Artificial Intelligence Review of Insurance Contracts – Are we there yet?

by Joe Chvasta, JD, MBA, CPCU

The use of Artificial Intelligence to review insurance contracts is growing as AI tools can rapidly compare a high volume of contract clauses and make human reviewers faster and more effective. First implemented in 2011, the technology automating the review of contracts has progressed dramatically, leading to the question if AI technology by itself is ready to totally automate the review of contracts or if it remains a tool to assist humans in their contract review.

Insurance companies are adopting AI contract review for greater efficiency.

Insurance companies are joining law firms and other large corporations by adopting AI for contract review due to its efficiency compared to using lawyers, paralegals and other persons. LawGeex, a leader in AI contract review, is used by three of the five largest insurance companiest. LawGeex estimates that contract review by AI saves 80% of the time and 90% of the cost and results in deals closing three times faster. Kira Systems, a provider with insurance company and law firm clients estimate cost savings at 60% while other sources estimate total cost savings at a still hefty 50%. One insurance company leading in the use of AI contract review is Swiss Re which writes about 20,000 contracts a year using an AI tool to compare single clauses against the entire data set expediting the underwriting process. Some in the industry believe that comprehensive contract review without AI is impossible. One analyst commented that an insurance company could not hire enough underwriters and lawyers to review a high volume of contracts with any depth unless they used AI.

What is Artificial Intelligence?

Artificial Intelligence consists of many related technologies including: “Machine Learning,” a subset of AI in which computers use algorithms and statistical models to identify patterns in data and adapt with experience and without the need for reprogramming. Machine learning makes predictions based on data and uses techniques such as decision trees.

“Deep Learning,” a type of machine learning loosely modeled on the neural pathways of the human brain. Deep learning is a self-learning model where algorithms attempt to model high level abstractions in data and requires more computational power than other types of machine learning. “Natural Language Processing (NLP),” a type of machine learning with the ability of a computer to understand, analyze, manipulate, and potentially generate human language. NLP is useful for identifying and tagging phrases in a document. “Optical character recognition (OCR)” is a type of natural language processing.

How does AI review contracts?

AI is used both to review individual contracts and a high volume of contracts. The process of contract analysis for a large volume of contracts typically involves five steps:

1. Importing contract documents into a database. This can be done multiple ways including drag-and-drop, upload directs, imports, etc.

2. The AI software automatically converts the files into machine readable forms. The software works with numerous document formats. Then OCR or another AI tool is used to extract key items and clauses as per specified by the user and tagged for further review.

3. Place the contracts with tagged clauses into a searchable repository. With some software the user may participate in the review before contracts are stored.

4. Use AI such as machine learning to identify patterns in the data and analyze them looking for variances from a set standard wording.

5. Provide customized reports with charts and summaries for almost any data point selected by the user.

By comparing many like clauses AI can identify the core concepts found in each clause and conversely, identify the variances from core concepts. For example, an insurance company may want to review all of the waiver of subrogation clause endorsements found in its policies to confirm that the waiver is mutual, the core concept of the clause. If the AI tool finds that some of these clauses lack mutuality it would flag these contracts for further review. Another example of a clause subject to review on a high volume in an insurance contract is a cure clause such as how long does the party at fault have to correct the default. AI can look for variations from the base wording on the number of days to cure a default and if there is a variance for further review flag it. Properly used AI would facilitate keeping a group of contracts uniform.

Human Review of flagged clauses. A human ultimately reviewing the contracts flagged by the AI tool is expected. At this time algorithms used by AI do not have the judgment for a nuanced read of a contract clause. Similarly, the algorithms lack the nuanced understanding of how conflicting contract clauses work together.

AI models have different approaches to review by a human. LexCheck streamlines the review process by automatically correcting contract provisions, highlighting issues for review and proposing possible fixes. LawGeex uses AI to analyze contracts one at a time where the user selects from a list of clauses and variations to require, accept or reject. The AI scans the contract to see what clauses and variations are present. The language at issue is highlighted with a green thumbs-up or red thumbs-down per the user’s guidelines.

