Q&A
Big data and predictive modeling are transforming how claim examiners identify and manage high-risk workers’ compensation insurance (WC insurance) claims. The results are helping to reduce return-to-work time frames and claim costs.
Vice President and Claim Data Scientist Paul Drennan is a driving force behind The Hartford’s innovative approach to predictive modeling. This modeling informs the company’s comprehensive solutions for workers’ compensation claim intervention and management.
Learn about big data and why it’s such a critical value proposition for large, loss-sensitive companies.
What is big data?
PD: Big data is a product of the digital age. It refers to the large streams of information that businesses collect every day through digital sources. These sources can range from websites and social networks to sensors and electronic communications. This data can be analyzed, often in real time, to quickly reveal insights and trends that aid in decision-making, improve business performance and build competitive advantages.
What makes big data different from traditional data?
PD: Big data is coming at us faster and in a volume that is exponentially greater than traditional data. For instance, we may collect a payload of data through our in-house claim system, but that’s still fairly manageable compared to the data captured within every click on all of a company’s web portals. Or consider an activity tracker like Fitbit®, which generates a pulse of data every 15 seconds for those wearing the device. Data of that velocity and volume is big data.
Big data is also characterized by variety, specifically in the form of freeform data that can be sifted through for insight. A claim handler can enter the standard data about a car accident to a claim system, but it’s the details in his/her text notes about the police report and the interview with the claimant and the claimant’s doctor that yield insight about potential fraud. That data doesn’t have a structure. There is no template for capturing it. Big data pulls the meaning out.
What kind of technological advances make predictive modeling possible?
PD: Workers’ compensation claims are data intensive and expansions in data capture capabilities have advanced to the point where we now have the ability to collect, store and manage the large volume of data that strong predictive models need.
With sufficient data and powerful computational tools in hand, data scientists and engineers who specialize in predictive analysis can create models for a range of insurance needs, such has helping identify potentially volatile claims, fraud, subrogation and other conditions.
Each model brings together unique combinations of data, such as loss and treatment plans, environmental factors and time periods. It then uses this data to:
- Find patterns
- Identify outliers
- Recommend the best next steps for the claim handler to better manage claims
How is big data processed so it can be put to practical use?
PD: We use advanced analytical methods to see patterns in the data and build models based on those patterns. Predictive models allow us to forecast what might happen in the future based on what happened in the past, providing the user with objective information on which to base decisions. A credit score is an example of predictive analytics that enables banks and other lending institutions to evaluate a potential customer’s ability to make loan payments on time.
While predictive modeling is an alerting mechanism, prescriptive analytics go a step further by offering advice, or the “next best action.” An everyday example is a traffic navigation application that identifies possible driving routes and recommends the speediest one based on an evaluation of multiple factors.
How are insurance companies using predictive modeling?
PD: Big data has application in many areas of insurance operations. A McKinsey analysis found that insurers using predictive modeling and analytics can see better loss ratios and new business premium growth, as well as retention.1 A primary application of predictive modeling at The Hartford is the in the area of workers’ compensation claims. As far back as 2008, we deployed a model to identify potentially high-risk workers’ compensation claims before they become volatile. We patented the model that launched and it’s been rebuilt and improved upon twice.
Why is predictive modeling important for workers’ compensation claims?
PD: Predictive modeling uses data analytics to identify potentially high-risk claims before they become volatile. Volatile claims are those that tend to spiral into a costly succession of complex conditions that the examiner could not have anticipated upon first review. That could be due to the nature of the injury, treatments, barriers to recovery or third parties, like a medical provider or attorney.
On average, only 3% to 4% of workers’ compensation claims are deemed volatile. However, they can generate 50% of a company’s loss costs according to The Hartford’s statistics. Predictive modeling helps handlers identify these claims to take corrective action before they escalate.
How would the claim examiner intervene on a nominated claim?
PD: The claim examiner can tailor intervention to the nuances of an individual case. For instance, the examiner may:
- Assign the claim to the most experienced adjusting, medical and legal resources
- Increase the level of management review
- Engage a nurse case manager
- Bring in a medical professional to help make treatment changes
- Conduct a drug review
- Obtain a second opinion
- Provide return-to-work coordination
Optimally, the claim team fosters a relationship with each claimant that allows for a customized solution. This doesn’t just help the claimant’s injury, but it also addresses other barriers to recovery, such as:
- Addiction
- Chronic pain
- Anxiety
- Depresión
Predictive models can assist here, too, drawing out claim-related conditions that might ordinarily remain hidden.
Can you provide an example of this?
PD: Imagine an employee slips and falls on wet flooring at work and injures their back. Their doctor prescribes physical therapy and painkillers, advising the employee that they should be safely back to work in a few weeks.
Unbeknownst to the claim examiner and the doctor, the employee is facing personal issues that are taking a toll on their emotional well-being. They also lack the motivation to take charge of their recovery.
The employee takes the prescribed medication for pain relief but misses most of their physical therapy appointments. When weeks pass without any improvement to their back condition, the doctor orders an MRI and recommends surgery. To make matters worse, the employee becomes dependent on the pain medication.
In this case, lack of incoming claims for physical therapy could prompt the model to issue an alert. Upon investigating the matter and its cause, a behavioral therapist could be assigned to help the employee work through personal issues that can prevent them from taking charge of their health.
How is this improving claim outcomes?
