Two Hours to Build One Redetermination Packet. Now You Hand Over the Goal and the CMS-20027 Comes Back Filled.

Two hours and ten minutes. That is roughly what it takes one experienced AR follow-up specialist to build a single Medicare redetermination packet properly: pull the remittance advice, find the claim as submitted, get the office note and any op report out of the client's chart system, look up the local coverage determination the contractor is leaning on, fill out Form CMS-20027, write a cover letter that actually addresses the reason code, and assemble the whole thing as one PDF that a reviewer will not bounce.

Which is why almost nobody does it properly. A denial worth $186 does not survive a two-hour packet. So it gets a phone call, a rebill, a "corrected claim" that changes nothing, and eventually a write-off with an adjustment code that makes the aging report look tidy. Multiply that by every denial under a few hundred dollars across your whole book and you are looking at the single largest pile of abandoned money in a billing company.

The math that makes small denials disappear

Every billing company runs the same unwritten rule: work the big ones. A $4,200 surgical denial gets the full treatment. A $186 office visit denied CO-197 for no prior authorization gets a call to the payer and then gets dropped, because at a loaded cost of $38 an hour, two hours of specialist time against a $186 balance earning you a 6% fee is obviously upside down.

The trouble is that the small ones are most of them. Denial volume clusters in the cheap codes: CO-197 missing authorization, CO-16 missing or invalid information, CO-11 diagnosis inconsistent with the procedure, CO-97 bundled into another service, CO-29 past timely filing. Your clients feel this as a net collection rate stuck at 93% when the contract says you will get them to 96%, and they cannot see why.

An appeal packet is not writing — it is fetching. Ninety percent of those two hours is moving documents between six systems that do not talk to each other: the clearinghouse worklist, the practice management system, the chart, the payer portal, the coverage policy, and Word.

What shipped in 2026, and why it is different from a chatbot

In January 2026 Anthropic released Claude Cowork, and on 9 July 2026 OpenAI released ChatGPT Work on GPT-5.6. Both do the same new thing: you give a goal, not a task list. It connects to your files and the applications you already run, breaks the job into steps by itself, works for a stretch without you sitting over it, and hands back a finished artifact — a completed spreadsheet, a filled form set, a document packet. Both were aimed at ordinary office staff, not programmers.

That is genuinely different from what you tried in 2024. Then, the assistant was a chat box: it answered a question and you did the work. You could ask it to draft an appeal letter, and it would produce a nice-sounding letter with no remittance advice attached, no claim copy, no policy citation, and no idea which of your clients' payers wanted it faxed versus uploaded. The 2026 version goes and gets those things and returns the packet.

flowchart TD
  A["Goal: appeal every CO-197 denial over $100 from last week"] --> B["Pull the denial list from the clearinghouse worklist"]
  A --> C["Pull claim as submitted and the 835 remittance"]
  A --> D["Pull the office note from the client chart system"]
  A --> E["Pull the payer coverage policy in force on the date of service"]
  B --> F["Assemble one packet per claim with CMS-20027 filled"]
  C --> F
  D --> F
  E --> F
  F --> G{"AR specialist reviews and signs"}
  G --> H["Submitted to the payer portal, tracked to a decision"]

The goal you would actually type on a Tuesday

Not "write an appeal letter." The goal looks like this: For client Ridge Family Medicine, take every denial from last week's remit with reason code CO-197 and a balance over $100. For each one, build a redetermination packet: cover letter citing the authorization on file if there is one, the claim as submitted, the remittance, the progress note for that date of service, and the coverage policy in force. Fill Form CMS-20027. Put them in the review folder and give me a one-page list of anything you could not find an authorization for.

Then it works. It reads the worklist in Waystar or your clearinghouse of choice, opens each claim in the client's system, finds the note, checks whether an authorization number exists in the referral record, and builds the packets. An hour or two later there is a folder with twenty-two packets in it and a short list of six claims where there is genuinely no authorization to point to — which is not an appeal problem, it is a front-desk problem at the practice, and that list is worth more to your client account manager than the twenty-two packets.