3. Most providers of AI contract review allow users to review features and increase or decrease the level of human review. More human review will be more expensive. On the other hand, there are situations where additional human review may be necessary as when the policy language is vague or where the clause at issue may at some point be subject to regulatory scrutiny. In September 2020, Kira Systems adopted “Answers and Insights,” a smart field for contract review. Users can ask the software a question about the data, such as “will this policy renew automatically?” and receive a straightforward answer. Alternatively, the software also allows the reviewer to pick from pre-selected questions and choose answers in either yes/no or multiple-choice formats, a capability which users have long requested as a way to speed up the review process.

When is AI better?

The tremendous speed of AI in reviewing a high volume of contracts for uniformity allows it to quickly separate contracts with acceptable clauses from those needing further review. This is perhaps the greatest value of AI in contract view – simply reducing the number of contract clauses needing human review.

Another advantage is that the results of the AI tool derived from the data can be validated statistically. The data results would be able to estimate the accuracy of a model’s predictions on unobserved cases. Accordingly, the user would have an understanding of the likelihood of a certain event. For example, if the AI tool found that the original sample of insurance policies all had a mutual waiver of subrogation, AI can use the data from the sample to predict the likelihood of other policies with the same characteristics having the same mutual waiver of subrogation.

AI is also better when the language of the contract is in a language other than the human reviewer’s native language as it can translate the contract to the reviewer’s native language. The value of this is apparent when the reviewer is managing a large volume of contracts in multiple languages.

In insurance policies AI is especially valuable in flagging policy renewal dates so that they can be reviewed for changes before the regulatory deadline for policy revisions on a policy renewal. AI can also identify when an insurance policy requires a future obligation by the insurer, such as a dividend or return of premium. Likewise, AI can monitor contractual obligations by an insured including premium payments and changes required by a safety inspection without the need of a human to track compliance by the insured.

When are humans better? There’s a reason that current AI software for contract review is designed to have additional review by humans. Humans are better at understanding the nuances of a vague contract clause. There are other functions in the contract process besides review where humans are better than AI. Humans are better at strategy, creativity, judgment and empathy and the activities that use these tools including:

· Advocacy.

· Negotiation.

· Structuring of the initial contract.

· Making revisions outside of a defined template.

· Correcting or explaining contracts subject to regulatory scrutiny.

· Understanding the nuance of vague contract wordings.

· Understanding the nuance of contractual clauses that conflict with other clauses.

Humans are better when there is an issue with a regulatory or compliance matter. One of the issues with AI is understanding how the algorithm made the decision. In insurance this would be especially important when a policy is being non-renewed or cancelled. A human, typically working from written guidelines, can better explain their decision process to a regulator than an AI tool that used thousands of factors in its decision. The insurance company representative in this situation would have a difficult time explaining which factor or factors the AI tool found determinative.

The list of activities where humans are better is generally with the contract process itself. For contract review humans alone cannot match the speed and efficiency of AI in reviewing a high volume of contracts or even a single contract. The areas of contract review where humans are better involve nuance such as vagueness in a clause or vagueness due to conflicting clauses and this review can be done in conjunction with AI tools, too.

Are we there yet?

The value of AI when reviewing similar clauses with a large volume of contracts is unmistakable. The question is what level of human review is necessary to ensure that the contracts under review are meeting the requirements of the insurance company and its regulators for acceptable language. One approach is to use AI for first and final review. In the initial review AI will separate the contracts that meet all user requirements from those requiring additional review. Then humans with or without further assistance from AI will review the clauses for corrections so that they are correct and uniform. Then AI will conduct a final review for uniformity.

Even as AI develops humans will continue to take the lead in creating new contracts where our strategy, creativity, judgment and empathy are not available to AI. Similarly, humans will continue to negotiate drafts of contracts. There is no answer to the question of “Are we there yet?” as it depends on what we are expecting AI to do. Once humans set the parameters, are we expecting AI to do a total review of a contract without further human involvement? Or are we expecting AI to tell us what’s wrong with the contracts by pointing out variances without having received human input during the review process? The question becomes how much thinking do we want AI to do when reviewing contracts? As there are no commonly accepted goals, I suspect that we will not have the answer to “Are we there yet?” anytime soon.