Underwriting is the nucleus of the insurance business. For centuries, human beings have performed this process, evaluating a risk to determine whether or not it is insurable at a profit for the insurance carrier. To this task they brought significant statistical and analytical skills, attention to detail, and judgment.
Executive SummaryVeteran insurance journalist Russ Banham interviews leaders of four InsurTechs focused on making insurance underwriting more accurate and less burdensome, freeing underwriters to take on more strategic, value-added work.
Well, move over people; here come the robots. Through the use of cognitive computing tools like machine learning, predictive analytics, robotics processing automation, and both image recognition and natural language processing, underwriting is becoming less manual and more automated. Providers of the tools offer novel ways for underwriters to better gauge risk, set premiums, save time, become more efficient and lower loss ratios.
We’ve profiled four such InsurTech companies here, each with a different set of products and services, but all with a similar value proposition: to make insurance underwriting more accurate and less burdensome, freeing underwriters to take on more strategic, value-added work.
Will the tools eventually replace the people whom they are currently helping? Read on.
Intellect SEEC: Expanding Information Boundaries
The unusually named Intellect SEEC (the two words reflect the consequence of a merger) is the first InsurTech enterprise in our lineup. Intellect SEEC provides cognitive computing solutions covering multiple insurance functions like underwriting and distribution via a cloud-based platform. The company focuses on commercial lines underwriting services for primarily medium-sized and smaller commercial insurers and specialty carriers.
Pranav Pasricha, Intellect SEEC’s CEO, said the company reinvented itself after the 2009 merger to bring the latest innovations in machine learning and big data to underwriting. “We’re confident that we’re the best source of structured, semi-structured and unstructured information in the world,” he asserted.
Pranav Pasricha, CEO, Intellect SEEC
This information ranges from publicly available legal filings and press articles to customer comments and social media feedback. Intellect SEEC’s tools capture this data and ferret out the most pertinent information from an underwriting standpoint.
“We’re able to distill fine-tuned alerts of information about each class of business—the different things that can go wrong and the insights drawn from this knowledge,” said Pasricha. “Such risk indicators often escape the attention of underwriters, yet are crucial elements of the overall risk picture. We’re expanding information a thousand times.”
He’s not necessarily boasting. A human being could not possibly collect and collate 10,000 pieces of information of import to a particular risk. However, using cognitive computing tools like predictive analytics and machine learning, this huge volume of data is compressed into digestible tidbits of underwriting import.
Intellect SEEC also canvasses historical and real-time data sources to make predictions on future loss likelihood. Examples include an upcoming regulation or possibly adverse legal ruling affecting a potential insured’s business prospects or a competitor’s research into the development of a new product or product enhancement.
“Our Risk Analyst product uses machine learning to look at events occurring around an insurance prospect’s business to assess potential risks down the line,” said Pasricha. “We capture this information and provide it to underwriters in the form of an alert.”
Prior to joining Intellect SEEC, Pasricha was the chief operating officer of QBE Insurance Group in Australia, leading the company’s global underwriting transformation effort. Intellect SEEC’s Chief Technology Officer Lakshan De Silva worked with him at QBE in driving this transformation.
“Next up for us is an extension of our current capabilities, incorporating more video into our telematics to further illuminate the risk profile,” said Pasricha. “We also see the Internet of Things as a huge growth platform, pulling and analyzing data from the embedded sensors to provide added insights to underwriters.”
DataRobot: Powering Predictive Models
DataRobot also digs through mountains of risk-based data to unearth underwriting insights, in its case via an automated machine learning platform. Underwriters interact with the platform to create better risk models.
“We help underwriters get an idea of what an insurance policy will cost over a multiyear period of time, presenting the opportunity for the carrier to improve its risk segmentation,” explained Satadru Sengupta, DataRobot general manager and data scientist.
Satadru Sengupta, General Manager and Data Scientist, Data Robot
The business of selling an insurance policy today is based on an assessment of a prospect’s historical risk and loss data to price the coverage terms and conditions on an annual basis. Scant thought is given the trajectory of the risk five years into the future and what the premium for the policy would need to be at that time. Predictive big data analytics offers a way to gauge this future cost of goods sold to create a more balanced underwriting portfolio.
