How Do You Catch Underpayments Against Contracted Rates?

“Paid” is not the same as “paid right.”

Billing teams spend most of their energy on denials and unbilled work. Underpayments sit in a quieter pile. The claim went out clean. The payer adjudicated. Money posted. The balance looks closed. Nobody opens a work item because nothing bounced. The contract still says the allowed amount should have been higher, and the difference just evaporated into contractual write-off habits that nobody rechecks.

Practice owners ask about this when net collection looks soft even though clean claim rate and denial rate look fine. That gap is the subject of this article.

Underpayment is not a denial and not a secondary

Three different leaks get mixed together in the same AR meeting.

A denial is an explicit no. You get a CARC, a reason, and a clock. Denial work is classification, fix or appeal, and deadline control.

A secondary claim is another payer still owing part of the same encounter after primary adjudication. Secondary work is COB order, primary payment data, attachments, and a second 837.

An underpayment is the primary (or the only) payer paying less than the contracted allowed amount, or applying the wrong fee schedule, multiple procedure reduction, or bundling logic for that plan and that date of service. The remittance may look normal. The write-off may be taken automatically. The leak is mathematical, not theatrical.

If you staff one queue for all three, the underpayment work loses every time. Denials yell. Secondaries age. Underpayments smile and disappear.

Why this leak is getting louder in 2025 and 2026

Industry estimates commonly put underpayment drag around 1 to 3 percent of net patient revenue each year when contract management is average. Vendor write-ups sometimes push 5 to 7 percent for shops with weak fee schedule upkeep. Treat those as directional ranges, not a quote for your practice. The point is that a quiet percentage of paid claims can vanish without ever appearing on a denial report.

The pressure is real. A February 2026 survey of revenue cycle leaders, summarized by Revecore and Fierce Healthcare reporting, found 62 percent citing denials and underpayment management as their top obstacle heading into 2026. Separate hospital analyses in 2025 showed net revenue leakage rising even while AR days improved. That is the underpayment pattern: cash moves, collection metrics look calmer, and contracted value still leaks out the side door.

Large settlements keep proving the problem is not imaginary. A federal court gave final approval in August 2025 to a Blue Cross Blue Shield plan to pay about $2.8 billion related to underpayments to hospitals and other providers. An arbitration panel earlier ordered UnitedHealthcare to pay Envision Healthcare $91.2 million over a reimbursement agreement dispute. Those are hospital-scale fights. The same mechanics show up in a small clinic when a commercial plan never loads last year’s rate update.

Where the money actually goes

Most shops do not lose underpaid cash because nobody has a fee schedule. They lose it in three boring places.

Auto-posting trusts the payer allowed amount and writes the rest to contractual adjustment without a second look. That is correct when the payer is right. It is a silent giveaway when the payer is wrong.

Fee schedules drift. A plan renews, a CMS update lands, or a local contractor edits a rate, and the practice management comparison table is still last year’s file. The CY 2026 Medicare Physician Fee Schedule is a live example: dual conversion factors and efficiency adjustments that move work RVUs on many non-time-based services. If commercial contracts peg to Medicare or refresh on a January clock, a stale table will bless the wrong payment as contractual.

Small misses never clear a manual review threshold. Five dollars on a code you bill two thousand times a month is real money. A human will not open two thousand EOBs for five dollars. Software will.

Secondary and patient-balance work can hide the same miss. A short primary allowed amount gets written off, then the remainder is pushed to patient or secondary without anyone asking whether the primary was correct.

What automation can take off the plate

Useful underpayment automation starts from the remittance, not from a quarterly consultant export.

It compares paid and allowed amounts to the contracted fee schedule for that payer, plan, place of service, and date of service, claim line by claim line. The comparison has to respect modifiers, units, bilateral rules, and site of service, or the flags will be noise.

