Paid is not the same as paid in full.

An underpayment does not bounce.

The remittance posts, the account closes, and the difference between what the contract entitled you to and what actually arrived sits in nobody's worklist. A denial announces itself. An underpayment looks exactly like a payment.

Recovering it starts with knowing what you should have been paid — claim by claim, against the contract terms actually in force — and then pursuing the difference with the payer.

How the money goes missing

Contracts nobody has modeled. Payer contracts carry rate schedules, carve-outs, stop-loss provisions, implant and outlier terms, and escalators, and most of them live as documents rather than as calculations. If the expected reimbursement for a claim was never computed, there is nothing to compare the payment against — so a short payment posts exactly like a correct one.

Changes that never reach the system. An amendment, a retroactive rate update, a network change. The contract moves and the billing system's expectation does not, so every claim after that date can be paid wrong in the same direction, quietly, until somebody looks.

Variance too small to chase one claim at a time. Many underpayments are modest per claim and material in aggregate. Worked by hand, each one costs more to pursue than it returns, so none of them is pursued — and the pattern that would justify pursuing all of them is never seen.

Third-party liability. When a patient is injured in a motor-vehicle accident, another insurer is frequently primarily liable — but that fact is rarely established at the point of care. The claim bills to health insurance, pays at a health-insurance rate, and closes. The liable insurer is never approached, and standard coordination-of-benefits processes don't find it: they check the fields that exist, and the relevant fact was never captured in one.

Why the obvious fixes don't work

"Our contract management module flags underpayments." It flags variance against the rates configured in it. If the contract was loaded incompletely, or its complex terms were simplified during setup, it validates the error rather than catching it.

A one-time audit. It recovers a look-back window and stops. The contract terms that caused the gap are still in force, and the gap reopens the following month.

"We'd see it in our AR." You wouldn't. The account is closed, in balance, and correct-looking. That is the whole problem.

"Our subrogation vendor handles TPL." Subrogation works from data your registration process captured. If the auto-liability fact was never captured, there is nothing to work from. This needs active discovery, not passive matching.

What actually moves the needle

Compute what you should have been paid, then compare every payment to it. Contract terms turned into an expected amount for each claim, tested against the actual remittance across the full population rather than a sample. Variance becomes a list — by payer, by contract term, by service line — instead of a suspicion.

Treat detection and recovery as two disciplines. Finding the variance is a data problem. Recovering it is payer follow-up: disputes prepared to each payer's requirements, citing the contract term and the computed difference, and tracked to resolution. Different skills, worked as two steps rather than one.

Bring in contract specialists where the contract is the question. Some variances are errors. Others are disagreements about what the contract actually says. The second kind needs someone who reads payer contracts for a living — and the answer often belongs in the next negotiation as well as in this appeal.

Fix the cause as well as the claim. A recurring variance is usually one contract term loaded wrong or one payer policy applied wrong. Correcting it at the source stops the next run of claims from needing recovery at all.

Where to start

RevDNA models expected against actual reimbursement for your largest payer contracts, sizes the variance by payer and contract term, and runs a third-party liability discovery sample against historical motor-vehicle encounters. The output is how much of what you were paid differs from what you were owed — and which of those differences recur.