There are three ways a practice gets its billing done. Providers handle their own, someone inside the building handles it, or a company outside the building handles it. Most owners pick one for a reason that made sense at the time, usually size or cost, and then treat the choice as settled. It rarely is. Each of the three has a failure mode that belongs to it specifically, and it produces a distinct pattern in your numbers. The useful question is not which model is best, because there is no answer to that. It is which of the three failures is currently running in your practice, because that tells you what to fix.
Model one: the provider bills their own work
The smallest practices start here because there is nobody else. The clinician sees the patient, writes the note, picks the codes, and the charge goes out. It is cheap in the sense that nobody new is hired.
The failure is that the billing competes with the clinical day and always loses. Nobody chose to deprioritize it. A patient runs long, the afternoon compresses, and the administrative tail of the visit slides to tonight, then to the weekend, then to whenever.
At one practice we assessed, 628 chargeslips had never been signed. Roughly $94,000 of care that had been delivered, documented, and never billed. It was not a backlog anyone had decided to accept. It accumulated a few at a time, invisibly, because nothing put the unsigned list in front of the person who could clear it.
The second signature of this model is concentration. When we grouped rework by the person whose work produced it, five providers accounted for 80 percent of it. The same documentation gaps repeating in the same way every week. None of those five were careless. None had ever been shown the pattern, because no report in the practice organized denials by provider, and a gap you have never seen described is a gap you keep making.
What this model looks like in your numbers: revenue that lags the schedule for no visible reason, a denial pattern that concentrates in a few names, and work that is complete clinically and absent financially.
Model two: someone inside the building does the billing
This is the step most practices take next, and it fixes the first problem immediately. Billing becomes somebody’s actual job instead of the last thing on a clinician’s night.
The failure that replaces it is distance. The person billing is now downstream of the person who created the work, and the two rarely talk in a structured way. A claim arrives on the biller’s desk already carrying whatever defect it was born with, and the biller’s job is defined as getting it paid rather than asking why it keeps arriving broken.
So a loop forms. The claim goes out, comes back denied, gets corrected by hand, goes out again. We traced one claim that had been submitted 18 times. Not 18 different claims. The same one, cycling, until it was near the filing deadline where it would convert to zero regardless of who was right about the coding.
Nobody did anything wrong. The biller resubmitted because resubmitting is what the system permits, and nothing in the process ever asked whether the next attempt would fail for the same reason as the last one.
Underneath that loop, the concentration is the same story as model one. At a different practice, five denial patterns drove 80 percent of the total rework. Five, repeating, each one a combination the payer’s system was always going to reject. Over $600,000 of that work was preventable, and it was preventable because it was predictable.
The tell here is a billing team that is busy, competent, and behind. Effort is high, first-pass rates are not, and the same reason codes appear month after month.
Model three: an outside company does the billing
Larger practices and groups move here for good reasons. Depth of expertise, coverage that does not evaporate when one person takes leave, and a cost that scales with volume instead of headcount.
The failure mode is the most misread of the three, so it is worth being precise. It is not that outside billers do poor work. Most do the job they were engaged to do, and do it well.
The issue is where the job starts. A billing company is engaged to collect on the claims it receives, and the standard arrangement pays a share of what it collects. That aligns the relationship correctly for its actual scope: everything from the claim onward.
The defects, though, are created before that point. A coverage detail nobody verified at scheduling. A note that supports a lower level than the one billed. An authorization that was never obtained. Those happen inside your building, in workflows the billing company does not run, cannot see, and was never asked to change. So the arrangement works exactly as designed and the upstream defect keeps arriving, because nothing in the structure pays anyone to eliminate it at the source.
That gap has a second effect: it hides things. At one practice, $650,000 a year in disputed charges sat inside a category labeled “other” that nobody had reason to open. At the same practice, $104,000 of accounts receivable read as collectible while it was being quietly reclassified through write-off codes nobody audited. Neither was concealed. Both were simply outside the boundary of what anyone had been asked to look at.
The tell for this model is a reporting relationship that looks healthy at the summary level and cannot answer a specific question. Collections look reasonable, the monthly summary balances, and nobody can tell you why one payer’s yield dropped or which five patterns drove last quarter’s rework.
The fourth arrangement, which is the most common of all
Describing three models is tidy, and almost nobody runs a clean one. Most practices past a certain size run a blend without ever having decided to.
