Streamlining Insurance Operations: Navigating Claims & Submissions

Greg Hauenstein: We will go ahead and get started. My name is Greg Hauenstein. I am on the Go-to-market team here at Hyperscience, and I am joined by Laura. Laura, would you like to introduce yourself?

Laura Heritage: Hi everyone. I’m Laura Heritage, I work closely with Greg, a little bit more on the product side of things. Great to meet everybody.

Greg Hauenstein: We value you spending some time with us today. Hyperscience has been doing a lot of research and getting feedback from organizations on the leading challenges in insurance operations and how to use enterprise AI to offer a differentiated solution. Our organization started in 2014 with a fundamental hypothesis that we could use machine learning and artificial intelligence to fundamentally improve the life of the back-office worker. One of the critical things all of these organizations have strived for is the idea of hyper-automation: how can they take the most fundamentally challenging human processes and apply machine learning to deliver better outcomes?

Greg Hauenstein: When you look at organizations achieving these kinds of results and leveraging best-in-breed enterprise AI through turnkey solutions, it is the idea of this 99.5% accuracy and 98% automation. These results were fundamentally not possible with the tools available to organizations previously. For example, the Department of Veterans Affairs previously had over a three-week backlog in the processing of claims. After implementing Hyperscience, they were able to reduce that three-week time down to about four hours, which equated to about $46 million a year in costs they were able to reallocate to serve veterans faster.

Greg Hauenstein: The reason previous attempts failed is because the tools fundamentally weren’t serving the organization. Rules-based OCR or early versions of IDP fell into critical blockages where they would get to about 50 to 70% automation, but the rest of the work required BPO processing or RPA to handle exceptions. Leading AI solutions are now giving a novel approach. When you use a rules-driven approach, you only have automation that looks in the rearview mirror. It can only handle things it has explicitly been told to handle. It is not using experience to make a decision looking into the future.

Greg Hauenstein: The process we propose is similar to how you would onboard new employees. First, define the goal you want the AI to achieve. Once we know what we want a model to achieve, we need to train it on what good looks like, usually with the information or documents it will be interacting with. That is the creation of ground truth. Once we have a model, we do quality assurance testing to fine-tune it before deploying it to partner with human counterparts. Supervision is where we use this model and, when it is not confident enough, it partners with team members to fill that gap and uses those interactions to learn and improve over time.

Greg Hauenstein: This interaction allows for high accuracy, which is very important in automation because, for customers in claims, it is the difference between recovering from a loss or not. The quicker we can help customers through automation by reporting their claims timely, we help them through the most difficult times of their lives. We see an advantage of highly automated solutions in giving time back and allowing us to process claims much more quickly. In a moment, Laura is going to go through a demonstration of the platform to give you real-world insight into ways you can use our toolset to create and deploy models. Laura, I will hand it over to you.

Laura Heritage: Thank you, Greg. In this section, we are going to look at the Hyperscience AI platform demo. At the heart of the platform is the Hypercell. This is where we are deploying that digital workforce Greg spoke about inside the enterprise, which is going to automate the processing of human-friendly data. Each of those digital workers is going to perform at any accuracy level you ask it to, regularly achieving 99.5% accuracy and 98% automation.

Laura Heritage: The demo begins with a customer who is submitting a new medical insurance claim. Hyperscience is going to receive that claim and begin to process it. We can see here that we have this inbound human-friendly information in the form of an email body describing the nature of the claim. There is also a slight negative disposition to the email; they are saying it was annoying last time. We also have a number of attachments accompanying this email. The first attachment is a claim form taken from a cell phone. It has real-world issues: it is skewed, there is a pen across the page, things are crossed out, and it is handwritten. The second attachment is an invoice describing the consultation. The third attachment is a handwritten doctor’s note, a truly unstructured document.

Laura Heritage: In order to process this claim, there are three underlying machine learning models working here. The first is a classification model, identifying what types of documents we are looking at. The second model identifies the interesting data across those different documents using advanced AI techniques. The last model is focused on the transcription of that interesting data. We use natural language processing here to reason with that data, helping us drive up automation rates whilst guaranteeing target accuracies.

Laura Heritage: At this point, our digital workforce has decided it can no longer meet those desired accuracy levels and is asking for help from a human worker. In this example, it is a transcription task. These human-in-the-loop tasks take a handful of seconds to complete. The Hypercell has zoomed in directly to the one particular field it wants us to look at, so we do not have to be distracted by the rest of the document. It also flashes red if I am entering something I should not be, helping to maintain the integrity of the data.

Laura Heritage: Once the human tasks are complete, the digital workforce can take this accurate data and build an overall picture of the claim. The Hypercell can integrate with backend systems or use its own knowledge store. In the knowledge store here, I have the conditions for our medical policy. We can see that the digital workforce has completed the view of the claim and determined there are issues. It highlights these issues to the claims assessor. We see the policy details, the claim details, and the negative sentiment picked up. We then see a recommendation from our digital worker to reject this claim. It provides a reason: the symptoms began in February, but the policy did not begin until March. From the knowledge store, the digital worker displays the offending policy to give us full visibility into why it is recommending rejection.

Laura Heritage: In addition, the Hypercell can integrate with Gen AI large language models. In this example, we are able to provide a summary of the supporting evidence, such as the handwritten doctor’s note, to get a quick overview. Our human claims assessor can review all of this information and agree with the digital worker that the claim should be rejected. This decision can kick off downstream actions, such as giving the claimant immediate feedback via a phone call or text, improving the customer experience.

Greg Hauenstein: Laura, thank you for going through all of that. It ties directly to one of the questions we received: Has Hyperscience moved beyond the processing of documents and into the actual delivering of outcomes? Can you expand on the abilities to do knowledge work?

Laura Heritage: Absolutely. The example of the medical insurance claim and our digital worker making a recommendation is a great use case of where Hyperscience has gone beyond data extraction. The Hypercell platform allows us to build comprehensive workflows of how we want document use cases to be processed. It allows a huge amount of flexibility: connecting with external systems, building custom business rules, connecting to Gen AI large language models, storing external data, and performing validations.

Greg Hauenstein: That dovetails nicely into another question regarding how Hyperscience gives clients the ability to reach in and out of other repositories or connect to other enterprise systems.

Laura Heritage: We have out-of-the-box methods available to be built into workflows, such as API lookups into external systems or database lookups. How that data is used within the workflow offers a huge amount of flexibility. We can write custom business rules within the workflows themselves. Each workflow can work differently; for example, medical insurance claims can be processed differently than travel insurance claims.

Greg Hauenstein: I appreciate you shedding color on those concepts. We are seeing a tremendous amount of interest in how to turn human-readable information into machine-actionable data, and then how we partner with our AI models to deliver outcomes. Thank you everybody for joining us today.