The use of artificial intelligence in the revenue cycle management space is accelerating as companies look to leverage this technology to reduce rejections and make financial workflows more efficient.
Ensemble says it is taking a unique approach by partnering with enterprise AI company Cohere to build RCM-native large-scale language models designed specifically for healthcare financial workflows.
Ensemble is a 12-year-old company that manages end-to-end revenue cycle operations for more than 30 health systems nationwide.
Many AI products wrap their prompts around a generic LLM. Ensemble Chief Technology Officer Grant Veazey said that with Cohere, Ensemble saw an opportunity to collaborate with enterprise AI partners to build completely custom models shaped by RCM insights.
The RCM-native LLM model is shaped by Ensemble’s operational expertise and fine-tuned for real-world RCM tasks, powering AI agents that support workflows from patient intake to account resolution, company executives said.
“With our expertise, we found that the revenue cycle is a very procedural and conditional world. We were looking for a partner who could suggest how to get past context engineering. And we had already partnered with Cohere. They recommended that we take very deep contextual data on how to do the revenue cycle properly and actually use that to post-train the model, so we could do things like context engineering and RAG (search augmentation). “We’re going to have something that’s not prompt engineering, but a model that’s truly trained on the best RCM data set in the industry, and that allows us to do RCM inference at a much higher level than anyone else in the industry, and certainly a lot more than the underlying model,” Veazey said upon getting a first look at the LLM model the two companies are building. told Fierce Healthcare.
According to Ensemble and Cohere executives, most AI tools attempt to “teach” models RCM logic at inference time through heavy context engineering, but this increases the cost of agent workflows, burdens model inferences with long context inputs, and plateaus accuracy. These AI tools may be inadequate to handle payer-specific behaviors, regulatory nuances, workflow dependencies, and complex multi-step processes that underpin healthcare financial operations.
The companies are developing an RCM-native intelligence layer that can understand complex clinical, financial, and regulatory language. The system is designed to adhere to multi-level rules and documentation requirements set by payers, and company executives claim it will drive productivity gains beyond what can be achieved with standard off-the-shelf large-scale language models.
The goal is to capture more revenue, reduce friction in the RCM process, and improve outcomes for healthcare providers and patients, Veazey said. “For every dollar saved and recovered in the revenue cycle, hospitals can reinvest another dollar back into their communities, improving existing facilities, opening new clinics, and providing better health care,” he said.
Executives said the model is not intended to replace or duplicate the workflow of electronic medical records systems. The RCM native model is designed to improve both accuracy and speed for users without changing EHR configurations.
“We’re in the business of generating insights and driving results. Being able to take that inference and bring it together with data and intelligence is really important to us. That’s what differentiates us, our ability to take those and deliver results to our customers,” he said.
Healthcare systems are increasingly adopting AI solutions for revenue cycle management, recognizing the potential of technology to improve coding and capture more revenue. According to a study by the Healthcare Financial Management Association (HFMA) and AKASA, 80% of health systems say they are considering, piloting, or implementing generational AI tools for RCM in 2025, a 38% increase in less than two years.
The work to build the custom LLM follows a two-year data partnership between the two companies. “Cohere has extensive experience working with enterprise companies where data and security are very important, and that was very important to us as we are in the healthcare industry and take security as a first-class citizen. They were also a cultural match for us: innovators, fast-moving, and industry leaders,” Veazey said.
This project does not use any identifiable client data or PHI to train the model. Rather, it leverages Ensemble’s insights from working with diverse health systems, including operator expertise, documented procedures, industry-wide patterns, payer trends, and denial behaviors, and is supported by synthetic datasets created from appropriately authenticated and anonymized sources within a HIPAA-compliant environment, executives said.
Cohere provides secure, enterprise-grade AI capabilities, said Joelle Pineau, the company’s chief AI officer.
“We are focused on secure deployment of AI. We have on-premises deployments where confidentiality and security are guaranteed. We are focused on sovereign AI, meaning the full stack, meaning the ability to control the model and agent platform. And we are 100% focused on enterprise deployments. We have developed expertise in how to partner, how to build AI models and agents in a way that meets enterprise needs,” Pinault said.
As part of the RCM Native LLM development effort, Ensemble and Cohere developed a revenue cycle benchmark dataset to measure model performance.
Ensemble plans to release the LLM model in the second half of 2026. “We hope this model will accelerate many of the existing AI projects that are already in production today,” he said.
Both companies’ AI initiatives for the revenue cycle come as health systems and providers face increasing financial pressures.
“Most providers are seeing an increase in initial denials. Providers really need a strategic partner who is capable of making this kind of investment and partnership with Cohere to have a fair chance in the new world of AI, because the payers that are well-funded are definitely spending millions of dollars in this space as well,” Veazey said.
Veazey said the two companies are focused on taking a real-world implementation-first approach to AI in RCM. He added that the company has learned valuable lessons from its AI investments over the past two years.
“Data is king in AI. Having the right data, having lots of data, and making sure that data is properly labeled and available to AI has been a huge learning experience for us over the last 24 months. We also need to understand where we sit on that line between being the consumer of AI and partnering with the creators of AI to be part of the creation of AI.” “A lot of people are easily impressed that they can pull out any type of underlying model in a chat situation and ask a question and get a reasonable answer. And that’s a good thing, but when you think about RCM and certain health data and protecting health data, being pretty good isn’t enough. What we’ve learned in clinical and administrative reasoning is that you need to take it to the next level and actually create your own models that are trained on the data.”
