What the first 30 days found in $6.39 million of PT accounts receivable

The engagement

The client is a billing company whose clinics belong to a multi-clinic outpatient physical therapy franchise group. Its clinics share one practice management and EMR platform, and its AR follow-up, prior authorization, verification of benefits and billing support are worked by our team on a full-time-equivalent model. The AR file we received was dated September 15, 2026.

We do not start an AR engagement by working the oldest claims. We start by measuring the whole book, because the shape of the AR decides how many people are needed, which queues come first and which problems are not AR problems at all. This case study reports what that measurement found. It does not report collection results, because the work has just begun.

First, make the file tell the truth

Before any finding can be trusted, the export has to be read correctly. Three traps in this file would have produced a wrong picture if taken at face value.

• Balance had to be taken as the sum of the six aging buckets. Last billed minus paid tied to the balance on only 51.6% of rows.
• The 8,151 rows with no billed-to-primary date were the secondary bucket, not unbilled claims. Treating them as unbilled would have overstated the billing backlog.
• The file kept only the last comment per claim, so every touch and productivity count drawn from it is a floor, not a total.

What the AR baseline showed

Most of the risk was not in the age of the AR. It was in what had never been touched. 32.8% of AR was over 90 days, but 58.3% of the dollars, $3.73 million on 17,441 lines, had no comment at all. Another $1.52 million had passed its own follow-up date.

The most time-sensitive queue was small in lines and large in consequence: $218,581 on 798 primary claims with dates of service between 300 and 365 days old. Medicare requires a claim to be filed within one calendar year of the date of service, and many commercial contracts are shorter, so this queue had to be worked before anything older or larger.

Concentration mattered too. The largest single clinic carried about $608,000 of AR, and 75% of it was over 90 days.

What the first production review showed

We then read the August and early September production logs against the AR file. The question was not how many lines were touched, but whether touches were moving claims to a new status.

34% of worked lines had already been paid by the payer. Across August and September, 496 paid primary lines worth $179,272 were still open in AR, including 306 lines and $123,002 in September alone. That is a payment posting gap, not a collection problem, and calling the payer about it wastes a follow-up slot.

Of the denials, 60% were front-end: invalid insurance or coordination of benefits, missing authorization and taxonomy. One taxonomy issue alone accounted for 108 lines on four patients with one commercial payer, $26,563 in total.

Work selection also needed to change. 30% of matched worked lines carried balances of $50 or less, $14,617 in total, while only 121 of the 798 filing-window lines had been worked and 10 clinics holding $755,011 of AR had no production at all. 47% of worked lines ended in a status asking the clinic for help or review.

• Paid but still open: 496 lines, $179,272 across August and September.
• Front-end denials: 60% of denials.
• Low-balance work: 30% of worked lines at $50 or less.
• Filing-window coverage: 121 of 798 lines worked.
• Clinics with no production: 10, holding $755,011.

Tying AR to the schedule

An operations export from the same platform let us join scheduled visits to claims by claim number. 25,324 of the 28,873 AR lines matched, with the same date of service on every match. The join surfaced problems that an AR file alone cannot show.

770 AR lines worth $141,777 were already closed in the practice system, and 1,626 arrived visits had no claim at all, 796 of them older than 30 days. Between March and August 2026 the clinics scheduled 89,960 visits: 85.5% arrived, 11.2% were cancelled by the patient and 1.9% were no-shows. At check-in, 3,976 visits were short of the expected copay by a combined $158,818.

What we changed in the first 30 days

None of the changes below required new software. They reordered the work so the people already on it moved claims instead of touching them.

• Put the 798 primary claims at 300 to 365 days at the top of every queue until each one had a dated action.
• Routed paid-but-open lines to payment posting instead of payer follow-up.
• Sent front-end denials, including the taxonomy cluster, back to their root cause with the clinic, rather than reworking each line.
• Batched follow-up by patient, payer and day, which is how 76% of recent lines were already being worked.
• Deprioritized balances of $50 or less unless they shared a root cause with a larger claim.
• Changed the productivity measure from lines touched to lines moved to a new status.
• Sized staffing from the measured book: at 40 claims per day the model called for 29.5 FTE for the first 90 days and 25.3 at steady state; at 80 lines per day, the level suited to batched follow-up, it called for 14.7 and 12.6.