Armed with this knowledge, an insurer may determine a particular risk provides a greater long-term return than another risk. “We’re providing a way for underwriters to make better predictions that improve risk segmentation and charge a more accurate premium,” said Sengupta. “We tap into different sets of data and automatically apply open source algorithms to help underwriters build highly accurate predictive models that tell a truer story of future risk.”
DataRobot’s cognitive computing platform also is marketed to carriers for claims, distribution and other insurance processes (underwriting represents less than one-third of its market). The platform can be used to underwrite personal lines and commercial lines products, as well as health and life insurance. Users interact with the platform to build hundreds of risk models in a single click, helping them make better predictions. “We make the process of building a risk model extremely simple,” Sengupta said.
Large global insurance carriers are DataRobot’s primary customers, although its modeling tools also are sold to other industry sectors like banking and health care. Nevertheless, insurance would appear to be the company’s sweet spot. Two former insurance executives—Jeremy Achin and Tom de Godoy (both from Travelers)—are co-founders of DataRobot. Sengupta also hails from the industry, serving stints at AIG and Liberty Mutual. And its chief data scientist is a former actuary.
“We’re insurance through and through, from product design and development through advisory and client interactions,” said Sengupta. “We speak the language of insurance and understand the challenges of underwriting.”
He added, “Oftentimes people think analytics is all about the application of algorithms. Not necessarily so, although they are important. What is most critical is designing the workflow. When you merge experienced data scientists with people who have deep insurance domain expertise, you get solutions that address real business problems.”
In 2018 DataRobot plans to incorporate so-called time series analytical modeling into its platform. Last year, it acquired data science company Nutonian to bolster its capabilities to create models involving time series data. The key word is “time.” As the name suggests, the analyses involve predictions generated by time-based data—years, days and hours.
DataCubes: Solving Underwriting Problems
Unlike DataRobot, DataCubes focuses exclusively on developing machine learning and data science tools for insurance underwriters. “It’s all we do,” said Harish Neelamana, DataCubes’ co-founder and chief product officer. “We solve two big problems: overcoming inefficiencies in how underwriters do their job and providing access to better facts to make smarter decisions.”
Harish Neelamana, Co-founder and Chief Product Officer, DataCubes
Regarding the first solution, by digitizing and automating the processing of insurance applications in real time, the company reduces the paperwork migraines involved in the quote-to-bind underwriting process. The solution also comprises a data integration engine that captures and organizes data from multiple external and internal sources.
“We start with a few pieces of information, like the name and address of a business, and then sift through the usual mountains of publicly available data and licensed data sources that describe various aspects of this entity,” said Kuldeep Malik, DataCubes’ CEO and co-founder. “This typically includes how long the company has been in business, the nature of the work it does, how many employees it has and all sorts of other information. We then apply machine learning to this data to answer specific underwriting questions, giving users an Amazon-like experience.”
An example is a landscaping enterprise that mows lawns, cuts hedges and removes dead leaves. These activities help describe the company’s risk profile for underwriting purposes, culminating in a premium charged for the related exposures. However, by scraping data off websites and social media, the underwriter may learn that the landscaper did a great job cleaning out the roof gutters of a particular customer. Unfortunately, this high-risk activity was neither realized nor reflected in the underwriter’s risk assessment and premium calculations.
DataCubes helps to solve this conundrum. “The underwriter can ask the question: ‘Does the landscape contractor do roofing work?'” said Malik. “The tool interprets this to go out and search data about the company. Up pops some information that the company did some roofing work a couple times. Well, roofers fall off roofs, changing the risk profile.”
Most of DataCubes’ insurance carrier customers are in the $50 million to $100 million range (gross written premiums), although some are in the $500 million to $2 billion category, and one is a top-tier $10 billion-plus insurer. “We focus on underwriters of workers compensation and BOP [businessowner policy] packages—general liability and property stuff,” Neelamana said.
Prior to launching DataCubes, Neelamana spent 15 years performing operational and strategic roles at Zurich Insurance Group and Allstate; Malik, on the other hand, is an experienced entrepreneur. He said, “Our team is a sort of happy medium of data technologists and insurance underwriting experts coming together to solve underwriting problems.”
