Skip to Main Content

Harnessing AI and ML through Discovery Navigator to save time and money on medical records review

It has been about 15 years since I last practiced law and it is fair to say that a lot has changed in that time. Back in the day, I would love it when the mail arrived (the USPS kind) because that is how we received the majority of important correspondences or medical records and it was a great way to not only move your cases forward, but also an opportunity to capture some billable time. Then, when it was time to write a letter back, I would break out the dictation machine and record on a small tape: “Dear John, I am in receipt of your correspondence and would like to …” The tape would “magically” (thanks to my amazing assistant) convert into a word document, and then it would be reviewed, printed, signed, and sent back via USPS. Unless it was time sensitive, then we would overnight it!

Fast forward to today - now everything above can literally be done on your cell phone while walking to your car and saying “Hey Siri…” The practice of law, the insurance industry, and the business world have all fundamentally changed with the evolution of things like the internet, email, PDF’s, DocuSign, and more recently the cloud. But “Hey Siri” is really interesting. It uses the power of Artificial Intelligence (AI)[1] and Machine learning (ML)[2] (Please be advised that, for demonstration purposes, we relied upon/cited to ChatGPT[3] as source material for certain information provided in this article and indicate as such in the endnotes below) and is an example of how quickly and easily these technologies can further evolve our daily routines. It is already making major shifts with things like Waze, social media, E- payments, and facial detection.  

In this article, I examine how AI and ML are revolutionizing the medical record landscape – and how Verisk can help you capitalize on this technology to reduce claim costs, improve resource efficiencies, and help get claims settled.

What exactly is AI and ML?

No bones about it, I chose law school because I couldn’t do math or sciences. So, my goal in this article is to be overly simplistic on the technology talk and leave the in depth understanding of these fields to the data scientists. But I believe laying the foundation is helpful because words and abbreviations like ML, AI, NLP (natural language processing), RPA (robotic process automation) all get thrown around a lot these days. So, let’s start by exploring AI and ML with the help of ChatGPT.  The emergence of Chat GPT– is just one example of how quickly this evolution is occurring. ChatGPT is estimated to have reached 100 million monthly active users in January of 2023, just two months after launch, making it the fastest-growing consumer application in history.[4] In comparison, it took TikTok about nine months after its global launch to reach 100 million users and Instagram 2.5 years.[5] As mentioned above, Chat GPT was used for many of the endnotes and basic research in this article highlighting its ease of use.

Artificial Intelligence (AI) simply refers to the development of intelligent machines that can perform tasks that typically require human intelligence such as perception, reasoning, learning, and decision making.[6] AI encompasses a broad range of techniques and approaches, including machine learning, natural language processing,[7] computer vision,[8] and robotic process automation.[9]

Machine Learning (ML), on the other hand, is a specific subfield of AI that focuses on the development of algorithms and models that enable machines to learn from and make predictions or decisions based on data. ML involves training a machine using large amounts of data, where the machine can learn and improve its performance over time. A good analogy to understand the distinction between ML and AI is to think of AI as a toolbox and ML as one of the tools inside. AI has the goal of creating intelligent machines and ML is one of the techniques used to achieve that goal.

These tools are not only here to stay, but they will continue to change the way we work and evolve our industries at a rapid pace.

The problem: time consuming manual processes

What we have set out to do at Verisk is to use AI and ML to bring together our unique data with our medical, legal, and claims experts with the goal of harnessing technology to solve a complex problem: organizing, sorting, reading, and analyzing unstructured medical records. And let’s be clear – this is no easy task.

Organizing medical records for review

If you are involved in the legal, insurance, or healthcare space and deal with injury claims (workers’ compensation, liability, no-fault), can you think of any other task that results in more work and more time spent than dealing with medical records? It doesn’t matter what type of litigation it is, medical records are the foundation of every bodily injury claim and it is critical that these records are complete, organized, and thoroughly reviewed.

It is also important to note that in a workers’ compensation claim this process – collecting and reviewing medical records – is a moving target as a claim progresses from date of injury to maximum medical improvement, to date of settlement. Updates and new records are typically being added to the claim file while discovery may be ongoing.

And yet, these records are critical to understanding things like causation, assessing damages, and forecasting treatment and settlement potential. The insights gathered from a review of these records, include: documentation of current treatment, identification of any previous injuries or pre-existing conditions, changes in condition, subsequent injuries, prescription drugs, ICD codes, etc. Therefore, this process is essential to ensure that the records are complete and tell the entire story.

Back in the day (again), when medical records would come in the mail at my law firm they would be:

  • Reviewed
  • Organized
  • Date sorted
  • Added to the medical file in chronological order
  • De-duplicated
  • Summarized, and then…
  • The client billed

Rinse and repeat when the next stack of documents came in the mail. And all of this was done manually, gobbling up huge chunks of time.

