By Sagar Shankaran, Founder of CallSphere
A 40-claim DME probe takes 16 working days one at a time and finds the gaps too late. Split four ways, the records requests go out on day two of forty-five.
Key takeaways
You tried handing an audit response to software in 2024. It produced a checklist, a folder structure and a reminder email, and every single packet still got assembled by your reimbursement manager, one claim at a time, on the same 45-day clock. It was a filing cabinet with opinions. Here is what is different now, and it is not that the software got smarter about Medicare policy.
The letter arrives from CGS on a Tuesday. Targeted Probe and Educate, round one, forty claim numbers listed, forty-five days to respond. Every one of those forty needs a complete packet: the Standard Written Order with a valid signature and date, the face-to-face chart notes that establish medical need, the delivery ticket with the patient's or designee's signature, the compliance download where the item requires one, the proof of continued need where the rental is past its first year, and a cover letter that tells the reviewer where each of those things is. Get any of it wrong and the claim is denied, and enough denials sends you to round two, then round three, then to prepayment review on that product line — which is the point at which a small DME company genuinely gets into trouble.
Your reimbursement manager does not get to work on this full time. She has the daily denials, the appeals already in flight, the month-end close, and the two prior authorization files that will expire if nobody touches them. Call it three usable hours a day.
Each packet takes about seventy minutes when nothing is missing: find the order in Brightree, pull the scanned documents out of the document management system, cross-check the dates against the policy article, pull the usage download from AirView, verify the delivery ticket signature is legible and dated, assemble it in the right order, index it, and write the cover letter. Forty packets is roughly forty-seven hours, or about sixteen working days at three hours a day.
Except that is the optimistic version, because roughly a third of the packets turn out to be missing something that lives in a physician's office. Those need a records request, and a physician's office takes ten to fourteen days to answer one. If you do not discover the gap until day thirty because you are working the claims in order, you are now requesting records with two weeks left on the clock. That is the actual failure mode in DME audits: not that the documentation did not exist, but that nobody found out it was missing early enough to go get it.
Notice what is not the problem here. Each individual packet is well understood — your reimbursement manager could describe the requirements for a K0823 file or an E1390 file from memory. The problem is that forty of them happen one after another, in the order the letter listed them, by one person who has three hours a day.
Hear it before you finish reading
Talk to a live CallSphere AI voice agent for healthcare in your browser — 60 seconds, no signup.
Several AI agents splitting one job means the work that used to happen one claim at a time happens on ten claims at a time, so the forty-eighth hour of work finishes on Wednesday afternoon instead of in week four. That is the whole idea. Agent Teams shipped as a research preview alongside Claude Opus 4.6, and multi-agent working is now a normal option rather than a laboratory demonstration: several agents take separate pieces of one large job, work at the same time, and their results get merged.
flowchart TD
A["45-day ADR letter lists 40 claim numbers"] --> B["Work split four ways"]
B --> C["Claims 1-10: pull order, notes, delivery ticket"]
B --> D["Claims 11-20: pull order, notes, delivery ticket"]
B --> E["Claims 21-30: pull order, notes, delivery ticket"]
B --> F["Claims 31-40: pull order, notes, delivery ticket"]
C --> G["Merged into one indexed response set with a gap list"]
D --> G
E --> G
F --> G
G --> H["Reimbursement manager reviews and signs each packet"]
The letter arrives Tuesday morning. By Tuesday evening, four agents have worked ten claims each: located every claim in Brightree, pulled the associated documents out of your document system, checked each packet against the requirements for that product code, assembled them in reviewer order, drafted the index and the cover letter, and produced one more thing that matters more than any of it — a gap list.
The gap list says: eleven of the forty are missing something. Six need chart notes from three physician practices. Three have delivery tickets where the signature date is illegible on the scan and the original needs pulling from the branch. Two have compliance downloads that stop short of the documented period. Your reimbursement manager opens that list Wednesday morning, and the records requests go out on day two of forty-five instead of day thirty.
Then she does the part only she can do: she reads all forty. Twelve minutes a packet, about eight hours across two days, checking that the clinical documentation actually supports what was billed, that nothing in the file contradicts anything else, and that the cover letter says what she wants it to say. She signs each one. The response goes out in week two with every gap closed, rather than in week seven with three claims abandoned.