The clinician picks the codes. Someone at the front desk handles eligibility, when there is time. An in-house person builds and submits the claim. An outside company works the denials. That is four hands on one dollar, and each pair of hands was added at a different moment to solve a different problem.
Blends are not worse than the pure models. What they add is handoffs, and a handoff is where responsibility gets thin. Each person in that chain is doing their own job correctly and none of them owns the outcome, so a defect can pass through four sets of hands without anyone being wrong.
You can see this in the questions a blended practice cannot answer. Ask who decides whether a claim is ready to submit and you often get two names and a pause. Ask who is accountable for the first-pass rate and you get a description of a process rather than a person. That pause is the finding. A number nobody owns does not improve, because improving it is nobody’s assignment.
The other effect is that blends conceal which failure you have. The provider concentration from model one, the rework loop from model two, and the upstream blind spot from model three can all be running at once, in different corners, cancelling each other out in the summary numbers while each one costs you separately.
Which failure is yours
You can identify it in an afternoon, and the sequence matters because each answer narrows the next question.
Start with timing. Take last month’s completed visits and count how many had a signed note and a submitted claim within two days. If a meaningful share did not, your problem is upstream of billing entirely and you are looking at the model one failure, regardless of who does your billing. Work that has not been captured cannot be worked by anyone.
If that number is healthy, move to first-pass. Of the claims that went out, how many paid on the first submission without a human touching them again. A low number here with a busy, competent team is the model two signature: the defects are arriving faster than anyone can prevent them, and the response is repair rather than prevention.
If first-pass is also healthy, go to yield. For each of your five largest payers, take dollars actually paid and divide by dollars they ruled on, then compare each payer against its own history rather than against each other. A payer drifting down here while every speed metric stays clean is the model three signature, because it is exactly the kind of thing that lives outside what anyone was asked to watch.
Run all three and you will usually find one dominant failure and one secondary. Fix in that order. The most common mistake is changing the billing arrangement before running this sequence, which swaps a known failure for an unknown one and costs a year.
What all three have in common
Read those together and the pattern is hard to miss. In every model, the problem is created upstream of whoever is being held responsible for the money.
The provider model puts the work on someone whose day is already full. The in-house model puts a person between the defect and the payment without giving them authority over the defect. The outsourced model draws a contract boundary between the two, and the defect sits on the far side of it from the person paid to handle the consequence.
Changing models moves the boundary. It does not remove it. Practices that switch usually get relief for a quarter or two, because the new arrangement is attentive and the backlog gets worked, and then the same underlying rate reasserts itself in a new shape.
The fix works in any of the three
The good news is that none of this requires picking a different model. It requires a check at the point where defects are created rather than at the point where they are discovered.
That means three things. Know your patterns, which is a matter of grouping twelve months of denials by reason and payer and sorting by dollars rather than claim count. For most practices the top 80 percent of the money sits behind a handful of combinations, not dozens. Put a check in front of submission that holds any claim carrying one of those combinations and routes it to a named person with the specific defect identified. And group the same data by provider, because the practice side concentrates the same way the payer side does, and a report organized by reason code will never show you that five people are producing most of the rework.
At the practice that ran this, rework dropped 62 percent and cash conversion moved from 81 days to 42, a 48 percent cut, with no new hires and no change of billing model. The team did not work harder. Work stopped arriving that should never have existed.
We find why. And we fix it.
Your check this week
Whichever model you run, one question separates a working setup from a busy one: who is responsible for making sure a defect does not happen again, as opposed to fixing it after it does.
Ask it out loud. If the honest answer names nobody, or names someone who has no authority over the step where the defect is created, you have found the gap and it is the same gap regardless of which of the three you use.
Then run the count. Pull last month’s denials, group them by reason code and payer, and see how few combinations make up the bulk of the dollars. The shorter that list, the more of your rework was preventable.
Related reading
- The Denial Machine, how payers automate the answers your team responds to by hand
- Your real denial rate, counted honestly instead of flatteringly
- Where denials are manufactured, the five places they originate
- The clean claim routine, the rhythm behind a high first-pass rate
We will run your denial history and show you your patterns, your provider concentration, and the dollars behind each. Grab 30 minutes with us. Prep nothing.