Honestly, other than the delivery method, that process really has not changed much in 15 years. A good question to ask is how many people are doing these activities and how much time are they spending on any one claim? There are usually numerous professionals involved:

  • Adjusters
  • Nurse reviewers
  • Attorneys
  • Paralegals
  • Administrative Assistants
  • The IME / QME / AME
  • Expert witnesses
  • Physicians providing a second opinion

I likely missed some, but you get the point – so many hands in the process and so much time spent dealing with medical records. Let’s plug in some data from Chat GPT and do some very rough math on a few of the above just as examples:

Adjuster’s time and skills are being wasted on clerical tasks

This one is painful for insurers and self-insureds. The average time spent per day by insurance adjusters reviewing medical records can vary depending on several factors such as the complexity of the case, the number of medical records to review, the experience level of the adjuster, and the specific requirements of the insurance company. However, in general, insurance adjusters may spend anywhere from two to six hours per day reviewing medical records.[10]

So, let’s take the median – and say that the average adjuster spends 4 hours per day reviewing and organizing medical records. That is literally half a day spent reading, highlighting, sorting, and getting an understanding of the medical treatment involved in a claim. If an adjuster had four extra hours a day, how much more effective could they be? How much more time could they spend on the complex and difficult tasks and the files that they are uniquely qualified to handle?

Between the great resignation and an aging workforce, every minute an adjuster can spend on the tasks they are uniquely qualified for, such as settling claims, is critical. Additionally, while adjusters may skim medical files to save time, is that really the most efficient use of time and is it driving the best quality? Even the best adjuster can miss critical information buried in a medical document. Skimming a file isn’t the answer: automation is. The proper use of automation plus the skills of an adjuster reviewing the right files at the right time will drive the best outcomes.  

Higher lawyer and physician review costs

The amount of time that a bodily injury attorney typically spends reviewing medical records can vary depending on several factors, such as the complexity of the case, the volume of medical records, and the lawyer’s experience and expertise. It is common, however, for lawyers to spend several hours reviewing medical records per day. For example, in a personal injury case, the lawyer may need to review medical records from multiple healthcare providers, including emergency room records, hospital records, diagnostic test results, and records from specialists. The lawyer will need to carefully analyze and understand the medical information in order to build a strong case for their client, which can involve identifying the nature and extent of the injuries, assessing the long-term impact, and determining damages. According to Chat GPT, cases may require several days or even weeks of analysis and review by the lawyer and their team.[11]

Similarly, the cost of reviewing medical records by doctors can vary depending on the complexity, the type of case, the amount of time required to review and/or organize the records, the geographic location, and the specialty of the doctor. According to a survey conducted by the American Medical Association, the average hourly rate for reviewing medical records in 2021 was $238 per hour for non-specialists and $301 for specialists[12]. It is important, however, to note that some doctors may charge a flat fee, while others may charge a per page or per hour rate. But if they are spending several hours reviewing and organizing records, the costs add up quickly.

As highlighted by the California Workers’ Compensation Institute, they reported that in 2021, the average payments for all comprehensive medical-legal evaluations climbed 67% under the new fee schedule in effect.[13] The new fee schedule provided a single flat rate of $2,015, which includes the review of up to 200 pages of records and the reimbursement rate for reviewing additional records is $3 per page. By way of reference, in our Medicare compliance business, our average record count is 600 pages per file, and it is not uncommon to see cases with 1000+ pages.  

You get the point: the costs of time and money dealing with medical records is significant for many people involved in a claim on both the Plaintiff and Defense side. That same effort is often duplicated many times over on the same file. The records come in through discovery at various times and multiple people are required to review them, organize them, or summarize and analyze them to properly move a claim forward. Over the course of a claim, which can last for years, these costs really add up.

So, how can we bring AI and ML to the table and not only help reduce expenses, but also give everyone back some time to focus on the difficult tasks that they are uniquely qualified to handle?

The solution: Verisk’s Discovery Navigator

The good news is you can leave the old behind and enter a new era of efficiency. Our Discovery Navigator is a purpose-built tool to meet the age-old challenges above with cutting-edge AI and ML. By automating the administrative efforts of medical record review, Discovery Navigator simplifies the process and helps claims professionals use their skills for much more targeted review and analysis—all while reducing costs. Discovery Navigator is the next wave of claims automation.   

Records processing is simplified and improved

Discovery Navigator can successfully review unstructured documents to identify and extract key terms and medical data resulting in expedited file reviews. Cracking the complexity of unstructured documents was key to developing this solution. Medical records are extremely complex because the data is not in a structured format. Just about every doctor, hospital, office, or provider has a different form or format for maintaining records. Some are typed and some handwritten. Some provide a background on the injuries and prior history, others do not. Some show lab results and others talk about the results in the body of the document. Some start with a description of the injury, others start with a prior history. You get the point.

Multiple unstructured records from multiple providers, in multiple formats, are provided at various times throughout the life of the claim. That is extremely complex to make sense of and it is also why so many people are involved in requesting, sorting, organizing, and summarizing medical records today. 

Enhanced capabilities for improved review and analysis

Discovery Navigator was developed internally by our expert medical, legal and analytics team to automate the administrative aspect of medical record handling within our business. As a data and analytics company, we have access to 19 petabytes of data across our proprietary databases. To put that into context, a petabyte could store around 200 million hours of high-definition video.[14]  Included in that data are millions of medical and claim specific data points that we used to build and test our models.