Your AR specialist's job on Wednesday morning is to open twenty-two packets, read the cover letters, check the reason code was addressed, and sign. That is roughly six minutes each instead of two hours each. The specialist is still the one who decides what goes out the door.

What has to change about how you assign work

This is the part owners underestimate. Your worklists are currently written as task lists — "call payer, check status, note account" — because that is what you hand a person who needs to be told the steps. A goal-shaped assignment needs three things a task list never needed: a stated outcome, a stated boundary, and a stated stopping rule.

• Outcome: a signed-ready packet in the review folder, not "research the denial."
• Boundary: which client, which reason codes, which dollar floor, which date range, which payers.
• Stopping rule: what to do when the note is missing, when the authorization does not exist, when the timely filing window already closed. Say it out loud in the goal, or you will get a confident packet built on a guess.

You will also need to decide who writes goals. In most billing companies it is not the newest AR clerk — it is the denials lead or the client account manager, because writing a good boundary requires knowing the payer's appeal window and the client's chart access. Budget an afternoon of your best person's time to write ten reusable goals, one per major denial reason, and treat those as company property the way you treat your fee schedule library.

Twelve months of small denials, with the assumptions on the table

Illustration figures. Assume your book generates 900 denials a month, that 62% of them carry a balance under $300, and that today you appeal 15% of that small-balance group. Assume a 48% overturn rate on appealed claims, an average recovered balance of $205, and your fee at 6% of collections. Assume a specialist costs $38 an hour loaded, and that packet build drops from 2.1 hours to 0.4 hours of human time.

Roughly the same human hours, four and a half times the packets, and about $1,820 a month of additional fee revenue on this one denial category — with $30,000 a month of recovered cash flowing to clients who will renew because of it. Prove it the honest way: run one client for one quarter, and compare overturn dollars per specialist hour against the same client's prior quarter.

Where this still needs a human, without exception

Three hard lines. Nothing goes to a payer unsigned — a redetermination is a formal filing under your client's provider number, and a machine-built packet with a wrong date of service is your client's problem, not a software bug. Nothing touches the medical record itself: the agent may attach a note, never edit one, never restate what the provider documented, and never suggest a diagnosis code the provider did not support. And nothing decides medical necessity. When the denial is CO-50 and the argument is clinical, a certified coder or the provider writes that argument.

There is a quieter failure worth watching for too. An agent that builds packets fast will happily build packets for claims that are past the appeal window, which wastes postage and, worse, teaches your staff to stop reading the output. Put the filing deadline in the goal as a hard stop and audit a sample every month.

Frequently asked questions

Can it get into our clients' systems at all?

Sometimes directly, often through exports. Start where you already have a clean export — the denial worklist and the remittance file — plus a folder of chart notes your client already shares with you. You do not need every system connected on day one to get value out of the first ten packets.

What about HIPAA, since this touches charts?

Same standard as any other vendor: a signed business associate agreement, access limited to the clients and date ranges the work requires, no reuse of the data, and an access log you can show a client during their annual review. If the vendor will not sign a business associate agreement, the conversation is over.

Do we still need certified coders?

Yes, and arguably more of them per denial dollar. What shrinks is document-fetching time. What grows is the number of arguments that need a CPC or CCS to judge whether the modifier and the documentation support the claim.

Which denial reason should we start with?

The one with the highest count and the most mechanical argument — usually CO-16 or CO-197. Save CO-50 medical necessity for after your staff has learned what a good packet looks like.

CallSphere builds AI voice and chat agents that answer business phone lines and web chat, book appointments, and capture leads around the clock. It does not build appeal packets — but the volume those appeals generate lands somewhere, and for most billing companies that is the practice line and the patient billing line. Answering those calls consistently is what keeps your specialists inside the worklist instead of on hold.