BEFORE YOU BUY AI FOR LAB BILLING, ASK WHAT IT CAN SEE
Most labs already have automation. What they don't have is the context to make it useful.
That gap helps explain a number that should give any lab billing leader pause. In an HFMA and FinThrive survey, 63% of healthcare organizations said they use AI or automation in the revenue cycle. Only 15% reported a positive ROI.
In a new white paper, Bolt-On or Built-In: AI in Lab RCM, Sunjay Mouli, Vice President of AI Systems at TELCOR, looks at why so much of that investment stalls in lab billing, where AI is paying off today, and how to tell the difference before you sign a contract. Here's a preview of the argument.
LAB BILLING HAS A CONTEXT PROBLEM
The administrative drag in healthcare billing is well documented. The AHA estimates hospitals spent $43 billion in 2025 trying to collect payments insurers already owed for care they delivered. Availity's 2026 Abrasion Index found that around 70% of denials are eventually overturned and paid. Most denial work is churn, not recovery. Often the information was right. It just arrived too late.
Labs carry a harder version of this problem. A lab has to establish medical necessity from outside the provider-patient relationship. A hospital coder can go back to the chart. The lab usually can't. When the supporting diagnosis lives in the medical record but never reaches the requisition, the claim goes out incomplete. Molecular and genetic testing raise the cost of every gap, and the staff who might otherwise work around it are harder to hire every year.
That's why so much AI in lab RCM disappoints. The tools are aimed at tasks. The real problem is missing context.
BOLT-ON AI: THE RIGHT FIRST STEP
The paper separates AI deployments into two layers. The first is bolt-on: a tool layered over the systems a lab already runs. Mouli is clear that this layer is real and, for most labs, the right place to start.
Correspondence processing is the usual first candidate. Faxes, PDFs, and lockbox documents are read, classified, and indexed to the right patient and claim without anyone dragging files between systems. The time saved is welcome, but the paper argues the bigger payoff is one most labs overlook, and it shows up directly in appeal win rates.
Choosing that first workflow matters more than most teams expect. The paper lays out a four-part filter for picking one, including the criterion labs skip most often. It's also the one that decides whether a second workflow ever gets funded.
Bolt-on tools have a ceiling, though, and it's usually context. A bolt-on tool sees only the documents it's pointed at. It doesn't see the denial in the billing system, the records in the document store, or the contract terms. Beyond well-bounded tasks, it has to fill in the gaps, and that's where accuracy drops.
BUILT-IN AI: WHERE THE VALUE CONCENTRATES
The second layer is built-in AI, with native access to a lab's data model and workflow engine. That access changes two things at once: the context available for a decision and the actions available after it.
The paper breaks lab RCM “AI” into four distinct capabilities and explains why most standalone tools handle only the first two. The value concentrates in the other two, and those depend almost entirely on how much of a lab's data the system can reach. Building a payer- and patient-specific medical necessity argument for an appeal, then writing the result back into the system of record, is a question of integration, not model choice.
“Once you do have that full context, the ability to do something with the data becomes exponentially larger,” Mouli says. “The usage of a language model depends entirely on the source data you have.”
The paper also takes on one of the most common objections to AI in billing: that it weakens auditability. Mouli argues the opposite and uses lab ICD coding to show why.
WHY TECHNOLOGY ALONE WON'T MOVE THE NEEDLE
The gap between 63% adoption and 15% positive ROI isn't a model problem. The paper identifies three ways labs get AI wrong. The most common is automating a list of tasks with no view of the process around them. You can make a single step dramatically faster, but if the process is broken, it stays broken.
The labs getting value share two habits. They redesign the process around automation instead of layering automation onto what they already do. And they redesign their context, breaking down the walled gardens around patient data, policy data, and contracts. Most of that work isn't AI at all, and none of the rest pays off until it's done.
TRY THIS BEFORE YOUR NEXT VENDOR CALL
You don't need a vendor to find out which problem you have. Pull your last 20 overturned denials. For each one, ask what information the win depended on and where that information lived.
If most of it came from a document someone had to go find, you have a context problem, not a productivity problem. Faster task automation won't fix it.
When you do talk to vendors, the paper offers six questions to ask. Start with the one Mouli calls the most diagnostic in the category: How many workflows do you have in production, and with whom? A vendor who can't answer specifically is selling a roadmap.
WHAT'S IN THE FULL WHITE PAPER
The four capabilities behind nearly every “AI” product in lab RCM, and which ones drive value
A four-part filter for choosing a first workflow that earns funding for the next one
A crawl, walk, run maturity model with realistic timelines for each stage
Why AI-driven workflows can strengthen auditability rather than weaken it
All six vendor questions, and why each one matters
WHERE TELCOR FITS
There is no shortage of healthcare AI vendors. There is a shortage of vendors who understand labs. TELCOR has spent decades inside the details that decide whether a lab claim gets paid. For labs whose billing system is already TELCOR, the context problem is largely solved by architecture rather than integration work. For labs on other systems, integrations into other lab RCM platforms and adjacent data stores already exist and have run in production.
The paper's six questions are the right ones to put to every vendor in this market, including TELCOR.