But data alone was not sufficient given the unstructured nature of the medical documents and the need for accuracy. Our clinical experts reviewed medical records line by line to augment the machine learning aspect of Discovery Navigator over an 8-year period resulting in industry leading accuracy and results. It truly was a combined effort involving legal and clinical experts and data scientists who used their expertise in statistical analysis, computer programming, medical, and injury claims knowledge to extract the critical insights necessary to build this tool and make Discovery Navigator so effective. It was not until we proved its impact in our own business that we realized the value this brought to the insurance, legal, and medical fields.

What will Discovery Navigator do for you?

  • Organize records and put them in chronological order
  • Remove duplicates
  • Provide a timeline of treatment
  • Provide an index with hyperlinks to corresponding treatment
  • Retrieve information such as ICD Codes, medications, and comorbidities
  • Tag medical terms to allow filtering
  • Identify information available in an easily uploaded table or document
  • Ability to use the data captured for analytics

Discovery Navigator delivers real results and savings

As a result, users can achieve significant cost and time savings. How much time can be saved? Verisk research has shown the average person,  takes about two minutes per page to review medical records. Discovery Navigator can do the job in six seconds per page, representing time savings up to 90% with accuracy rates up to 95%.

This tool was designed for ease of use and can integrate directly into your platform or claim system via API. It is also available instantly via Verisk’s website.

As a data analytics and technology partner to the global insurance industry, we are excited to share this solution with you and your organization. Verisk has only begun to explore Discovery Navigator’s applicability and its potential to revolutionize the handling of medical records for the insurance and legal industries.

Learn more about Discovery Navigator

We are confident Discovery Navigator can help you take claims handling to the next level and would look forward to setting up a call to discuss how we can help. 

Download this brochure to get additional details. 


Please do not hesitate to contact the author if you have any questions or contact us to learn more.

[1] AI has applications in many industries, including healthcare, finance, transportation, and entertainment, among others, and it is expected to play an increasingly important role in shaping the future of society. (ChatGPT, personal communication, May 1, 2023).

[2] Machine learning algorithms are designed to identify patterns in data and use these patterns to make predictions or decisions. Machine learning has many applications, including image recognition, speech recognition, natural language processing, recommendation systems, and fraud detection. (ChatGPT, personal communication, May 1, 2023)

[3] ChatGPT is an AI language model created by OpenAI, based on the GPT (Generative Pretrained Transformer) architecture. It is one of the largest and most powerful language models available, with the ability to generate human-like responses to text input. ChatGPT was trained on massive amounts of text data, allowing it to generate coherent and contextually relevant responses to a wide range of topics and questions. It can understand and respond to natural language input, including conversational text, making it well-suited for use in chatbots, customer service applications, and other conversational interfaces. As an AI language model, ChatGPT can be used to generate text, answer questions, provide recommendations, and even carry on a conversation with users. It has the ability to learn from new input, which means it can adapt and improve over time as it encounters new data and experiences. (ChatGPT, personal communication, May 1, 2023)


[5] Id.

[6] (ChatGPT, personal communication, May 1, 2023)

[7] NLP is a branch of computer science and artificial intelligence that deals with the interaction between humans and computers using natural language. NLP is concerned with various aspects of language processing such as syntactic and semantic analysis, machine translation, sentiment analysis, speech recognition, and text-to-speech synthesis, among others. The ultimate goal of NLP is to enable machines to understand human language and to communicate with humans in a way that is natural and intuitive. (ChatGPT, personal communication, May 1, 2023)

[8] CV is a field of study within computer science and AI that deals with enabling computers to interpret and understand visual information from the world around them. This includes the development of algorithms and techniques to extract and analyze information from images and videos. Computer vision aims to enable machines to automatically recognize objects, detect patterns, and extract meaningful insights from visual data. Some of the applications of computer vision include facial recognition, object detection, image segmentation, and autonomous navigation. The ultimate goal of computer vision is to enable machines to perceive and understand the world in the same way that humans do. (ChatGPT, personal communication, May 1, 2023)

[9] RPA is technology that allows software robots to automate repetitive, rule-based tasks that are typically performed by humans. These robots can interact with digital systems, such as computer software and web applications, to perform tasks like data entry, data extraction, and report generation. RPA works by replicating the actions of a human user interacting with a software application. The robots are programmed to follow a set of rules and procedures, and can work 24/7 without interruption, increasing efficiency and productivity while reducing errors.

[10] (ChatGPT, personal communication, May 1, 2023)

[11] (ChatGPT, personal communication, May 1, 2023)

[12] (ChatGPT, personal communication, May 1, 2023)


[14] (ChatGPT, personal communication, May 1, 2023)

Robert T. Lewis

Robert T. Lewis is a Senior Vice President of Innovation and Business Development at Verisk.

You might also like

Visualize Subscribe

Get the best of Visualize!

Get the latest news and insights straight to your inbox.

Subscribe now

You will soon be redirected to the 3E website. If the page has not redirected, please visit the 3E site here. Please visit our newsroom to learn more about this agreement: Verisk Announces Sale of 3E Business to New Mountain Capital.