PD: Predictive modeling helps put the right claim into the right process and assign the right resources at the right time. This enables claim teams to deliver the attention that volatile or complex claims require. And it helps minimize over-handling claims that are more routine in nature.
The effects may not be obvious to the employer, but when predictive modeling achieves its goal, they realize earlier closure rates for claims and faster return-to-work timeframes for their employees. Their cost of risk is also likely to go down. This is especially true for risk managers and insurance buyers with large deductibles given that they share a significant part of the workers’ compensation risk.
Predictive modeling helps to ensure that claimants receive the timely, tailored support they need – beginning with the day their claim gets filed all the way through to its resolution.
Tell us more about The Hartford’s use of predictive modeling for claims.
PD: Our claim predictive modeling portfolio spans property and casualty lines of business as well as group benefits. We have models for everything ranging from fraud, subrogation and reinsurance to large loss, claim volatility and triage.
Within claims, we’ve concentrated our energy on text mining of unstructured data. That’s where the secret sauce is, where the nuances of the claim are revealed. We have patents on the processes we use to take in that data every day, translate it from whatever language it’s in. This can include:
- Nomenclature
- Shorthand
- Medical terms
- Prescription drug names
Then we apply labels to it and recognize patterns that allow us to make conclusions to improve claim handling. Our text mining engine is very robust, which enables us to quickly identify higher-risk claims to ensure that these claims are assigned to our more experienced adjusters.
What staffing resources make your predictive modeling at The Hartford possible?
PD: The success of our models depends upon input from a cross-functional team with a broad range of skills and experience, including claim handlers and supervisors, experts in strategy and workflow processing, data scientists and risk engineering consultants. We also bring in our actuarial and finance partners to evaluate how well our models are working.
Our texting mining engine is the work of an in-house team of five “knowledgeable engineers,” who build the dictionaries, develop the pattern-matching rules and evolve these capabilities to new use case and sources of data.
Do you think predictive modeling could ever replace the role of the claim handler?
PD: Both data and people are critical to success. At The Hartford, the claim handlers are primary. Our objective is to wed the qualities of great claim handlers with the tools that allow them to do their jobs better and more efficiently. The predictive model is just one of the tools within the handler’s book.
How does predictive modeling improve the claim handler’s performance?
PD: An effective model puts every handler’s performance on par with the best. It studies what top claim handlers do, looks for what they look for, systematically finds what they find and alerts everyone in the claim process who needs to know. The model is also consistent. Even the best handlers aren’t at the top of their game all the time, but the model never has a bad day.
We’ve also trained our model to detect symptoms of lack of rapport, whether it was lost at some point in the claim process or never established in the first place. A handler who is alerted to this potentially negative sign is better positioned to:
- Restore communication among all parties
- Address any issues
- Get the claim back on track
What do you see ahead for the future of big data?
PD: At The Hartford, big data will play an increasingly important role, helping us achieve greater efficiencies and better serve our customers, while also empowering a new generation of employees.
Are there any best practices to share with employers looking at insurers who use predictive modeling?
PD: When evaluating a potential insurance carrier, risk managers should first look for insurers whose models are based on years of historical claim information. Robust models are built on data. More data is better. Models that capture the fine details of claims in process will be better equipped to reveal hidden factors and identify potential outliers.
Successful intervention is equally essential. The insurer should have the expertise needed to respond with an action plan that delivers results.
Finally, it’s one thing to have data science available, but more important to find an insurer with a demonstrated track record of implementing predictive modeling algorithms effectively in their workflows. The real magic happens not when the analyst designs the program, but when the business uses it to its fullest advantage to drive better decisions and outcomes.
Learn more about The Hartford’s specialized large business insurance offerings and the unique services and programs that are part of its WC insurance cobertura.
1 McKinsey & Company, “How Data and Analytics are Redefining Excellence in P&C Underwriting”
La información proporcionada en estos materiales brinda información general y de asesoría. It shall not be considered legal advice. The Hartford does not warrant that the implementation of any view or recommendation contained herein will: (i) result in the elimination of any unsafe conditions at your business locations or with respect to your business operations; or (ii) will be an appropriate legal or business practice. The Hartford assumes no responsibility for the control or correction of hazards or legal compliance with respect to your business practices, and the views and recommendations contained herein shall not constitute our undertaking, on your behalf or for the benefit of others, to determine or warrant that your business premises, locations or operations are safe or healthful, or are in compliance with any law, rule or regulation. Readers seeking to resolve specific safety, legal or business issues or concerns related to the information provided in these materials should consult their safety consultant, attorney or business advisors. All information and representations herein are as of February 2022.
Links from this site to an external site, unaffiliated with The Hartford, may be provided for users' convenience only. The Hartford no controla o revisa estos sitios. La provisiòn de cualquiera de estos enlaces no implica la aprobación o asociación de The Hartford con dichos sitios. The Hartford no es responsable y no ejerce ningún tipo de representación o garantía relacionadas con los contenidos, integridad, precisión o seguridad de cualquier material publicado en dichos sitios. Si usted decide ingresar a sitios que no pertenezcan a The Hartford, lo hace bajo su propia responsabilidad.
The Hartford Financial Services Group, Inc., (NYSE: HIG) operates through its subsidiaries, including the underwriting company Hartford Fire insurance Company, under the brand name, The Hartford,® and is headquartered in Hartford, CT. For additional details, please read The Hartford’s legal notice at https://www.thehartford.com.
