Artificial intelligence in the revenue cycle management space is heating up as companies look to leverage the technology to reduce denials and make financial workflows more efficient.
Ensemble says it is taking a distinct approach by teaming up with enterprise AI company Cohere to build an RCM-native large language model designed specifically for healthcare financial workflows.
Ensemble, a 12-year-old company, manages end-to-end revenue cycle operations for more than 30 health systems nationwide.
Many AI offerings wrap prompts around general-purpose LLMs. With Cohere, Ensemble saw an opportunity to work with an enterprise AI partner to build a fully custom model shaped by RCM insights, Ensemble Chief Technology Officer, Grant Veazey, said.
The RCM-native LLM model is shaped by Ensemble’s operational expertise and fine-tuned on real RCM tasks to power AI agents that support workflows from patient intake through account resolution, company executives said.
"We found in [the] revenue cycle, with our expertise, it is a deeply procedural and conditional world. We were looking for a partner to say, how do we go past context engineering? And we'd already been partnering with Cohere. They recommended to take our really deep contextual data around how to do revenue cycle right, and actually use it to post-train a model, and we could collectively go together and build RCMs first specifically-trained model, not context engineering, not RAG (retrieval augmented generation), not prompt engineering, a truly trained model on the best set of RCM data in the industry, and we would have something that would allow us to do RCM reasoning at a much higher level than anybody else in the industry, and certainly much more than a foundation model could do," Veazey told Fierce Healthcare in a first look at the LLM model that the companies are building.
Most AI tools attempt to "teach" models RCM logic at inference time through heavy context engineering, which raises the cost of agentic workflows, strains model reasoning with long-context inputs and hits a ceiling in accuracy, according to Ensemble and Cohere executives. These AI tools can fall short in handling payer-specific behavior, regulatory nuance, workflow dependencies and the complex, multi-step processes that underpin healthcare financial operations.
The two companies are developing an RCM-native intelligence layer that is capable of comprehending complex clinical, financial, and regulatory language. The system is designed to navigate the multi-step rules and documentation requirements set forth by payers, driving productivity improvements that surpass those achievable by standard off-the-shelf large language models, company executives assert.
The aim is to help capture more revenue, reduce friction in the RCM process and improve outcomes for providers and patients, Veazey noted. "For every dollar saved and recovered in the revenue cycle, that's another dollar hospitals are able to re-invest back into their communities: improving existing facilities, opening new clinics, and providing better care," he said.
The model is not meant to replace or replicate electronic health record system workflows, executives said. The RCM-native model is designed to improve both accuracy and speed for users without making any changes to EHR configurations.
"We're in the business of creating insights, creating outcomes, and it's really important for us to be able to take that reasoning and be able to ingest it with our data and that intelligence, that's what differentiates us—the ability to take those and deliver those outcomes for our customers," he said.
Health systems are eager to adopt AI solutions for revenue cycle management, seeing the promise of the technology to improve coding and capture more revenue. Eighty percent of health systems say they’re exploring, piloting or implementing gen AI tools for RCM in 2025, a 38% jump in less than two years, according to a survey by the Healthcare Financial Management Association (HFMA) and AKASA.
The work to build a custom LLM builds on a two-year data partnership between the two companies. "Cohere has extensive experience working with enterprise companies in which data and security is extremely important, and us being in the healthcare industry and taking security as a first-class citizen, that was extremely important to us. They also matched us from a cultural perspective—innovator, fast-moving, industry leader," Veazey said.
The project does not use identifiable client data or PHI for model training. Rather, it draws on Ensemble’s insights from working with diverse health systems, including operator expertise, documented procedures, industry‑wide patterns, payer trends and denial behaviors, supported by synthetic datasets created from properly certified, deidentified sources within a HIPAA-compliant environment, executives said.
Cohere provides secure, enterprise‑grade AI capabilities, noted Joelle Pineau, the company's chief AI officer.
"We focus on secure deployments of AI. We do on-premise deployment, where confidentiality and security is guaranteed. We focus on sovereign AI, meaning the ability to have control over the full stack, so the model, as well as the agentic platform. And, we are 100% focused on enterprise deployments. We've developed an expertise in terms of how to partner, how to build the AI models and the agents in a way that satisfies the needs of enterprise," Pineau said.
As part of the development work on the RCM-native LLMs, Ensemble and Cohere are to develop a benchmark dataset for the revenue cycle to measure model performance.
Ensemble plans to release the LLM model in the second half of 2026. "We expect it to fuel a lot of our existing AI projects that we have already in production today," he noted.
The two companies' work on AI for revenue cycle comes as health systems and providers are facing increasing financial pressures.
"Most providers are seeing an increase in initial denials. A provider really needs to have a strategic partner who has the ability to make these types of investments and partnerships that have with Cohere to have a fair shot in the new AI world because the well-funded payers are certainly, certainly spending millions in this area as well," Veazey said.
The two companies are focused on taking a real-world, implementation-first approach to AI in RCM, Veazey said. The company has learned valuable lessons in its AI investments in the past two years, he added.
"Data is king in AI, having the right data, having the multitude of data, having that data properly labeled and consumable to AI, was a huge learning for us over the last 24 months. You also have to figure out where you are on that line of a consumer of AI versus partnering with creators of AI and being a part of the creation of AI," he said. "A lot of people are very easily impressed with being able to pull up any type of foundation model in a chat situation and ask it a question and get a reasonable answer back. And while that's good, with RCM and specific healthcare data, and the protection of healthcare data, pretty good is not good enough. What we learned in our clinical and administrative reasoning is that we needed to take it to this next level, which is actually creating proprietary models that are trained on our data."