Healthcare document intelligence-focused startup Tennr Inc. today announced it has raised $101 million in a funding round to fix what’s wrong with patient referrals.

The Series C investment was led by IVP with participation from existing backers Andreessen Horowitz and Lightspeed, bringing the company’s valuation to $605 billion. The fundraise comes less than a year since the company’s last $37 million round in October.

Tennr uses proprietary artificial intelligence document analysis tools to help healthcare organizations sift through thousands of paper documents to help organize what is otherwise a complete confusion. The company has developed a suite of algorithms to identify the type of medical document received and accurately tag medical referrals, including those containing handwritten notes.

According to the company, referrals are the first line where patients find themselves in limbo because of missing documents, delayed workflows or denials. Any delay in care, or a denial of necessary care, and a patient could decide not to return to the doctor, leading to a worsening illness.

A referral is the process by which a primary care doctor recommends a patient see a specialist to receive a specific medical service, basically a formal request for an evaluation. The paperwork goes through multiple hands, including insurance, billing and medical coordination. Any portion of this pipeline can be muddied if the paperwork is unreadable or is left on a desk or forgotten in a filing cabinet.

To speed up this process, Tennr trained its large language models on more than 100 million medical documents, 2.3 billion distinct data fields and 8,000 sets of criteria. In the real world, eligibility checks can be wrong almost 8% of the time, the company said, so Tennr determines what questions need to be asked to confirm information. Its RaeLM model extracts data points that are purpose-built for administration to help reduce this error rate.

Although big-name LLM companies such as OpenAI and Anthropic PBC have risen onto the scene with generalized closed-source models, Tennr Chief Executive Trey Holterman told Fortune these models are not fit to the challenge. Instead, what’s required is making a specialized medical AI document model and algorithms that use hyper-specific healthcare data.

“Betting on open source, betting on a proprietary data set that we’ve accumulated, continues to totally smash benchmarks,” Holterman said.

Tennr also offers a system that can understand and parse medical call transcripts into actionable data. Called T3, or Transcript Translation Technology, it combines carefully designed heuristics trained on medical data and probabilistic models to understand the unique cadences of medical calls. This is because most medical phone calls involve someone reading the same things over and over, such as member IDs.

Holterman said the solution isn’t about automation but about providing a tool that will support patients and deliver a better method for resolving an immediate need. “It’s really about making sure that the patient actually gets the service and understands what it’s going to cost, so that they show up,” he said.

Photo: Pixabay