For as long as there has been insurance litigation, there has been a need to efficiently review and capture information from insurance policies and to analyze it effectively. Insurance policy reviews are valuable tools for coverage counsel, but they can be a significant upfront cost to policyholders, who may already be suffering from a lack of insurance cover. Policy reviews can also take weeks to months, limiting their application under tight deadlines.

Luckily, the methods for reviewing policies continue to evolve. I have heard horror stories of young associates making mountains of copies, then cutting and pasting clauses into different binders. I can recall many hours as a young consultant spent on time-consuming, manual data entry to capture text — a process that necessitated equally detailed quality control measures, making sure each comma and space was correctly captured.

Since my first policy review at KCIC in 2004, we’ve successfully leveraged improvements in Optical Character Recognition (OCR) technology to decrease the amount of typing. At the same time, we have cultivated a database of nearly 300 common policy forms and thousands of standard clauses to further reduce the time and energy spent on data entry. Still, when policy documents are partially legible or we haven’t seen them before, we must resort back to more manual and time-consuming methods.

Where do we go from here? Perhaps you have heard of Technology-Assisted Review (TAR) and predictive coding as emerging techniques being utilized in legal document reviews. The general process is to teach a computer model to identify important documents by feeding it coding data created by human reviewers on a subset of the documents. The remaining documents are then put through the computer model, resulting generally in three sets of documents: relevant, questionable and irrelevant. Rather than reading every line of every page, the team of reviewers focuses mainly on the questionable documents. This practice has already been accepted in some courts as a way to reduce the costs of document production.

At KCIC, we are exploring whether this technology could be adapted to expedite insurance policy reviews. Our database of standard insurance policy language will “teach” the computer model to locate identical or similar clauses in different insurance policies and capture the text directly into our Ligado policy database. Our consultants would then focus on confirming whether every relevant clause was captured. A full read of the policy by an expert will still be needed, but automated extraction of even just 60-70 percent of the relevant policy language would translate into significant time savings.

As we begin developing and testing our concept, I’ll provide an update to share what we’ve learned.