Every AR department runs on the same arithmetic, whether or not anyone writes it down. A claim is worth working if the expected recovery beats the cost of the touch. Premier Inc. puts the average cost to rework a denied commercial claim at $63.76, which means a large share of the queue fails that test the day it arrives, and everyone in the department knows which part.
That arithmetic is what AI agents change. Not the payer rules, not the contracts, not the clearinghouse. The per-touch cost. When a status check stops consuming a person's attention, the line between worth working and write off moves, and it moves in one direction. Here is what actually changes on the floor when it does.
Making the bottom of the queue economic again
Work queues have always been sorted by dollar value, because attention is the scarce resource. High-balance commercial claims get worked. Small balances age. Timely filing does the rest.
Premier Inc. puts the all-payer average cost of reworking a denied claim at $43.84, and the commercial average at $63.76. Set either number against a small balance and the sort order explains itself. The Change Healthcare Revenue Cycle Denials Index, also cited by the American Medical Association, found that 65% of denied claims are never reworked at all.
AI revenue cycle management does not make those claims more valuable. It makes them cheaper to touch, which has the same effect on where they land in the queue.
- Small-balance claims re-enter the queue instead of aging into timely filing
- Prioritization stops being a proxy for staff capacity and starts tracking recovery probability
- Secondary and tertiary balances get followed rather than closed at primary payment
- The write-off threshold becomes a policy decision instead of a staffing constraint
Queue economics in action
At Nutex Health, a 24-campus US health system, fully loaded cost per claim fell from $22.28 to $2.00. At $2.00 a touch, claims that used to sit below the line are worth a call, and then a second call. Nobody reset the threshold. The cost underneath it fell out.
Turning claim status follow-up into background work
Status follow-up is the largest single consumer of AR staff time and the hardest to defend. The CAQH Index estimates roughly 24 minutes of manual effort per claim status follow-up. Almost none of that is judgment. It is hold music, phone trees, portal logins, and re-keying the same claim number into a fourth system.
Claim status automation takes that work off the clock. Agents query the payer, read the response, update the work queue, and surface only what needs a person: a denial with a real reason code, a claim the payer has no record of, a payment that lands under contracted rate.
The change is not that status checks get faster. It is that they stop being rationed. A claim can be checked on day 7, day 14, and day 21 without anyone deciding in advance that it deserves three touches.
- Follow-up cadence is set by payer behavior, not by how many claims a specialist can reach in a shift
- Staff open a queue of exceptions rather than a queue of unknowns
- Portal and phone work runs overnight and between business hours
- The first human touch happens after the facts are already gathered
Claim status automation in action
Across the LunaBill portfolio, agents handled 90,000+ payer interactions per month against $64.4M of claim value in the trailing 30 days. At Nutex Health, staff time per claim follow-up fell from 22 minutes to 4 minutes. The 4 minutes that remain are the part that needed a person.
Shifting denial work from cleanup to prevention
Denials get worked after the fact because that is when they become visible. By then the cost is fixed, the appeal clock has started, and the claim competes with everything else in the queue for the same limited attention.
Premier Inc.'s figures show how uneven that cost is by payer. Reworking a denied Medicare claim averages $0.79. Medicare Advantage averages $47.77. Commercial averages $63.76. Same department, same staff, very different economics depending on who is on the other end of the claim.
Denial management automation changes the sequence. When agents are already reading payer responses at volume, patterns show up before the denial does: the payer that started requiring a modifier, the plan that changed an authorization rule mid-quarter, the coding pattern that draws a specific reason code from one payer and not from three others.
- Denial reasons aggregate across the whole book of business instead of being handled claim by claim
- Front-end fixes arrive with evidence attached, so registration and coding changes have a number behind them
- Appeals get filed while the documentation is still easy to assemble
- The share that never gets reworked shrinks, because rework cost is no longer the binding constraint
Denial prevention in action
At Nutex Health, 60+ day commercial claims moved from 45 days in AR to 7 days within three months of go-live. That 45-day window is where most denials used to be discovered. Closing it does not prevent every denial. It turns most of them into a problem with time left on the clock.
Keeping the judgment calls with people
Some of the queue should not move. The parts that require judgment are the parts that pay, and they stay with staff.
A clinical appeal that needs a physician's narrative is human work. So is escalating to a payer rep who has gone quiet, arguing a contract interpretation, and deciding whether a patient balance goes to collections or gets written off. Those decisions have consequences that reach past the claim.
What agents do is clear the path to them. They run inside Epic, Cerner, GoRev, Waystar, and Availity, so the work stays in the systems staff already use and the audit trail stays where compliance expects to find it. The engagement is HIPAA compliant and SOC 2 Type II, with a BAA executed in under a day.
- Specialists spend the day on appeals and escalations rather than on discovery
- Senior staff stop absorbing volume and start absorbing difficulty
- New hires ramp on exception handling instead of portal navigation
- Headcount planning comes loose from raw claim volume
Human judgment in action
Nutex Health worked $32M+ of AR in 30 days with zero added headcount. The team did not grow. The work it could reach did. Every hour that came back went to claims where a person's decision changed the outcome, which is set out in the Nutex Health case study.
Rebuilding the queue around what actually pays
Four changes, and they compound.
- Low-balance claims become economic to work, so the bottom of the queue stops aging out
- Status follow-up runs in the background, so staff open exceptions instead of unknowns
- Denial patterns surface early enough to fix upstream, not only to appeal downstream
- Human attention concentrates on the claims where judgment moves money
The real constraint on an AR department has never been effort. It has been arithmetic: how many touches fit in a shift, and which claims clear the bar those touches set. When the per-touch cost falls the way it did at Nutex Health, the bar falls with it, and a queue that was always theoretically workable becomes actually workable. That is worth planning for now, because the sorting logic sitting in most work queues today was written for a cost structure that is on its way out. How pricing works follows the same logic: 1.5% of net collections recovered, billed after the money lands.
Sources: Premier Inc. (cost to rework denied claims, commercial, all-payer, Medicare, and Medicare Advantage); Change Healthcare Revenue Cycle Denials Index, also cited by the American Medical Association (share of denied claims never reworked); CAQH Index (manual effort per claim status follow-up); LunaBill engagement data for Nutex Health and portfolio figures, trailing 30 days.