How AI can save time, frustration in supporting the claims process
Removing touchpoints and as much manual intervention as possible can bring multiple benefits to providers and payers alike.
Among the many areas in which artificial intelligence could make a major difference in the claims process, ensuring all parties are up to date on a claim’s status is near the top of the list.
Managing the claims resolution process is difficult enough, and many believe that AI and machine learning can provide a positive return on investment, especially in the convoluted process of revenue cycle management that lies after claims submission – that’s where one out of every three claims need some type of review and additional work before they can be resolved.
Given the extensive inefficiencies in RCM processes, there’s much that advanced technologies can do. For example, a previous article examined payer masters and assigning payer IDs to payer names and addresses as one aspect of RCM that AI and ML can address.
The heightened attention being paid to artificial intelligence and machine learning re-emphasizes the importance of overseeing the use of these advanced technologies when they are applied to the revenue cycle management process, so they can cost-effectively improve business processes.
Automating electronic claims status
There’s renewed interest in applying technology to electronic claims status because of the introduction of RESTful API responses from payers, which add additional data elements to the 277 claims status responses.
In these cases, a RESTful API response tells the client what happened with a claim status response, provides more information about the response and delivers requested data in a consumable format, obviating the need for batch files. Thus, APIs function like a server talking with a payer server without any files being uploaded or downloaded. This accelerates the process to a matter of seconds.
The additional data elements in RESTful API responses enable touchless transactions and thus reduce labor related to claims issues. This precludes the use of ineffective processes such as scraping payer web portals.
The data elements that can create touchless transactions include original claim receipt data, allowed amount, non-covered charges, deductible, co-payment, co-insurance, rich denial description and check-cashed date. These elements aren’t included in the rigid HIPAA 277 transaction, and the extra information they provide can enable business offices to direct labor toward value-added activities like reviewing complex denials and enabling the use of analytics to better understand reasons for payer denials.
Touchless transactions capture enough adjudication data to enable the application of business rules and eliminate labor related to reviewing and following up on claims. Some examples of touchless transactions include small balance write-offs when the paid amount is near the contracted amount that’s expected for payment or pending responses that have no further current action but warrant a trigger or second claim status request in a few days.
Removing touchpoints is a significant driver of improved productivity. Touchpoints like calling payers, logging into payer web portals and manually entering adjudication data into workflow or EHR systems create a drag on productivity. Touchless responses integrated into workflow or EHR systems present opportunities for efficiencies and cost savings.
A best practice for dealing with payer claim status responses is to categorize and interpret category code D and E responses, which are hard to decipher. Responses from payers can be difficult to decipher, and the time to review payer companion guides is significant, but automated mechanisms are now in place to decipher ambiguous responses.
Business rules are considered a basic form of artificial intelligence, and business rules use predefined human-coded logic like “if-then” statements to make decisions and automate processes, rather than learning from data like machine learning models. While they are a type of AI, they are not the same as the more advanced data-driven AI systems in use today. However, machine learning can be used to automatically generate appeal letters based on pre-defined rules and the reasons for denials – and this can save time.
Eliminating hold times on payer calls
Another potential area for AI benefits involves facilitating phone calls to payers and navigating payer phone trees.
There are several reasons why it’s still crucial to make phone calls to payers. For example, revenue cycle management firms were not issued credentials to payer web portals in a timely manner by their provider client and do not use the EDI 277 response with additional claim adjudication data made available via RESTful API responses.
Also, claim status error responses are vague and have not been labeled appropriately based on payer companion guides. Finally, claim status 277 responses are used without the additional claim adjudication data available with RESTful API responses. This can occur if NPIs and Tax IDs are not appropriately registered with payers.
Beyond eliminating the need to navigate payer phone trees, a more advanced approach is embedding error handling into the automated process to avoid convoluted payer phone systems that frequently result in disconnected calls.
Intelligent tools like a scheduler create a touchless event with software applications to improve the return on investment on innovation. Imagine a scenario where a scheduler tool places callbacks from payers to denial representatives throughout the day with a payer representative on the phone waiting to talk with no time wasted on hold. All the denial representatives need to do is answer the phone to begin the claim resolution process with the payer rep.
AI as a solution
Large language models (LLM) are a specific type of machine learning that analyze and generate human language by training on massive amounts of data.
For example, LLMs can navigate payer phone trees by understanding natural language, identifying user intent and using a structured reasoning process to choose the correct path. LLMs can process spoken or typed requests, which is important because payer phone trees offer touch tone and speech recognition.
Clearly, the use of AI and machine learning can transform revenue cycle management operations and enable denial representatives to focus on complex activity and avoid the rote, routine and repetitive nature of calling payers, responding to phone tree prompts and waiting on hold to speak to a payer representative.
While the return on investment for these activities is significant, they also provide the additional benefit of removing exhausting, needless activities that include waiting on hold on multiple calls just to talk to payer representatives. Providers’ denial response teams will be able to better focus their attention and provide the biggest returns on their time and expertise.
Ken Poray is CEO of Integrex Health with 20 years of experience in working with claims status, including EDI and web portal transactions.