After the pandemic, can insurers still trust analytics?
What can I say about 2020? Things have changed, people have changed, the business environment has changed, and the insurance industry is changing. Yet, the importance of data and analytics has not changed.
The pandemic has forced processes to go digital. More data is being generated from more points of origin today than at any time in human history. Data generated and the insights derived have triggered even more change. The question is: Are the insights and resulting decisions correct and appropriate? When I price a risk or reserve for a loss, are the decisions I’m making correct? We may not fully know for some time. However, it is crucial to take a step back and ask these questions.
The Pandemic’s Impact
The efficacy of analytics and predictive models is now questionable because of the potential for accelerated decay of those models. Changes in underlying assumptions and relationships between data elements due to COVID-19 have triggered this decay. Insurers have started to assess the inventory of models they use to determine whether they are still adding value with the insights they generate.
Numerous examples of this exist, including “next best action” models and models that rely on specific data points, e.g., credit scores. Exclusion assumptions within models that evaluate workers’ compensation and business interruption losses may no longer be valid in some states that no longer honor exclusions. Assumptions about the severity of auto claims may be flawed since people are driving faster due to less traffic (even when state regulators reduced premiums in response to fewer miles driven and other mobility trend changes). There may be increased D&O claims and medical malpractice losses. The gradual opening and closing of states in response to transmission rates may also impact data into 2021 and beyond.
The use of third-party data has grown exponentially in insurance. Insurers are ingesting it into predictive models. Sometimes, the ingested data may be inferred or modeled. Many insurance use cases require high-quality data that approximations may not fulfill. Bad data leads to untrustworthy insights, no matter how good the model. Garbage in; garbage out.
Determining the quality and suitability of external data is labor-intensive and requires time-consuming testing. The cost for this data and the effort to incorporate it into processes is substantial; insurers require a return on this investment. Insurers need to vet the data and the vendors that provide the data. Insurers need to do proofs of concept, validate that they will see the returns they expect, and confirm that they are getting the right data in the right amounts. Data scientists and folks responsible for data governance processes need to be engaged.
Easy-to-assess success criteria need to be part of insurer vetting processes for data and data vendors. These criteria can include knockout items or what use and business cases are behind the data initiative. SLAs for data providers also require definition. Remember, the assumptions around how good or bad the data may change.
Deployment of Analytics
One area for insurers to explore is how analytics models are becoming part of day-to-day operations. Analytics need to be part of the underwriting decisions in real-time; insurers need to leverage analytics leveraged to determine whether they should price risk at all.
For claims, analytics needs to be part of the reserving process. Insurers can use analytics to trigger interactions with policyholders to avoid losses in the first place. The use of IoT devices to detect water leakage and shut down infrastructure or monitor electric grid variations to prevent fires may impact overall losses. Machine learning algorithms overlaid on smartphone pictures of a vehicle damaged in an accident can determine whether it is totaled, which can reduce the frequency with which adjusters have to assess vehicles in person.
The damage from the Derecho that hit the Midwest in August 2020 was estimatedusing shades of color from satellite and drone images. Our research shows that between 15-25% of insurers, depending on the segment, are increasing investments in big data technology, AI, machine learning, unstructured text analytics, sensors, and drones.
Talent—Here Today, Gone Tomorrow
Finding and retaining strong data people is a critical factor in the success of any analytics program. We have heard that the importance of training and recruiting talent is something most insurers consider paramount. However, coming up with creative ways to do so remains a challenge for insurers.
Many are working from their homes today. Would an insurer be willing to have a geographically dispersed workforce to get the right capabilities? Will satellite offices have a place in the future for necessary in-person collaboration? When the pandemic ends, how will insurers retain this talent? Given the ever-increasing importance of data, more companies will be knocking on your data scientists’ doors in the years to come.
This blog entry has been reposted with permission from Novarica.