The Other Side Already Has One

The first AI you ever met in this job was working for the payer.

A clean claim goes out on Tuesday morning. The denial is back before lunch. Nobody read that claim. An edit engine read it, matched it against a rule, and returned a CARC with two remark codes that between them tell you almost nothing you can act on. Then you spend forty minutes establishing that a human being ought to look at it. Eventually one does, and it pays.

So when the two loud stories about this technology reach you, you’re not starting from nothing. You have met the thing. It has already cost you an afternoon.

The first story says coding is finished. There is a slide for it at every user conference, a line going up and to the right over a percentage of charts handled with no human touch. A relative asks, gently, at a family thing, whether you have thought about what you would do next.

The second story is the same technology pointed the other way, sold from a booth. This one ends denials. Clean-claim rate in the high nineties, days in A/R cut in half, the aging report finally behaving itself. Thirty days free, card details now.

Both describe a year nobody has worked yet, and neither moves anything on your desk this afternoon. The ERA that posted this morning still has thirty-odd denials in it. The 90-plus column is still the wrong shape. The provider still hasn’t answered the query you sent a week ago Tuesday, and the appeal clock on the big medical-necessity denial runs out in eleven days.

This book has no opinion about the future of coding. It is about that ERA.

Start with the first story, because a short list finishes it, and every entry on that list is something you did this week.

It cannot read the chart. It can read text, and a note is text, so this sounds arguable until you watch it happen. It cannot tell whether “follow-up, stable, continue current medications” means a condition is being managed or the physician was fifty minutes behind. It cannot know that history means one thing in this doctor’s notes and something else in his partner’s. It knows what was typed. It does not know what was meant, and the distance between those two is where your whole profession lives.

It cannot assign a code you could defend. It will produce one — fast, confident, formatted, with a tidy justification underneath. In an audit, tidy is worth nothing. The only question anyone asks is what the documentation supports, and the person answering has a credential number, a name, and a chair to sit in. No model has any of the three.

It cannot make a payer honor its own contract. It cannot hold for thirty-five minutes. It cannot get a rep to actually open the claim, catch the small change in tone when they see the denial was wrong, and take a reference number before the call drops. It cannot be the person who calls a third time about one claim, which is often the entire mechanism by which claims get paid.

It cannot catch a physician between patients. That is a twelve-second conversation in a corridor, and it works because you are a person he knows rather than a task in his inbox.

And it has never seen your contract. Ask what your appeal window is and back comes the industry-standard answer in a steady voice, and that answer is wrong every time your contract says otherwise. On the payers that hurt you most, it usually does.

That is the list, and it covers the parts of this job that carry consequences. The chart. The code. The credential. The phone call.

That took less than a page. A reassurance needing a whole chapter is usually a sales pitch.

The unease behind the first story is still worth taking seriously. It is just aimed slightly off target.

Software has been eating the clean end of this work for twenty years. Scrubbers, edit engines, auto-adjudication, the suggestions that surface in the encoder before you have finished reading the note. What that software takes is the volume that follows a rule: well documented, template driven, nothing surprising in it. What it leaves behind, every time, is exceptions. A rule engine doesn’t remove work. It sorts work into the part it can handle and the pile it can’t, and that pile is denials, appeals, medical necessity, audit response, contract quirks, and anything that needs a person to decide something.

That pile is where your job already lives, and it is growing rather than shrinking. That is a reason to read on, not a reason to stop. The billers and coders who come out of this decade in good shape are fast on exceptions, fluent in how each payer behaves, and visibly the person everyone else asks. Chapters 7, 8 and 9 exist for that.

Now the harder half. If this book claimed nothing was changing, Chapter 4 would give it away, and you would have no reason to trust the rest.

Something is moving, and it is not the coding. It is the writing that surrounds the coding.

Nobody mentions the writing while you are studying for the credential. You learn guidelines, anatomy, edits, modifiers, the difference between a new problem and an established one. Then you take the job and find the day is mostly sentences. The appeal letter. The medical-necessity narrative that lines documentation up against published criteria without going an inch past the record. The query rewritten four times to be sure it isn’t leading the answer. The letter explaining a deductible to somebody upset and not wrong to be. The front-desk script nobody has ever written down. The escalation letter to the payer that has ignored three follow-ups. The note after the call. The month-end report. The encyclopedia of payer quirks that lives in your head and leaves the building with you every night.

None of that is coding. None of it is billable. Most of it happens at twenty to five, or on your lunch, or not at all.

That stack is where these tools earn their keep. Not the decision about whether an appeal is worth filing. The nine paragraphs that turn a decision you already made into a letter.