RiskPossible: Continuous Underwriting
RiskPossible is the newest kid on the block, a startup still getting its footing. Like the other InsurTech companies, its founder and CEO Michael DeSiato hails from the insurance industry. His mother and two uncles launched the small Granada Insurance Company, a Florida-based property/casualty carrier, in the 1980s. “My mom introduced both insurance and entrepreneurship when I was a little kid,” said DeSiato, who was in Des Moines, Iowa, taking part in a global accelerator program for startups when interviewed for this article.
Michael DeSiato , Founder and CEO, RiskPossible
The company has yet to make its official launch, although it has participated in several pilot projects. RiskPossible also leverages data access and analysis tools, but for a somewhat different purpose. “We help underwriters find out if a policyholder’s risk profile has changed dramatically since binding,” said DeSiato. “We provide this information through our continuous underwriting engine.”
Rather than underwriting being a once-and-done exercise with an annual reappraisal of client risk, DeSiato wanted to make it more of an ongoing process throughout the life of the policy. His thinking was that important risk-based data was escaping the attention of carriers—information that may compel it to cancel the policy.
“We’ve partnered in a pilot program with a nursing home, providing a continuous feed of risk-based data that our tool has scraped off different public and private sources of information, including social media,” he explained. “Once you go down the rabbit hole, the amount of information is incredible. Based on the insights we learn, an alert would be sent to the underwriter to re-evaluate the risk.”
DeSiato provided the following scenario: a nursing home whose fire and smoke doors were recently inspected to ensure compliance with a new rule from the U.S. Centers for Medicare & Medicaid Services (CMS) covering the installation, care and maintenance of many types of doors and assemblies in a healthcare setting. If the company fails the test, this information typically would not reach the underwriter until just before the policy renewal.
“Say you have a restaurant regularly failing inspections for pests or with multiple infractions of people not washing their hands. Wouldn’t the carrier want to know this immediately?” asked DeSiato. “This way you could send out your own inspector to do a renewal review much earlier in the process. Depending on the state, you may have the ability to do a midterm policy cancellation.”
RiskPossible currently is engaged in a joint venture with a provider of IoT-enabled sensors measuring temperature and moisture. The plan is to feed this data into its continuous underwriting engine in time for the company’s imminent launch.
“We want to put the sensors inside freezers in restaurants to detect drops in temperature causing potential food spoilage, and in commercial buildings to discern evidence of a leak, with the data going to both the insured and insurers,” said DeSiato. “We’re also working with another partner that has developed a tool that counts the number of people going in and out of a facility. All this risk-based data coming from multiple sources has import for underwriting, well before the renewal.”
Back to Those Robots
As these stories relate, machine learning and data science technology should make the job of underwriting easier and more efficient and productive. But will the tools eliminate the need for underwriters in the future?
All the interviewees demurred on the point. “The day a machine does what human underwriters do is the day there is nothing left for anyone to do,” said DeSiato. “Underwriting requires three things: intellectual curiosity, domain knowledge and creativity. This is what human beings provide. At best, the tools will help underwriters enhance their portfolios and productivity. They won’t replace people—not any time soon.”
Pasricha from Intellect SEEC has a slightly different perspective. “In the future, every job is going to be disrupted by machine learning, including those of underwriters,” he said. “But this doesn’t mean underwriters will be replaced entirely. An important job in the future will be training the machines to underwrite—something that only the best underwriters will inevitably do.”
DataRobot’s Sengupta concurred: “Underwriters will be different in the future, but the jobs are not going away. As machines take over the rote jobs, underwriters will have more time on their hands to focus on emerging risks like cyber, where there isn’t much data yet to draw from. Machines will extract this data as it increasingly becomes available, but human beings will be needed to assess its meaning.”
“As robots allow underwriters to be more efficient and make more intelligent decisions, they will be freed to spend more time on building a better book of business,” said Neelamana from DataCubes. “The position itself will be occupied by highly intelligent people of enormous importance to the profitability of the carrier.”
Instead of robots replacing people, the interviewees contend that humans and machines will fuse together as one—not in a mechanical sense, of course, but in an intellectual one. Underwriters will not disappear. Instead, they will become uber-underwriters.