Medical bill review: the four days before the opinion

What a 3,000-page case actually costs

One case I prepared by hand had more than twenty provider bills across roughly 3,000 pages of scanned PDFs. Out of those 3,000 pages, about 150 were billing pages. Everything else was records, correspondence and duplicates. Finding those 150 meant paging through all 3,000, because there is no index and no reliable way to tell a billing page from a treatment note without looking at it.

The 150 billing pages produced more than 1,000 rows in a spreadsheet. Every charge line on every bill, every EOB and every ledger became a row: date of service, provider, code and billed amount, with columns waiting alongside for the fee schedule figure and the UCR benchmarks.

Three to four days, start to finish. None of it was analysis.

The expert does not start until the sheet is finished

Expert bill review splits cleanly into two jobs, and only one of them requires a license.

The first job is assembly. Locate the bills inside the record, capture every charge line accurately, and put the comparison figures next to them. It is careful work and it has to be right, but a trained assistant can do it, which is exactly why experts hire one as soon as their caseload makes it impossible to keep doing it themselves.

The second job is the opinion. The expert reads the recommended amounts against the fee schedule and the UCR benchmarks, decides what the charges should have been, and narrates that reasoning in a report she signs. That is the work the referring attorney is paying for. Software can help her write up a conclusion she has already reached, which is a different thing from reaching it.

The problem is the order. Job two cannot start until job one is finished. So the expert sits behind three or four days of data entry before she can spend three hours doing the thing she is qualified to do.

How long does prep take on a large bill review?

Three to four days for a 3,000-page, twenty-provider case, from a preparer who has done it for years. A small case with two providers and 200 pages is a couple of hours. Prep time tracks the page count almost linearly, because every page has to be looked at before it can be ruled out.

That relationship is the part worth sitting with. The opinion on a 3,000-page case is not meaningfully harder than the opinion on a 300-page case. The reasoning is the same, the benchmarks are the same, the report is a similar length. What changed by a factor of ten is the assembly. Page count sets the delivery date, and page count has nothing to do with the part that requires expertise.

Every expert in this field has felt that and most have stopped noticing it, the way you stop noticing a commute. It is also why large referrals get quoted long turnarounds, or declined.

What a transcription error actually costs

Writing 1,000 rows by hand over four days produces errors. Not because the preparer is careless, but because attention is not a constant across four days and a thousand repetitions. I was good at this, and roughly one sheet in ten came back to me for correction.

There are two ways it goes wrong, and they cost very different amounts.

The error you catch yourself.You spot the transposed amount or the wrong date of service, you fix it, you redo the affected section. Time extends. Nobody outside the room ever knows. This is the visible, survivable failure, and it is why “the prep took four days” was always an understatement. It took four days plus the loops.

The error nobody catches. The expert does not re-verify 1,000 rows against 150 source pages. That would defeat the entire point of having a preparer, and no expert has the hours for it. So an error in the sheet becomes an error in the recommended amount, and it rides into the signed report. The person who finds it is the referring attorney, after the report has been delivered.

That second failure is not a spreadsheet problem. The expert looks unreliable to the firm that sends her work, the report comes back for correction, and the next referral is smaller or does not arrive. In this business the cost of a wrong report is measured in future cases, not in rework hours.

The missing bill is the expensive error, not the wrong number

A wrong number has a chance of being noticed. A figure that looks strange next to the fee schedule invites a second look, and an experienced reviewer has caught bad rows that way.

A bill that never made it onto the sheet has no such chance. There is nothing on the page to look wrong. It is not a row with a mistake in it, it is an absence, and an absence cannot be spot-checked. When you are hunting 150 billing pages inside 3,000, that is the mistake waiting to happen: not a mistyped amount, a provider whose bills you paged straight past at hour six.

This distinction should drive how any automated version of this work is designed, and it drives ours.

What changes when the assembly is automated

The Expert Report Engine does the first job. It reads the full record, identifies the billing pages, extracts every charge line, and returns the spreadsheet with fee schedule and UCR benchmark columns already populated. The expert opens a sheet she did not have to build.

What that removes is the page-count penalty. A 200-page case and a 5,000-page case come back the same day, because reading 5,000 pages costs a machine almost nothing and cost a person a week. The expert’s own work is unchanged: she still reads the numbers, forms the opinion, writes the report and signs it. Nothing signs itself, and the review step is not optional.

The practical effect is on capacity rather than on quality. An expert who can return a large report in days instead of weeks can accept referrals she used to decline, and attorneys send more work to the expert who turns things around. One of our clients put it more directly than we would have: when he sends reports faster, he gets more work, so he makes more money.

What happens when the system is not sure

The design principle is deliberate: the cheap error is an extra row, and the expensive error is a missing bill, so the system is built to make the cheap one.

When a page is ambiguous, it goes in. When confidence on a line is low, the line is flagged in the sheet rather than dropped from it. The result is that a reviewer occasionally deletes something that should not have been captured, which takes seconds, instead of discovering months later that something was never captured at all, which cannot be fixed by looking harder at the sheet.

This is also the honest answer to the question every buyer asks, which is what happens when the system is wrong. The answer is that the expert catches it, the same way the expert has always been the last line, except that now she is reviewing a sheet she did not build at hour thirty of building it. The reasoning behind that design is the subject of why we are called System 2.

Where this does not help

We sell the thing described above, so treat the recommendation accordingly, and here is the case where it does not hold.

If your typical case is 40 pages with two providers, prep is not your bottleneck. You can build that sheet in twenty minutes and the arithmetic never favors changing anything. The work described here matters when page counts run into the thousands, when a single case has a dozen or more providers, or when you are already declining referrals because you cannot promise a date. Below that, an expert working alone with a spreadsheet is a perfectly good system, and we would rather say so than sell into a bad fit.