Your OTs Were Trained to Treat, Not to Type

AI will not fix a documentation problem that doesn't have a defined process underneath it. Before any clinic engages with automation, the owner needs to answer a specific set of questions — and be honest about what the answers reveal.

• Do your therapists document in the same EMR consistently, or are there paper notes, personal systems, or hybrid workflows still running? If session data lives in three different places, AI has nothing reliable to pull from.
• Does someone in your practice own billing — meaning they are accountable for it, not just involved in it? Coding automation only helps if there's a person whose job it is to act on the flags it generates.
• Can you articulate what a correct progress note looks like for your primary payer mix? If your therapists would give different answers, the AI will produce inconsistent output and you'll spend more time correcting it than writing notes manually.
• Have you had a payer audit or a pattern of denials in the last 12 months? If yes, this is actually a green flag — it means there's a defined problem with known parameters. If you have no visibility into your denial reasons, fix that first.
• Is your front desk currently handling prior auth requests, or is it falling to the treating therapist? This answer shapes which automation you build first.

The honest disqualifiers: if your clinic has fewer than two full-time therapists, documentation volume likely doesn't justify the build cost yet. If you haven't standardized your note templates across therapists, do that first — AI scales your current process, and a broken process at scale is worse than a broken process contained to one person. If your EMR vendor has locked APIs or won't allow third-party integrations, the technical feasibility conversation needs to happen before anything else.

AI implementation that starts without clear answers to these questions almost always stalls in week three when the process gaps become visible. Better to find them now.

The single highest-impact automation for most occupational therapy clinics is session documentation assistance — not scheduling, not billing lookup, not chatbots. Documentation. This is where therapist time disappears, and it's where the clinical-administrative collision is most painful.

Here's how it actually works in practice. After a session, the therapist opens a structured input form — either inside their EMR or in a connected interface — and answers a short set of prompts: what treatment activities were performed, what functional outcomes were observed, how the patient responded, and what the plan is for next session. These aren't open-ended fields. They're structured, and they're built around the specific CPT codes and functional goal language your primary payers require.

That structured input goes to a language model — in Oaken's stack, this runs through the Claude API with a clinic-specific prompt layer — and within seconds produces a draft progress note. Not a template with blanks. A full draft, written in clinical language, with measurable functional goal references, appropriate treatment justification, and the session minutes documented in a way that supports the CPT codes being billed. The therapist reads it, makes any corrections, and signs it. The note that used to take 25 minutes now takes 4.

What the owner notices on day one: therapists leaving closer to on time. That's it. It sounds small and it isn't — therapist burnout and turnover is a real cost, and overtime documentation is one of the clearest drivers. (Source: American Occupational Therapy Association, 2022 Workforce Survey) The AOTA has consistently documented that OTs cite paperwork burden as one of the top contributors to job dissatisfaction.

What the owner notices in month three: denial rates on documentation-related rejections dropping, progress reports going out on time because the data is actually being captured session by session, and a cleaner audit trail if a payer ever asks for records. The system connects to PostgreSQL for session data storage, integrates with your EMR via API where available, and uses pgvector for retrieving prior session context so notes reflect actual progression, not just today's session in isolation.

The clinic owner who gets real results from AI automation is not necessarily the most tech-forward. They're the most process-forward. They've already standardized their note templates. They have one person accountable for billing outcomes. Their therapists document in the EMR the same day as treatment — not two days later, not on Sundays before Monday's notes are due.

Size matters, but not in the way most owners think. A three-therapist clinic with clean processes will see faster, better results than a ten-therapist clinic where every OT has their own documentation style and the billing is outsourced to a company that doesn't communicate denial patterns back to the practice. Process maturity beats headcount every time.

The staff situation also matters. If you have a billing coordinator who understands OT-specific coding — who knows why 97530 and 97535 can't be billed together in most payer contexts, who tracks authorization limits by insurance — that person becomes dramatically more effective with AI-generated flags and summaries. If you don't have that person, the coding automation produces output that no one acts on correctly.

Clinics that specialize — pediatric OT, hand therapy, neurological rehab — tend to see stronger results than generalist practices, because the documentation and coding patterns are more consistent and the AI prompt layer can be tuned tightly to a specific clinical context. A hand therapy clinic billing primarily 97530 and 97597 with a defined payer mix is a cleaner build than a mixed-population clinic with wildly varied treatment approaches and 12 different payer contracts.

Who isn't ready yet: the solo practitioner who is also the biller, scheduler, and front desk. The practice that recently switched EMRs and hasn't stabilized workflows. The clinic owner who wants AI to solve a staffing problem — if you're short-staffed, automation helps at the margins but it doesn't replace a treating therapist. Be clear-eyed about what the actual problem is before investing in a solution.

(Source: AOTA, 2022 Workforce Survey — OT practitioners report spending an average of 30% of work time on documentation and administrative tasks)

Misconceptions about AI in healthcare settings are common, and in OT specifically, three beliefs keep showing up that lead either to bad implementations or to clinics ignoring tools that would genuinely help them.

Myth 1: "Our EMR already has AI — we're covered."
Most EMR vendors have added AI features in the last two years, and most of those features are documentation templates with smart autocomplete. That's not the same as a system that understands the relationship between your documented treatment activities, your functional goal language, and your CPT code selection. The EMR AI tells you what you told it. A purpose-built documentation assistant understands payer-specific LCD requirements, flags code mismatches before submission, and learns from your practice's specific correction patterns over time. They solve different problems.

Myth 2: "AI can't handle OT documentation because it's too clinical and individualized."
This is the most understandable misconception and also the most limiting. The clinical judgment stays with the therapist — always. What AI handles is the translation of clinical observations into structured, compliant documentation language. Therapists aren't describing their clinical reasoning to the AI and asking it to think for them. They're inputting structured session data and getting back a draft that reflects what they observed, in language that satisfies the payer. The individualization happens in the structured input. The AI handles the formatting, the functional goal framing, and the documentation logic. Those are not clinical tasks. They're administrative ones wearing clinical clothing.

Myth 3: "Our denial rate is acceptable — we don't have a documentation problem."
Denial rate is a lagging indicator. By the time a pattern shows up in your denials, you've already been underpaid, delayed, or flagged for months. The more telling number is how long it takes to appeal a denied claim and what the clinical time cost of that appeal is. According to the Medical Group Management Association, practices lose significant revenue not from outright denials but from underpayments and claims that are simply never corrected after initial denial. (Source: MGMA, Stat Poll on Claim Denials, 2022) If your therapists are writing appeal letters, that's clinical time with a dollar value that never shows up in a denial rate metric.