Lost in Translation: Insurance Companies Need AI that Speaks Their Language

Natural Language Processing - August 14 2019

In a recent report, VisibleThread evaluated the websites of 54 of North America's largest insurers. The results of their analysis showed that complex legal language was included in 79 percent of their websites, making it difficult for average consumers to easily understand product descriptions, terms and conditions. Among the report’s key findings were these gems:

  • Just two insurance websites are simpler to read than Moby Dick
  • Complex word density is an issue for 100 percent of insurance websites analyzed
  • No insurance site scores at the recommended level of long sentence use

The issue of language use and intelligibility gets a whole lot more interesting when you look at insurance documents like applications, policies, and binders. Travel insurer Squaremouth made news earlier this year when it gave a Georgia schoolteacher a $10,000 reward for actually reading her insurance contract. The insurance company placed an Easter egg seven pages deep in the fine print of the contract inviting the first customer who made it that far to call in to claim their cash reward.

Symptom of a Wider Problem

Things sure don’t get any simpler when you move from personal insurance to commercial lines; far from it. Few would call the language used in commercial insurance documents a light read. Not only do the length and number of pages multiply for commercial insurance documents, but the legalese and jargon tend to increase as well. This can be a real problem for commercial insurance brokers and carriers looking to use artificial intelligence (AI) to extract business insights from mountains of digital insurance documents of various types. Commercial insurance is complex, and the highly specialized vocabulary used by the industry can leave general-purpose AI solutions, well, at a loss for words.

Context Matters

Humans have an innate ability to recognize and understand the nuance and context of language. Machines? Not so much. Here’s a simple example: An AI-powered data extraction solution may recognize a word in a policy document, but if there is no meaning or context to the word, the data extracted from that document is not actionable. For example, a system may recognize that “John Smith” is a name, but is it the name of the broker or the policyholder? Likewise, is the word “orange” in an insurance document a color or a county in California? In order to be effective, AI solutions not only need to read like humans, they need to master the unique lexicon and business logic of insurance at least as well as an insurer’s human knowledge workers on their best day.

A Problem for Humans, Magnified for Machines

It’s worth remembering that even the most highly skilled and experienced knowledge workers and underwriters don’t get it right 100% of the time. In fact, a study found that a whopping 90 percent of E&O claims are the result of errors, not omissions. Human error is an issue that has long plagued the commercial insurance industry, costing insurers billions in E&O exposure.

In a press interview with Insurance Business, Saad Mered, CEO of Zurich Canada, describes the problem:

“The issue of mistakes or leakage in insurance documentation, whether that’s a wrong address, a wrong deductible, or a wrong sublimit on page 36 of an 84-page policy binder, is a huge cause of concern for the whole insurance industry. The industry’s re-work ratio is quite high, but it can be improved dramatically with innovation and technology.”

In another article in Canadian Underwriter, Mered goes on to say that:

“Often commercial policies have dozens of pages with several endorsements, which someone needs to read for accuracy before the policy is issued to the customer. Someone needs to confirm, for example, that all references in a policy and its endorsements to insured address, deductibles and sub-limits are correct…. Quite often, they have to open up three or four screens and compare and contrast visually with their eyes. And you know what happens after page 52. Your eyes get tired and with the best of intentions, people still miss things.”

While humans have the advantage of innately understanding language, AI has a few advantages of its own when it comes to reducing errors and omissions. An article in Time magazine summarizes one of the main strengths of AI:  

"Computers don’t suffer from important limitations that plague human beings. They’re not restricted by biology, they don’t get tired, they can crunch numbers for long hours, and they’re exceptionally smart while doing repetitive mathematical tasks, according to Satya Mallick from LearnOpenCV.com and the founder of Big Vision LLC."

Natural Language Processing (NLP) and AI give brokers and carriers the ability to extract, interpret, classify and analyze unstructured data in submissions, applications, quotes, binders, endorsements and policies – regardless of document type, format, or structure. More importantly, AI solutions designed specifically for commercial insurance can read and understand what these terms or named entities mean in the context of commercial insurance.

Purpose-built AI solutions can recognize hundreds of named entities and insurance-specific data points including name, city, limits, premiums, deductibles, types of coverage, exclusions, endorsements, territories of coverage, outstanding conditions, and more. Basically, these AI solutions can read insurance documents just like skilled knowledge workers do – only hundreds of times faster and with greater accuracy.

Unlike general-purpose AI solutions, purpose-built Natural Language Processing and AI solutions understand the language of insurance. When checking a policy against a binder, for example, they don’t get tongue-tied or confused by specialized language, terms, or legalese. And they never get tired, not even on page 28 of a policy.

Because AI solutions leverage machine learning, they aren’t constrained to rigid rules-based logic. This means that they can extract salient policy-level data from digital documents regardless of document format or structure. This allows the solution to more easily accommodate common changes in insurance documents or forms such as changes to coverage, language, organization, content, or the introduction of new lines of business – without requiring recoding.

Simply Put

By automatically extracting policy-level data from digital documents and handling the heavy lifting of checking policies to mitigate E&O risks, purpose-built AI solutions can free up human knowledge workers and underwriters to do what they do best: listening to and advising customers, assessing and pricing risk, developing new insurance products and growing the book of business. By augmenting human staff with purpose-built AI solutions or “digital employees,” commercial insurance brokers and carriers have an opportunity to boost their underwriting capacity and margins, increase their quoting capacity, and deliver a better digital customer experience. It’s results like these that have commercial insurers exclaiming, “Now you’re talking my language!”

Shopping for an AI solution?

If you’re considering an AI solution that can help you meet your growth objectives, check out our Insurance AI Buyer’s Guide & Evaluation Worksheet. Download your complimentary copy today.

Download Your AI Buyer's Guide & Worksheet

Browse different topics

Recent Posts