must be considered when considering use of electronic health data for insurance purposes is the consent of the consumer. GDPR, POPIA and other data legislations enforce service providers to only use data when consumer consent has been provided, that the data must be treated securely and only used for the specific purpose (underwriting) that the consumer provided consent for. This consent, however, can be obtained electronically at quotation or underwriting stage – provided the consumer is properly informed as to the usage of their data and assurances that it will be kept securely and only used for specific risk assessment processes.

Advancements in AI and ML are significant enablers to the underwriting process. Algorithms use the digitised health data records to create scoring metrics, which provide underwriters with more insight into the consumer health risk. Individual data sets, using trained ML, are compared to broader population statistics to infer risk of diabetes, stroke, heart attack and others in the future and this insight can be condensed into a single Health Risk Score. Providing a single score of an individual's health, while taking all known risk factors into account, frees up an enormous amount of specialist underwriting time, which would normally be spent analysing medical reports, self-assessment questionnaires and other, usually paper or manual, data points.

The ability to source, ingest and make sense of historical health data is only one part of the equation. The second is the ability to complement this data processing and decisioning with up-to-date and real-time health measurements that allow insurers to help their policy holders avoid developing chronic conditions. Diseases such as obesity, diabetes and heart disease cost insurers billions of dollars in claims every year and the ability to mitigate this risk by even a few basis points annually would be substantive.

The ability to proactively, and cost-effectively, test and triage for risk factors in all consumers, rather than only test where a family history exists, allows insurers to manage future risk more effectively, both pricing the risk in at new business but also proactively helping consumers manage their future risk more effectively. Even better when the instantaneous results are factored into the decisioning processes and ongoing policy holder engagement strategies.

rPPG, a technology that measures subcutaneous blood flow, has emerged in recent years as an interesting technology that many insurers are experimenting with. Some of the data points able to be extracted using rPPG include heart rate, heart rate variability, blood pressure, diabetes risk and other health risk factors. The ability to obtain real-time health measurements and insights from a 30- to 60-second facial scan enables insurers to classify individual risk without the need for onerous paperwork and blood work. While these risk indicators may change slightly daily, they still provide good insight into the broad individual risk and can be used to determine very early in the underwriting process which consumers should be allowed to apply for cover and which individuals should either be avoided, sent for further blood testing or to have their policies appropriately loaded. Further, expectations are that by 2033, saliva-based genomic sequencing will be made available at retail stores to identify potential risk and to enable preventative treatment of costly and deadly diseases (Celent Insights, Sep 2022). Cost-effective and digital testing will only continue to be available – so insurers need to anticipate how these data points will be embedded to increase efficiencies in their processes, enable better decisioning and enhance customer experiences.

Individuals themselves gain good insight into their own individual health, so there is personal benefit in doing the facial scanning process, especially if this can fast track their policy approval without requiring blood tests to be done. rPPG technology can thus be used at initial point of underwriting, but it could also be used to monitor an individual's health changes over time alongside wearable and other health data sources. If consumers are incentivised to regularly do a facial scan, they can be rewarded with loyalty offerings or even decreased policy premiums if their health profile changes positively. Insurers should therefore start to proactively work with the health risk across their book of business and this additional customer engagement should strengthen the relationship between insurer and policy holder.

The key to leveraging technology effectively, though, is to enable underwriters to examine risk profiles and make decisions without physically having to work through the vast quantities of data available. Making a platform available to underwriters, which combines all the disparate data in a structured way and a centralised location, is the first step to driving efficiencies. The next step is to enable certain the underwriting decisions to be made automatically.

This can be achieved through decision engines which, given the correct parameter inputs, can provide an output decision – provided all the required inputs exist, and the decision engine rules have been set up to cater for scenarios encountered. Decision engines are really a form of RPA (Robotic Process Automation) in that the rules must be set up carefully upfront, and the system will then step through the decision tree structure in a logical fashion, reaching an outcome that is predefined.

A natural evolution from traditional decision engines is to use another form of technology known as digital twin modelling. Digital twins are AI trained models that replicate human decisions, given specific data inputs and required output decisions. These models allow insurers to model underwriting decisions without the need for complex decision tree structures.

