Intellect SEEC, a global fintech company just concluded analysis on the incoming quote submissions to seven major United States commercial carriers between January 2017 and January 2018. The results are stunning.
While it’s common knowledge in the insurance industry that clients or brokers, on occasion, portray risks in a better light to obtain better rates or to get sub-par risks through underwriting, the extent of this misinformation and its impact on pricing and loss ratio has been statistically proven for the first time through the use of Intellect’s big data and AI toolkit.
Intellect found that up to 30 per cent of submissions had material errors on key rating and underwriting fields, such as employee count, revenue, details of business operations, and key disclosures about operational practices, past trading, and safety history.
A way to plug underwriting leakage
Historically, it has been near impossible for insurers to check submission accuracy through structured data and traditional means. They relied on denying claims if a client misled the company – which isn’t effective for many reasons. Most policies do not result in claims, and for those that do these errors mostly remain undiscovered.
Consequently, insurers face revenue leakage through lower premiums and higher claims incidence rates. The potential loss ratio impact of such errors is estimated between three and five per cent and Intellect’s work with US Insurers in 2017 to remediate such data discrepancies has validated this.
Building a risk’s digital footprint
There has been a lot of hype about the use of social network data. However, this is not a practical source for insurance as most clients do not have underwriting-pertinent information on their social profiles. Furthermore, to use such data, its veracity and authenticity must be proven and it should be appropriate from a regulatory perspective.
Risk-pertinent information is usually found hidden in many different sources, such as legal and government sources, press and review sites, trade and occupational sources, geo-spatial imagery, and other proprietary sources which are often not openly accessible. Even with access to these sources, manually assessing a client’s digital profile is simply not possible.
For computers to accurately extract underwriting information with certainty from an unstructured source is a complex art. The consequences of having false positives or wrong data, particularly when used in automated risk assessment models are very severe – financial, reputational and regulatory.
Confluence of AI and big data technology
Intellect has been building a network of data sources, which over the last four years has grown to 1800 plus, to cover different risk types. To analyse this varied data set requires the deployment of multiple AI branches such ML-based NLP and image recognition.
Furthermore, different sources can have different versions of information about the same fact, specific algorithms are required to triangulate and validate data. Given the complexities of insurance, creating algorithms to deal with such issues at scale, speed and certainty has been a challenge which requires highly specialised development across multiple technologies and is why progress was slow despite the hype.
As newer technologies like IoT and blockchain mature, and the connected device world grows, the universe of data will expand and the applicability of these techniques will become ubiquitous to all aspects of insurance.
Pranav Pasricha is the CEO of Intellect SEEC.