Coronis Health: 100,000 Claims a Week With Agents

March 2025: signing with UiPath, and the choice to build on top

In March 2025 Coronis Health announced its platform agreement with UiPath, starting from a twin strategy: traditional RPA on one side, intelligent reading of clinical documents on the other. Eighteen months later, the setup has grown in a different direction.

The company has built proprietary layers on top of the platform, welding together RPA, document processing, agentic automation and orchestration. CTO Doug Marcey calls the result "accountable intelligence", intelligence someone answers for.

Here lies the difference that matters. Buying off-the-shelf agents remains a purchasing decision; building your own layers on top of a platform is an architecture decision, and the two roads lead far apart.

Marcey starts from a lean principle: people should do the work that requires judgment, machines everything else.

US healthcare as a testing ground

Healthcare revenue cycle management is one of the environments most hostile to automation. Thin margins, fragmented systems, dense rules: every technical choice is paid for at the level of the individual transaction.

Coronis Health takes clinical records from providers, turns them into claims, sends them to payers and then handles denials, appeals and collections.

The revenue model is contingency-based: the company earns a share of what it recovers. That changes the math of every tool adopted, because cost per transaction flows straight into the income statement. A tool that works eighty percent of the time leaves a visible hole in the books.

In a context like this, days in accounts receivable on a claim are cash. Every manual step lengthens the cycle and pushes collection further out in time.

The last mile: putting data back inside the client's systems

The technical friction point is precise and far from glamorous. Reading data out of clients' systems is often possible; writing it back into those same systems the right way is much less so.

Marcey explains it clearly in the interview published by diginomica^[1]: healthcare technology is still catching up to the 2010s when it comes to interoperability, and often there is simply no route to write account or claim information back the way you would need.

Hence the choice of ground: bots and agents hold the last mile. They take the processed data and put it back into the source system, whatever that system is.

The benefit is twofold. Teams work on a single, consistent panel, and the client finds their own systems updated by end of day. Marcey names exactly this capability as the biggest operational result achieved so far on the platform.

The volume: 700 clients, 154 systems, 100,000 cases a week

The scope numbers explain why the last mile carries so much weight. The company works across a surface of systems few sectors can match for variety.

• roughly 700 clients served
• 154 electronic medical record (EMR) systems
• more than 25 practice management systems
• around 100,000 cases a week
• around 60,000 documents a week in the anesthesia division alone

One hundred and fifty-four EMRs mean one hundred and fifty-four ways of naming the same things. The anesthesia division is the extreme case: the most valuable and most idiosyncratic service line, with clinical records, explanations of benefits and lockbox flows arriving every day.

The figures come from the interview with Doug Marcey conducted at UiPath Fusion 2026 in Las Vegas and published by diginomica^[1]. They should be read for what they are: company statements, precise on scope and lacking any time-over-time comparison.

The course correction along the way

Every useful story contains a correction, and here it is visible to the naked eye.

The starting plan from March 2025 ran two tracks in parallel: traditional RPA for repetitive processes, intelligent reading to extract data from records. Eighteen months later that design has become something else. The proprietary layers built in-house hold together pieces the platform offers separately.

Calling it a failure would be wrong. It is a correction of scope: the company realized the value sat in the stitching, more than in the individual bricks.

The willingness to revisit the initial design, after eighteen months in operation, is a sign of operational maturity. Anyone presenting a plan unchanged since day one is usually reading from the brochure.

Fusion 2026: the ledger of human corrections

On the day of the interview, UiPath made Cartographer, its process mapping product, generally available, along with Process Atlas and Coding Agents. The same announcement introduced the Map of Work concept and the decision ledger.

The decision ledger is the interesting part for anyone who builds. It continuously records the corrections people make on top of machine decisions and feeds them back into the process definition.

It is the feedback loop every automation team, sooner or later, ends up drawing by hand.

Several of these announcements touch problems Coronis Health was already tackling on its own. When a customer builds in-house the feature the vendor later puts in the catalog, the direction taken was the right one. It is the clearest validation an enterprise tool can receive.

What it would take to call this ROI

Here comes the uncomfortable part. The public figures describe scope (clients, systems, documents, cases), while the operational result rests on the CTO's judgment.

What is missing is the data point that closes the loop: a before/after on a metric with a denominator. How many days it took to close a claim before the automation layers, how many it takes now. How far the denial rate has fallen, at a comparable client mix.

This distinction must be held firm. A result declared by the company and a result verified by an independent source carry different weight, and the speaker's credibility does not close that gap.

Anyone reading with an investor's eye should ask for exactly that: the denominator. Anyone reading with an engineer's eye can meanwhile take the design, which stands regardless of the missing number.

What you can take away

The transferable lesson is compact and cheap to test in your own context.

In fragmented environments, the value of automation concentrates at the point where data goes back into someone else's systems. It is the dullest stretch of the chain, and the hardest to buy ready-made. Whoever holds it with their own layers stops depending on a vendor's roadmap.

For an SMB founder the message is encouraging: the last mile is within reach, because a few targeted automations are enough, chosen on the processes the team knows by heart.

For anyone running a product the question is different: which piece of your chain requires real judgment, and which is keeping people busy with robotic work? Coronis Health answered by separating the two. Does your organization already know where that line runs?