Tennr

Where Tennr fits

Healthcare's front office runs on documents nobody chose: faxes, scanned referral forms, portal uploads, insurance cards photographed on phones. Tennr started there in 2021 — Axios framed its origin as fixing the fax — and has grown into what it now calls an agentic patient orchestration platform: it ingests patient orders and referrals from whatever channel they arrive on, extracts what matters, checks it against payer criteria, and pushes the patient to the right care setting without the weeks of manual back-and-forth that normally happen first. The problem is real and measurable: according to the company's own framing, more than half of the roughly one-third of Americans referred for specialty care each year never complete the next step.

What I look for in healthcare automation is narrow but strict: does it understand messy clinical documents, does it make fewer mistakes than the intake staff it supplements, and can it survive an audit. Tennr is unusually explicit on all three. Its pitch is not "an LLM for healthcare" but a set of specialized vision-language models — RaeLM — trained on medical documentation and evaluated against payer criteria, wired into an orchestration engine that does operational work: eligibility checks, benefits investigation, prior authorization submission, chart updates, routing decisions. That distinction from generic AI assistants is the core of the product, and it is the right battleground.

What it actually does

The intake side classifies and routes patient data from fax, portal and internal orders into the right operational workflows, including handwritten notes and scanned forms. Missing documentation — the company says the number one reason patients stall — is identified per patient and per service, requested from the right party, and tracked to completion. Coverage is handled continuously: payer and plan selection, change detection, automatic reverification before scheduling. The prior authorization module assembles proof against medical necessity criteria, submits, and tracks across benefit types; the same criteria drive routing recommendations, so a patient whose authorization status is unresolved gets a different path than one who is clear.

The communication layer closes loops that used to be phone tag. Tennr runs inbound and outbound calls — status questions, missing information, intake confirmation, co-sign requests to providers — in the patient's language at all hours, and gives referring providers real-time visibility into script status. That visibility productized in June 2025 as Tennr Network, a coordination layer where referring providers see every patient they sent, receiving providers see which referrals need documentation and which sources convert, and patients see acceptance, scheduling and expected cost. Finally there is an operational view of bottlenecks: which patients are stuck, what can move today, what affects value-based care.

The scale numbers reported in press coverage are substantial: roughly 10 million documents processed per month, millions of patients across hundreds of provider organizations, and a company in the eight-figure revenue range that says it tripled revenue in the two quarters before June 2025. The RaeLM model family is claimed to be trained on over 100 million anonymized healthcare documents, 2.3 billion distinct data fields and 8,000 criteria sets. Those are vendor figures, repeated consistently across Fortune, FierceHealthcare and MobiHealthNews but never independently audited.

My testing notes

First observation: Tennr is demo-gated end to end, and that shapes the evaluation. When I worked through the public surface, every path — product, network, pricing — funnels to "Book a demo." There is no self-serve trial, no public pricing, no developer documentation, no API reference. For a hospital evaluating a vendor this is normal; for a reviewer trying to verify claims it means every number on the marketing site is on trust. What I could verify is the consistency of those claims across independent coverage: the 10 million documents per month, the training corpus figures, the eight-figure revenue and the tripled-since-Series-B line all held up across Fortune, FierceHealthcare and the Series C press release without drifting.

Second observation: the eligibility angle is the sharpest thing in the pitch. The company cites an industry reality — eligibility checks are wrong almost 8% of the time — and positions RaeLM's purpose-built extraction as the fix, noting that it asks the follow-up questions needed to confirm information rather than guessing. The design implication is what matters operationally: a system that knows what it does not know and asks is fundamentally safer in medical administration than one that produces confident answers. I could not measure Tennr's own error rate — no public accuracy benchmarks exist — but the architecture of "flag potential denials and assemble proof before submission" targets first-pass denials, which is where the money actually is for specialty practices.