It knows which reductions are expected. Multiple procedure rules, bilateral modifiers, assistant surgery, capped visits, and timely filing penalties are not all “payer mistakes.” A dumb variance engine that flags every reduction will bury the team. A useful engine ranks unexpected variance above expected reduction.

It opens work items only when the dollar gap clears a threshold you set, or when the same CPT and payer pattern repeats enough times to matter in aggregate. One $12 miss is noise. The same $12 miss on the same CPT from the same plan three hundred times is a project.

It drafts the recovery packet with the contract clause or fee schedule row, the claim identifiers, and the remittance lines, so a biller is editing and sending, not rebuilding from scratch. AI helps match remittance lines to contract language and assemble a clean packet. It should not invent contract terms in freestyle letters.

It tracks payer response and closes the loop into payment posting so recovered dollars do not land as mystery patient credit.

It can also watch post-contract windows. New rates and amendments often fail in the first 90 to 120 days after a renewal because the payer loaded the wrong exhibit. That is a monitoring job, not a heroic audit once a year.

Vendor case studies often quote recovery rates in the 85 to 95 percent range once a variance is validated and appealed. Treat those as vendor-reported until they survive your top payers. Detection at scale is the hard part. Appeal quality is the second. Both beat quarterly sampling.

What still needs a human

Contract interpretation is not a pure rules problem. Evergreen amendments, silent PPO wraps, rental networks, and “lesser of” language need a person who can read the deal. Automation can surface the line that looks wrong. A human decides whether the contract supports a fight.

Clinical context still matters on some lines. A modifier that looks like a payer error may be a documentation problem on your side. Blind appeals create payer fatigue and hurt the next real case.

Relationship capital is finite. You do not want a bot firing fifty low-value disputes a week at your top commercial plan. A human sets the batch rules and the stop list.

Legal and compliance boundaries matter. Recovery work must stay inside your BAA, your contract dispute terms, and your documentation standards. No PHI in loose spreadsheets. No copy-paste of full EOBs into consumer AI tools. Commercial reconsideration windows are often short. A perfect variance found six months late is a report, not cash.

How to tell if this is your leak

Before you buy another dashboard, answer five questions with your own data.

What is net collection versus expected allowed on your top twenty CPT and payer pairs for the last ninety days.

How often does auto-post write contractual adjustment with no secondary review on those pairs.

When was each major payer fee schedule file last refreshed in the system that posts payments.

How many underpayment appeals did you send last quarter, and what was recoveries versus labor.

What share of AR over ninety days is pure “paid, closed, never variance-checked” versus open denials and unbilled secondaries.

If the top pairs show a steady negative variance and nobody owns a recovery queue, you do not have a motivation problem. You have a comparison and threshold problem.

Also separate underpayment from takebacks. A payer clawing money back after payment is a different workflow with different appeal logic. Mixing takeback defense into underpayment recovery muddies both queues.

A practical sequence that does not boil the ocean

First, pick the top payers and CPT pairs by volume times dollar, not by anecdote.

Second, refresh fee schedules and plan mapping for those pairs only. If a commercial rate is pegged to Medicare, load the current conversion logic instead of last year’s static table.

Third, run a retrospective variance on ninety days of remittances and rank unexpected gaps. Keep expected reductions out of the worklist.

Fourth, stand up a small recovery worklist with dollar and pattern thresholds, not a firehose. Assign an owner. Unowned variance reports die.

Fifth, template the appeal or reconsideration packet for the two or three payers that dominate the miss. Include the contract cite, the fee schedule row, claim IDs, and the ERA lines.

Sixth, only then expand to broader AR automation, denial appeal drafting, or secondary COB work if those queues are still the bigger leak. Underpayment recovery on a dirty denial shop just invents busywork.

That is the kind of leak we map in a free 30-minute Waste Audit. We look at the path on top of the software you already use, put a number on preventable underpayment drag, and only talk about automation where the math holds. No savings, no fee. Start at /free-audit, see the workflows at /solutions, or read how an engagement runs at /how-it-works.