Expert underwriters create a matrix of the data inputs they consider, and sample data is then presented to the underwriter for decisioning. The AI then learns from the expert and creates an algorithm that mimics the underwriters decisioning and this algorithm can then be deployed into the business process as a virtual expert that makes automated decisions on data presented to it. This type of AI embraces human judgment and bias in the decision-making process, something that a decision engine is simply incapable of doing, as it needs to follow very specific rules that have been pre-programmed. These decision models will obviously only ever be as good as the human decision-makers they were modelled on, therefore the best people in a business need to be used when engaging with this type of technology. By modelling the best people within a business means that that insurer is maintaining its underwriting expertise and USP, so a digital twin of one business will not be the same as another company’s. While still relatively new, this technology has been used very successfully in many industries across the globe, including banking and insurance verticals in recent years.

Making digital data sets available to AI-driven algorithms for decisions allows for an automated business process, involving little to no human intervention. Only where a digital twin model is unable to create a decision would a human underwriter be needed to evaluate the data presented, potentially request additional information, and then make a final decision. This end-to-end automated process is something that, until recent years, has only been discussed theoretically. Now that the right technology toolsets are available, this theoretical process is now a reality.

The automated process described above is equally applicable in other insurance business processes, most notably claims and risk renewal processes, where senior human decision-makers typically need to spend vast amounts of time analysing different data sets and making decisions that are both costly and time-consuming. This results in efficiency gains for the insurer and a far improved customer experience for the consumer.

The Alula solution, which is offered to the southern African market through SilverBridge and the HealthCloud platform, is an integrated platform that combines access to individual health data, credit bureau information, industry fraud risk and government identification services. Integration to an insurer’s own database and business rules is achieved through exposed Restful APIs. The total data set is complemented with native integration into an rPPG scanning application for real-time health insights. The platform consolidates data sets from all available sources, whether they be external or internal, historic or real-time, and exposes this data to digital twin AI models, enabling straight through processing (STP) of both underwriting and claims decision-making. The health data available through the platform includes hospital, clinic and doctor visits, pathology lab results, medicine usage and pharmacies, wearable device data plus historical health claims.

The data made available through the Alula platform offers an insurer a 360-degree view of the consumer, providing health, credit and fraud propensity risk in a single, real-time API call. In addition, the ability to create company-specific digital twin models, using the best decision-makers in their business, allows insurers to customise their appetite for risk in an automated fashion. The improved accuracy that automating these processes bring, by eliminating the risk of human error, results in financial savings as well as more consistent and higher quality decisions.

The HealthCloud platform is currently being used by eight life insurers in South Africa and four health insurers. Current volumes of health data records accessed through the platform by insurers are increasing exponentially. To date more than 2.5 million unique health risk scores have been generated. Facial scanning is being implemented by key insurers who are trialling the new technology with an aim to make the new process available to the market in 2023. The HealthCloud platform speeds up decision-making time on underwriting cases – whether for new applications or claims – significantly. The cost savings per underwriting decision simply by having direct access to consolidated, digital healthcare records is estimated at 50%, given currently available figures.

The digital twin technology is currently deployed in four production insurers in both life and general insurance and several other insurers are considering how the technology can be most effectively deployed within their environments. The concept of AI decision-making is still relatively new, as was the concept of cloud environments no more than five or six years ago. Nowadays, most insurers have embraced cloud and are running their production systems there. Those still on physical infrastructure are all planning the move in the short term and the same will happen with AI decisioning over time. RPA processes have been used for many years, but the shortcomings are now being clearly understood, embracing AI and ML to replace these systems is a natural progression.

This does not, however, mean that there are no challenges to overcome. Insurers and healthcare providers must be willing to embrace change and adapt to a more open and transparent process with clients. Facial scanning needs clear communication with the targeted consumers to illustrate both the potential financial and well-being benefits of this new technology. Without transparency and public education, no real change can be forthcoming. Having a platform that connects to an entire ecosystem of data, solving data interoperability, and providing a unified view of a consumer’s individual health profile, however, delivers strategic value that can help insurers and healthcare providers unlock new business growth and reduce risk across their books of business.

The road to a completely integrated data environment that can positively impact on underwriting and health decision-making is something that should be embraced by the insurance industry. At a time when digital transformation underpins virtually every industry sector, those insurers and healthcare providers who embrace a data-driven, automated and integrated environment to enhancing the customer experience, while making more informed decisions, will be the ones that will be successful.