Most conversations about denials never get to the money. They get to the denial rate, which is a process metric and then stop.
Here is the frame I use with finance leaders. EBITDA is the output. It sits on the right side of the equal sign. On the left side there are four inputs: how much you pay to collect, how much you expect to collect, how much you actually collect, and how quickly you collect it.
A denial touches all four. That is why it is easy to talk about and hard to price.
The three buckets
First a caveat: all of this depends on the practice, and hospital billing and physician billing are genuinely different. These are the proportions we see, not a published study. Roughly:
Twenty to thirty percent goes to cost. You are spending too much on people and software to collect the money.
Fifteen to twenty percent is working capital. Every day you do not get paid costs you something, and you borrow to finance the gap.
The remaining half or so is bad debt and underpayments. Balances you write off because you never collected them, plus claims where you did get paid, but not enough for the care you delivered (eg downcoded claims)
Notice the shape of that. The bucket everyone manages is the smallest one, because it is the only one with a budget line attached to it. The biggest bucket has no line item at all. It shows up as revenue that never existed.
Bucket one: cost
This is the bucket with an owner. Somebody signs the invoices for the billing team, the outsourcer, and the software.
It is also the bucket where the industry's answer has been the same for thirty years. Friction appears, so you add people. That answer worked, in the sense that it produced collections. It just produced them at a high price.
Bucket two: working capital
This is the one finance teams see and operators often do not.
The other side of accounts receivable is net working capital. The longer you wait to get paid for a service you already delivered, the more working capital you have to carry to finance operations, and carrying it costs money.
A backlog is the clearest version of this. A growing backlog is not a productivity problem, it is money you are owed for care you already delivered and costs you already paid. You bought the provider's time, the room, the syringes. The offsetting revenue has not arrived. You are financing the difference, whether or not anyone calls it that.
Bucket three: bad debt and underpayments
Bad debt is what happens to denials when you write off a claim as no longer collectible. It is visible if you look for it, and most organizations know their number. What good looks like, in my experience: bad debt at about 2% of net revenue. We see 6, 7, 8 percent and higher. On any reasonably sized organization, the gap between 2 percent and 7 percent is millions of dollars.
Underpayments are harder, because they don’t look like a loss. Money arrived, the claim’s no longer in your denials queue, even though you’re not getting paid what you’re contractually owed.
Down-coding is the clearest example. You submit a claim with an E&M code that reflects a certain amount of clinical complexity, say a level 4. The payer says, we don’t agree, and we’re going to pay you for this service as though it were a level 3.If your reporting tracks denial rate and cash collected but not yield per encounter, this bucket is effectively invisible.
What actually moves each bucket
Cost to collect. This is where AI agents have the most direct effect, because most of the cost of working a denial is diagnostic labor: the Substrate Claim Status Agent checks the clearinghouse, the payer portal, the practice management system, the lockbox, and will even wait on the phone. Across the workflows we have touched, we have driven at least a 70 percent reduction in cost to collect in those specific workflows.
Speed. Instead of queues that build over days and get worked down over days, the moment something lands in a queue we touch, it gets resolved that day about 8 times out of 10. That pulls days in accounts receivable down, which pulls cash forward, which shrinks the working capital you carry.
Recovery. Of the bad debt that is genuinely recoverable, you can cut it by 75 to 80 percent. This increases net revenue.
Putting a number on it
Here is an illustrative model, and I mean illustrative rather than a promise. On a practice with $100 million in net patient revenue, the opportunity is in the neighborhood of 3.5 points of EBITDA. Roughly half of that is revenue recovery. Roughly half is lower cost to collect.
For Heading Health, an outpatient behavioral health practice, the recovery side looked like this: 88 percent of appeals submitted each month without a person assembling them, more than 40 percent of denial revenue recovered, and a 2 percent lift in net collections attributable to Appeals. Read the case study.
What this doesnt fix
Not all bad debt is recoverable, and I would rather say that plainly than let a model imply otherwise.
If a service needed a prior authorization and nobody obtained one, that claim is not getting paid. That is an operational problem upstream of billing.
Substrate covers roughly 60 to 70 percent of the denials workflow today. We don’t handle charge entry, pre-bill, or coding. When thousands of claims are being underpaid on a contract basis, somebody has to call your contracting representative at the payer and work it out. The contracting rep (if you can reach them) isn’t going to take a call from an agent. Same with peer to peer clinical reviews.
The point of automating the routine diagnostic work is because it enables your team to spend time on the denials that no machine can solve.
Watch the demo
Your team runs RCM. We give them time and leverage. See our agents work a real denial, on demand, no meeting required: watch the demo.