They are bad at things too, and the failure mode matters more here than in most jobs. They invent, and the invention arrives sounding precisely like the parts that are true. Ask what a remark code means and you may get a meaning that is plausible and wrong. Ask about a filing limit and you get a confident number belonging to somebody else’s contract. Worst of all, ask for support and you may get a policy citation with a realistic-looking number on it, in a letter going out over your name to a payer that keeps a file on you.

Every workflow in this book therefore cuts the job along one line. Facts, codes, figures, deadlines and policy language stay with you, verified against the current books, the guidelines, the edits, and the payer’s own materials. Structure and sentences are its half. Where the line is easy to lose sight of, a WATCH OUT box names the check and says how long it runs. A minute, near enough. That minute is the whole price of using it, and the evening it buys back is worth more.

The commercial part, once. An appeal filed the week the denial posts is a different animal from the same appeal filed four days before the deadline, and a denial nobody appeals is never recorded as a loss at all. It leaves as an adjustment, and adjustments do not argue back. You know this already. The only thing that has moved is the price of the letter. It used to take an evening. Now it takes ten minutes plus the verification you were doing anyway.

Picture a Thursday in the last week of the month. Nothing unusual happens on it, which is the point of it.

The ERA posted before you got in. By eleven you have cleared the mechanical denials — the corrected claims, the eligibility ones, the two that are plainly the front desk. Three real ones are left. All medical necessity, all worth money, all winnable.

You write the first properly at ten past two. You pull the current policy off the portal, map the documentation against the criteria one line at a time, notice a gap and decide it is a query rather than a stretch. What comes out is good. It will probably win.

You build the second out of the first. Half an hour, and most of that goes on finding the places where the other claim’s details are still sitting inside a paragraph.

The third does not happen. It’s twenty to five, and there is a patient at the front window holding a statement she does not understand and getting louder about it. She is a person standing in front of you, so you take that instead. The appeal goes back on the worklist with a note saying what it needs.

You see that note again the following week, and the week after. Then one week it isn’t there, because the filing window closed and it left the building as a write-off. Nobody calls it a loss. No report in your system will ever describe it as one.

Look at where the problem actually sits. Not in your writing: you produced an excellent letter at ten past two, from a blank page, in the middle of a bad day. It is arithmetic. A real appeal takes forty minutes, Thursday asked for three of them, and you are one person.

Now the contents, in plain terms, so you can decide whether to carry on.

This is a working manual for one thing: using an ordinary AI chat tool on the words that surround a claim. ChatGPT, Claude or Gemini; the book does not much care which. Denial decoding and appeal skeletons. Medical-necessity narratives built from the record and never past it. Provider queries that don’t lead. Patient letters that explain a balance without giving money away. Front-desk scripts, escalation letters, call sheets and call notes. Worklists that sort themselves by deadline risk, monthly reports that explain the numbers instead of listing them, payer-quirk sheets, bulletin triage, study plans for the next credential, and the resume that still describes your last job. All 150 prompts are numbered and grouped, and they repeat in a library at the end, so you can put your hand on one while a hold recording plays in your other ear.

What it is not: a coding manual, an encoder, a compliance officer, or any authority on what a payer published last quarter. It will not tell you what to bill, and nothing in here asks it to. Two lines never move. Nothing that identifies a patient goes into a chat window, and no code goes on a claim because a machine suggested it. Where the tool comes close to something costly, a RULES CORNER box marks the boundary, and Chapter 10 sets them all out together. Read that chapter before real work goes anywhere near a chat window.

You need no technical background for any of this. A phone or a computer, an email address, and half an hour. The setup is Chapter 2. Chapter 3 is the one skill that matters: saying what you want clearly enough that the first draft comes back close. You already have it, because a compliant provider query is that skill under harder constraints than anything in this book. That is not flattery. It is the identical discipline, pointed at a different reader.

One more thing about scale, and then you can start.

Nothing here is life-changing. It works on the stretch between four and six, and after a month, on most of the week. The Promise Page opposite lays it out chapter by chapter and the total at the bottom is about eleven hours. That figure is honest, and it is not a gift. Week one you will be slower than you are now; tools charge up front. Somewhere in week three the arithmetic turns over.

Two things is enough to cover what you paid, and two is what most people take. Usually the denial decoder and the appeal skeleton, worked daily for a month until they stop feeling like a new step, Chapter 9 never opened, four hours a week back, and that is that. It is the intended result, not a shortfall. There is no credit for using all twelve chapters.

Your credential is not the thing under threat here. Somebody still has to read the note, decide what it supports, and then say so out loud when a payer disagrees, with a name and a number attached. What this offers sits well below both stories you were sold and is worth more than either: the appeal filed the day the denial posts, the query answered without a second round, and a queue that stops following you home.

The next few pages show where the hours come from.