InsurTech for Risk Management

Alexander Laurie
Principal, Risk Modeling Services, PwC US

Steve Walsh
Director, Risk Modeling Services, PwC US

Kris Kogut
Principal, Risk Modeling Services, PwC US

Part 2: Risk Analytics

Risk analytics play a key role in building and maintaining critical insights that can make smarter business decisions possible

In addition to continual protection of amassed data, there’s a business imperative to use information to better understand risk. In the case of a risk manager, the insights gained from risk analytics can affect how you manage risk and insurance programs and help your business recover from negative events.

This article builds on the series InsurTech for Risk Management Part 1: Data Infrastructure, where we showed that by delivering a single source of truth, businesses can establish a transparent view of risk. By leveraging that unified understanding of the risk landscape, risk analytics can be applied to allow your company to address more sophisticated business questions.

Increasing sophistication of demands

Companies produce copious amounts of information in the course of business. Demands for insights into this data continually increase. But how well are you able to access and use your information?

“65% of risk management professionals are increasing overall spending on risk management technology.”

- PwC, 2022 Global Risk Survey

Increasingly sophisticated and accessible risk analytics can help you uncover those insights. Start with the routine occurrence of an employee on-the-job injury, for example. Incident investigation usually begins with backward-looking questions about what happened, then progresses to forward-looking questions about how to implement procedures that can help better safeguard employees in the future. The same progression scales up — from the risk to an individual worker to the risks to an enterprise. And the progression of questions remains the same, easily categorized by a data maturity level, which is shown below:

Descriptive and diagnostic analytics

Analytics almost always begins at the descriptive level. Gathering basic facts provides the foundation for future insights. Ten years ago, gathering descriptive data across the breadth of risk metrics would have been a herculean effort for most organizations. Now, the increased prevalence of visualization tools linked to connected data platforms allows many companies to gather descriptive data and perform diagnostic trend analyses quickly.

The descriptive and diagnostic levels largely report what happened in the past. However, there’s much to glean from that information because it allows for the predictive level to transition from reporting to forecasting, where true risk analytics begins.

Predictive analytics

Risk analytics incorporates not just the average result but also the variance of key metrics and how they combine to generate a distribution of possible results. For example, the risk manager for the owner of commercial multifamily residential properties will be concerned about potential hurricane losses. If the manager expects $1 million in damages on average and there is no variance, then the property owner needs to hold $1 million of capital. However, a distribution captures a more realistic scenario, where damages could range from $0 to $100 million. The capital required is better estimated through predictive simulations than averages.

Prescriptive analytics

The prescriptive level of analytics combines the underlying risk analyses with overall business objectives to recommend a course of action. Risk managers often employ prescriptive analytics for litigation management. Prescriptive analytics considers factors such as detailed medical notes about an injury, jurisdictional reputation and continually updated metrics about the case management to inform recommendations. If an incident occurs, should your company assign the case to counsel? Should you offer a settlement agreement, and if so, in what amount?

Cognitive analytics

Cognitive analytics is common in fields outside of risk management, in everything from choosing advertisements to place on a mobile device to executing navigational control in a toy drone. Few companies currently deploy cognitive analytics for risk purposes, such as automatically booking a balance sheet charge for predictive hurricane loss or automatically offering a settlement for litigation management.

However, operational usage of cognitive analytics ultimately raises questions for risk analytics. For example, ride-sharing companies provide routing recommendations to drivers (and in the absence of autonomous vehicles, let’s assume drivers follow directions). While cognitive analytics incorporate distance and traffic, do they:

incorporate the higher risk of an accident from a left turn versus an alternate route?
allow for the risk manager to use predictive analytics to quantify the cost of the risky left turn versus the cost in time and distance of an alternate route?
have organizational support to provide those insights to operational teams and refine their cognitive analytic models?

Some of the answers to these questions require a technologically advanced combination of tactics. Predictive simulations, prescriptive recommendations and cognitive analytical decisions quickly exceed the limitations of analytic tools such as spreadsheets — which, incidentally, still use a 40-year-old design.

Fortunately, there are new tools that use artificial intelligence (AI) and machine learning (ML), to combine advanced tactics more easily. The National Association of Insurance Commissioners (NAIC) found that 70% of 193 private passenger auto insurers currently use AI/ML models for operations, with an additional 10% having models under construction. Common uses of claims models are to:

detect fraud
determine settlement amounts
make assignment decisions
evaluate images depicting loss

In our experience, we rarely work with risk managers who are using or building AI/ML models in addressing these challenges. Historical barriers to entry for analytic tools may explain low usage.

More easily accessible

Coding skills and statistical knowledge of candidate models are high barriers to entry for teams seeking to leverage risk analytics. Advances in usability of cloud environments have lowered or removed these barriers.

Consider the evolution of AWS SageMaker on Amazon Web Services (AWS). Sagemaker is an app for developing analytic models, such as risk analytics. When released in 2017, it provided a significant advancement by incorporating machine learning with a fully managed cloud infrastructure, tools and workflows. However, if you did not know how to use those tools, the barriers to entry remained intimidating, as illustrated by this snippet of code from an auto insurance claim fraud detection model.

In 2019, AWS added AWS Sagemaker AutoPilot to automatically build, train and tune machine learning models with control and visibility. This lowered the barrier to entry for statistical knowledge of candidate models. Users can train dozens of algorithms and then become comfortable with the statistics behind the models that ideally fit the training data.

Most recently, in 2021, AWS added Sagemaker Canvas. This lowered the barrier to entry for coding skills by providing a no-code, visual interface for machine learning model development. It builds upon the features of AutoPilot to guide the user through the model building and evaluation steps.

These increasingly powerful and user-friendly analytics tools can be combined with the cloud data infrastructure described in Part 1 of this series to enable risk managers to migrate analytics away from traditional spreadsheets to provide greater business insights.

Takeaway

Businesses increasingly expect risk managers to move beyond basic descriptive information and provide predictive and prescriptive insights about risks. Risk analytics can help you uncover those insights. The barriers to entry for performing sophisticated risk analysis have decreased through user-friendly analytics tools such as AWS or other cloud services. A migration from spreadsheets to the cloud allows risk managers to provide more sophisticated insights more easily.

Look for a detailed examination of risk analytics’ results in our Next in Series: Reporting.