70% of PT patients never finish their plan of care. The leak is usually visit 10.

A patient comes in for a knee. The therapist prescribes 12 visits over 6 weeks. The insurer authorised 10. On a Tuesday in week 5 the front desk books visit 11, nobody checks the count, and the clinic either treats for free or calls the patient to say they need to wait while paperwork goes through. Nine days later the extension comes back approved. The patient does not.

That is not a billing error. It is a timing problem, and it happens in the middle of an episode where almost nobody is looking.

Research on outpatient physical therapy puts roughly 70% of patients as never completing the full course of care their therapist prescribed, with about 20% gone inside the first three visits. Clinics that run well land somewhere between 65% and 75% completion. The industry at large sits closer to 55% to 60%. At around $95 a visit and a 40 patient caseload, that gap is $60,000 to $90,000 a year of work the clinic already earned the right to bill.

I've published the full build blueprint for this as a Pro Prompt: Physical Therapy Clinics, Plan of Care Completion & Insurance Authorization Tracker. It is a complete master prompt you paste into Lovable, plus the payer rule model, the risk scoring weights, an 11 step build guide, a 9 step sell and run playbook, and what to charge a clinic for it.

Two leaks that share a cause

Ask a clinic owner why patients drop off and you'll usually get an answer about motivation. People feel better at visit 6 and decide they're done. Some of that is real, and honestly it's sometimes a good outcome.

But a big slice of the drop-off is the clinic's own admin. An authorisation that ran out on a Tuesday. A progress note the payer wanted at visit 6 that nobody wrote, so the extension got refused for insufficient documentation. A cancelled appointment during the paperwork gap, after which the patient found something else to do with that hour on a Thursday.

The two leaks look separate on a spreadsheet. In the clinic they're the same event. The authorisation running out is what creates the gap, and the gap is what loses the patient.

What the numbers actually look like

Commercial plans differ wildly, which is the whole difficulty. Most insurers allow somewhere between 6 and 12 initial visits before any authorisation is needed. Some Blue Cross commercial plans require no review at all for the first 16. HMO plans almost always want authorisation from the start, often in blocks as small as 6. Private plans commonly cap annual visits somewhere between 20 and 60.

Then there's the filing deadline, which is the part that bites. A typical payer rule says the extension request has to be in at least a week before the last covered visit. So the real deadline is not "when the visits run out", it's "visits remaining minus that payer's actual turnaround time". A clinic with an HMO that answers in 5 days and a workers comp payer that answers by fax in 12 is running two completely different clocks on two patients sitting in the same waiting room.

On the Medicare side the mechanics change shape. There's no visit cap, but for 2026 the therapy threshold sits at $2,480 with a conversion factor of $33.40, and a claim that goes past that line without a KX modifier gets denied outright. Per 15 minute unit, therapeutic exercise (97110) runs about $30.10, manual therapy (97140) about $27.17, therapeutic activities (97530) about $34.61. A standard visit billing 3 units is roughly $90. Nobody is tracking a running dollar total in their head across 300 active patients.

What the tool does on a Tuesday morning

The build is a layer that sits beside the EMR rather than replacing it. The clinic schedules a nightly CSV export of patients and visits, which most EMRs can do, and the app reads that against a payer rules table you load once from the clinic's actual contracts.

At 6am every weekday it runs three jobs.

First, the burn-down. For each active episode it counts attended billable visits against the authorised block, works out the date of the first visit that would fall outside the authorisation, and subtracts that payer's real turnaround time to produce a hard filing deadline. Not a reminder. A date, in red, attached to a named patient and a named payer. Episodes where the payer wants a progress note every 6 visits and no note exists get flagged as an authorisation risk too, because that's the single most common reason an extension gets refused.

Second, the risk score. Every active episode gets scored 0 to 100 on how likely that patient is to stop coming. The heaviest weight goes on attendance in the first four weeks, because missing more than 20% of early appointments makes a patient roughly 3.5 times more likely to drop out before the plan finishes. After that: the gap since the last attended visit measured against the prescribed frequency, whether a next appointment exists on the books at all (the most common silent exit), home exercise compliance, pain trend across the last three visits, and movement on the outcome measure against the goal. The score shows its working. A front desk staffer sees "no next visit booked, 2 visits left on authorisation, missed 2 of 5 in the first month", not a bare number out of 100.

Third, the digest. One email to the front desk listing authorisation actions due today with their hard deadlines, red risk patients with a recommended action each, patients with nothing booked, and the week's revenue at risk as a single dollar figure. That last number is the one that keeps the tool being used.

The extension packet drafts itself

This is the part that saves the most hours per week, and it's the part clinics are most sceptical about until they see it.

When the biller opens an extension request, the app pulls that patient's own data: visit count and dates, outcome measure baseline and current score and goal, the pain trend, the functional notes from recent visits, and the diagnosis. It sends only that to a language model, with the patient's name stripped out first, and gets back a draft medical necessity narrative. Measurable progress to date, the functional deficit that remains, the specific interventions planned, and a clinical reason for the exact number of extra visits requested.

Then it puts the draft on the left of the screen and every source field on the right, so the clinician checks each claim before signing. Nothing submits without a human clicking approve. The approved packet renders to PDF and goes out by whichever route that payer uses, portal, fax cover sheet or email.