Third observation: Tennr Network is the strategic move in the product. When I read through the June 2025 launch coverage alongside the site, the significance is that it turns a single-provider automation tool into a two-sided coordination layer: referrers get delivery-tracking-style visibility that the company openly compares to food delivery and e-commerce. Whether that network effect compounds depends on adoption density in each specialty market, and the company is early — but every party in a referral has an incentive to join, which is more than most healthcare networks can say.
The funding history is worth a paragraph of its own because in healthcare IT the cap table is a reliability signal. Tennr went from a Y Combinator pre-seed in 2023 to a $18 million Series A led by Andreessen Horowitz in March 2024, a $37 million Series B led by Lightspeed that October, and the $101 million Series C in June 2025 led by IVP with Google Ventures and ICONIQ joining. Frank Slootman, the former Snowflake CEO, invested personally. The company said at Series C that it had not yet drawn down its Series B cash. Backing does not fix a referral backlog, but it does mean the vendor will be there in three years — a real consideration when you are rebuilding intake workflows around it.

Limits

Tennr does not do clinical decision support; it is administrative. It does not replace an EHR — it updates charts and works alongside them — and there is no public list of supported EHR integrations, so compatibility is a demo conversation. Pricing is entirely custom and undisclosed, which makes ROI comparison against staffing or against rivals like Concord Technologies impossible from the outside; you buy on the vendor's case studies. Implementation in referral-heavy organizations is a change-management project: the tool supervenes on fax lines, portal inboxes and phone scripts that real staff use daily. There is no self-serve evaluation, no public API, no documentation for developers. And the accuracy claims — denials reduced, conversions increased — come with no published methodology; the "eight figures of revenue" and valuation figures come from the company's CEO to journalists. The system handles protected health information under HIPAA through de-identification practices, but compliance certifications are not itemized publicly the way enterprise buyers may expect.

The honest framing of what Tennr sells is time. Its own CEO put it plainly to Fortune: the goal is that the patient actually gets the service and understands what it is going to cost, so that they show up. That is a conversion and no-show argument dressed as automation, and it is the right argument for a specialty practice where a completed referral is revenue and an abandoned one is a sunk marketing cost.

Verdict

Tennr is the most credible answer I have seen to a genuinely broken process — referrals disappearing into fax machines — and its funding trajectory backs the traction story: $101 million Series C at a $605 million valuation in June 2025, led by IVP with GV, ICONIQ, a16z and Lightspeed, on top of $37 million and $18 million rounds the year before. The specialized-model bet is the right one for medical documentation, and the Network layer could become the real asset over time. I rate it 3.9: deep core functionality aimed squarely at referral operations and denial prevention, strong momentum and reliability signals for a company its age, offset by undisclosed pricing, demo-only evaluation and the absence of any independent accuracy benchmark a buyer can check.

Sources

• https://www.tennr.com/
• https://fortune.com/2025/06/18/tennr-health-tech-ai-patient-referral-ivp-a16z-lightspeed-iconiq-series-c/
• https://www.prnewswire.com/news-releases/healthcare-referrals-are-where-patients-get-lost-tennr-raises-101m-to-bring-the-visibility-our-system-desperately-needs-302485255.html
• https://www.mobihealthnews.com/news/tennr-raises-101m-automate-referrals-hits-605m-valuation
• https://www.fiercehealthcare.com/health-tech/tennr-clinches-101m-build-out-ai-automates-patient-referral-workflows
• https://siliconangle.com/2025/06/18/ai-startup-tennr-raises-101m-revolutionize-patient-referrals/
• https://www.axios.com/pro/all-deals/2025/06/18/pro-rata-premium-first-look-mars-eu-kellanova-tennr-watco
• https://www.digitalhealthnews.com/google-ventures-backs-101-mn-round-for-healthtech-startup-fixing-patient-referrals
• https://multiples.vc/private-comps/tennr
• https://www.tennr.com/product
• https://www.tennr.com/lp/document-processing