Illustrative, based on a forty-claim round-one probe. Adjust the packet time to whatever your own team actually experiences.
| One person, serial | Split four ways, then reviewed | |
| Packet assembly | 70 min each, 47 hours total | overnight, plus 8 hours of review |
| Usable hours per day on the audit | 3 | 4 |
| Working days to a complete set | about 16 | about 2 |
| Day the missing-document gaps are discovered | rolling, through day 30 | day 1 |
| Days left to chase physician records | about 15 | about 43 |
| Claims likely abandoned for lack of time (illustrative) | 3 | 0 |
| Value of those claims at $410 average | about $1,230 written off | - |
The written-off claims are the small number in that table. The large number is the one you cannot put a figure on honestly: what a failed round one costs you. Round two is another probe, round three after that, and a referral to prepayment review on a product line means every claim in that family gets held pending documentation — which for a company carrying rental equipment on its own balance sheet is a cash problem long before it is a compliance problem.
Do not let anything sign an audit response. The signature is an attestation from your company to a Medicare contractor, and it belongs to a person with a name and a credential.
Still reading? Stop comparing — try CallSphere live.
See the healthcare AI agent handle a real call — complete, industry-specific, and live in your browser. No signup.
There is a specific and important limit here. Agents are good at gathering, checking against a stated requirement, and assembling. They are not good at the argument — the judgement about whether this particular patient's chart notes genuinely establish medical need for this particular item, which is the exact question the reviewer is asking. When a chart note is thin but arguable, that is a conversation between your reimbursement manager and possibly your compliance consultant. Handing that judgement to software is how you end up attesting to something you cannot defend.
Two more. Splitting work across agents means each one sees only its slice, so patterns across the whole set — the same physician's notes being thin on eight of the forty, or a delivery ticket template your branch has been using wrong since March — get missed unless somebody looks at the merged result as a whole. That is a job for your manager, and it is often the most valuable ten minutes of the entire exercise, because it tells you what to fix so round two never happens. And every one of these files is protected health information moving between systems, so the business associate agreement and the access controls need to be settled before the first letter arrives, not during.
Yes, and audit response is just the most painful example. The same serial-versus-parallel problem shows up in the capped rental conversion review at month eleven, the annual oxygen recertification sweep, cleaning up an equipment inventory against serial numbers after an acquisition, and repricing your item master when the fee schedule updates. Anything where you do the same careful review a few hundred times in a row is a candidate.
No. Reading scans, faxes and handwriting is exactly what the current generation of models does well, including a delivery ticket signed with initials. What genuinely breaks it is documents that were never filed against the right order, and no amount of parallel work fixes a document that is not there.
More than you would expect, because the bottleneck being removed is precisely the one-person bottleneck. The honest requirement is that somebody has to define what a complete packet looks like for each of your top product codes. That is a day of work by someone who knows your business, and it is the same day of work whether you have one billing person or nine.
They will, occasionally, which is why the review step is not optional and why the first probe you run this way should be checked packet by packet against how your manager would have built it. If the assembly is wrong in a consistent way, the definition of a complete packet was wrong — fix that once and it stays fixed.
Write down, on one page per product code, what a complete documentation packet contains for your five highest-volume codes. Most DME companies have never done this; it lives in one person's head, which is also the single biggest operating risk in your billing department. Do that now, while nothing is on fire, and the day a forty-five-day letter arrives you are configuring a job rather than inventing a standard.
One connected note. An audit round generates calls in both directions — physician offices returning your records requests, patients asking why you need a new order signed. CallSphere builds AI voice and chat agents that answer your phone line and web chat around the clock, capture who called about which patient, and route the message to the right person, so the callbacks during an audit month do not land on the same manager who is trying to close forty packets.
Written by
Sagar Shankaran· Founder, CallSphere
LinkedInSagar Shankaran is the founder of CallSphere, where he builds production AI voice and chat agents deployed across healthcare, hospitality, real estate, and home services. He writes about agentic AI, LLM engineering, and shipping voice agents that handle real calls in production.
See how AI voice agents work for your industry. Live demo available -- no signup required.
A 640-line BOM scrub eats three days of your buyer's week. Here is what splitting the RFQ across several agents does to quote throughput at an EMS shop.
A 250-claim PBM desk audit eats three weeks of a technician's time. Four agents splitting the pile turn it into an afternoon and a 19-claim exceptions list.
A 6,100-page open-records request takes 23 days serially. Splitting the first pass across parallel agents cuts it to 6, with every redaction signed by a clerk.
The 19-day lag between a distributor price increase and the new shelf tag quietly eats food margin. What parallel agents change, with the arithmetic.
12,000 objects at 15 records a day is 800 working days. How AI agents working in parallel cut a museum's collections-online backlog to about five months.
Why a 45-line RFQ takes five days in a precision machining shop, and what changes for the estimating desk when several agents read all the prints at once.
© 2026 CallSphere Inc. All rights reserved.
Made within San Francisco
Watch how CallSphere handles real customer calls, schedules appointments, and processes payments — live.
Try Live DemoBook a DemoCalculate Your ROI