A biller doing this by hand spends 15 to 25 minutes per request and writes the same five sentences in a slightly different order every time. The PT authorisation blueprint brings that down to about 3 minutes of reading and correcting, which matters because the slowest step is the one that makes extensions get filed late.

A worked example

Marissa, 44, rotator cuff, referred by an orthopaedic practice. Her HMO authorises in blocks of 6 and takes 5 working days to answer an extension. Her therapist prescribed 14 visits at 2 a week.

Week 1: she attends both visits. Baseline on the shoulder outcome measure is 28 out of 80, goal is 64. Green.

Week 2: she no-shows on the Tuesday, attends the Thursday. The score picks up 30 points for early attendance, because one miss out of four scheduled in the first fortnight is already past the 20% line. Amber. A rebook message goes out the morning after the no-show, automatically, and she books a Saturday slot.

Week 3: she attends both. Visit 5 of her 6 visit authorisation. The burn-down fires, because visits remaining has hit the clinic's trigger of 3 or fewer and the HMO needs 5 days. The authorisation queue now shows a hard deadline of the following Tuesday, which is the date of visit 7. The biller opens the extension builder on the Thursday afternoon. It drafts a narrative citing her move from 28 to 41 on the outcome measure, pain down from 7 to 4, and a remaining deficit in overhead reach, and it requests 8 more visits to reach the prescribed 14. The therapist reads it, corrects one line about her work duties, approves. Submitted Thursday, approved the following Wednesday for 6 of the 8 requested.

Week 4: approved for fewer visits than asked, so the burn-down recalculates immediately and the new count shows 11 total against a prescribed 14. That gap gets flagged early rather than discovered at visit 11.

Week 6: she misses a Tuesday and books nothing. No next appointment plus a gap past double her expected interval pushes her to red. The system queues a call task rather than a text, because a red patient needs a voice. The front desk rings, finds out her shifts changed, moves her to a 7am slot.

She finishes 13 of 14 visits. Without any of that she finishes 6, which is the median outcome in a clinic where nobody is watching. The difference is 7 visits, about $665, on one patient.

Who builds this, and what it's worth

The buyer is an independent clinic with 1 to 6 clinicians and 150 to 600 active patients. The owner is usually a treating clinician putting in 30 hours a week on the table, which is exactly why nobody is tracking visit counts. Hospital-owned groups and private equity backed chains already have a revenue cycle team and a 9 month procurement cycle, so skip them.

The qualifying question is one line: what percentage of your patients complete their full plan of care? Most owners can't answer it. The ones who can usually say something under 60%. Either answer opens the conversation.

On price, the blueprint sets out a setup fee of $3,500 to $6,000 and a monthly retainer of $497 to $897, with a performance element on top. The version I'd push is $75 for every completed plan of care above a baseline taken from the clinic's own 90 day EMR export, which on a 40 patient caseload moving from 55% to 68% is about 5 extra episodes a month. For clinics that resist a per-episode number, 8% of recovered authorised revenue with a monthly cap gives them something they can forecast. Full pricing structure and the sales script are in the Pro Prompt.

The setup fee has to cover the compliance work, because this app holds protected health information and that's real work. Business associate agreements with the clinic and with every vendor touching patient data. The clinic on a paid Supabase plan rather than the free one. Patient names stripped before any call to the drafting model. Row level security so a therapist sees only their own caseload, enforced in the database and not just in the interface. Charge for it properly. It's also the reason a cautious clinic owner says yes instead of "let me think about it".

Three things that will slow you down

The EMR export. Some systems schedule a nightly CSV without complaint. Others make you train the front desk to run it manually each morning, which takes 2 minutes and works fine, as long as one named person owns it.

A2P 10DLC registration. Patient SMS to US mobile numbers needs the clinic registered as a brand and the tool registered as a campaign. Approval commonly takes several days and gets rejected for vague use case descriptions. Start it on day one of the build, not the week you plan to go live.

Consent. Don't import 400 patients with SMS consent set to true by default. Set it false, give the clinic the opt-in wording for their intake paperwork, and let the front desk tick each patient as they confirm. Slower, and the only version that doesn't produce a complaint in month 3.

Why build it now

Prior authorisation requirements for outpatient therapy have been expanding, not shrinking, and the 2026 Medicare changes give every clinic a fresh reason to look at how they track thresholds. At the same time the completion rate problem has sat at the same 55% to 60% for years, which tells you nobody has solved it with scheduling reminders alone. Appointment reminder software is everywhere. Software that knows a specific HMO needs 5 days and this patient's visit 7 falls on Tuesday is not.

One last piece worth doing properly: ship it as a real app, not a browser bookmark. The blueprint's final step wraps the finished build through AppBuild.DIY into native iOS and Android apps plus a PWA, with push notifications and camera access switched on from the Native Capabilities panel. An authorisation expiring in 2 days sitting in an email that gets opened at 11am is useless. The same alert on a lock screen at 7am gets dealt with before the first patient arrives. Therapists get pushed when someone on their caseload turns red, so they raise it at the next visit rather than reading about it after the patient has gone.

If you build local-business tools for a living, outpatient rehab is an underworked vertical with hard numbers, a buyer who can measure the result, and a problem that recurs every single week. Grab the full blueprint here, load a clinic's real payer list into the seed data, and run the demo on their own HMO. That's usually the whole sales call.