Surveys of occupational therapists consistently put documentation at somewhere between a quarter and a third of total working hours. That's more time than most therapists spend in direct patient contact each week — spent writing down what already happened, in a format built mainly to satisfy compliance rather than to help the next decision.
This isn't a workflow problem that a better template solves. It's a structural one: documentation, as most systems are built, is backward-looking by design. A note describes a session that's over. It gets filed, and unless someone deliberately goes looking, it rarely gets read again until the next audit or the next handover.
Two different jobs, one form
Clinical notes are actually being asked to do two unrelated jobs at once. The first is legal and administrative: a defensible record that a service was delivered, to a standard, by a qualified person. The second is clinical: a living account of a patient's trajectory that should inform every future decision about their care. Most documentation systems optimise hard for the first job and treat the second as an accidental byproduct — which is exactly backwards from a therapist's actual priorities, and it's a large part of why documentation feels like a tax rather than a tool.
What changes when notes become data instead of text
The shift underway isn't about AI writing notes for therapists, though that's the part that gets the most attention. It's about what happens once notes — and the structured data increasingly sitting alongside them — are treated as a queryable record instead of a filed document.
- Evidence retrieval instead of memory. Instead of a therapist trying to recall which assessment tool fits a specific presentation, an AI layer can surface the relevant gold-standard measure — Barthel, Lawton IADL, COPM — mapped to the framework the therapist already uses, in the moment it's needed.
- Pattern surfacing across a caseload. A therapist managing forty patients cannot manually notice that six of them are showing the same early pattern of decline. Structured data can.
- Documentation that writes itself from what actually happened. When sensing data and session notes share a record, a draft summary can be generated from the objective activity pattern, with the therapist editing and signing off rather than starting from a blank page.
The goal isn't less documentation. It's documentation that pays the therapist back, instead of only ever being asked of them.
The PEO framework as the connective layer
None of this works if the AI layer speaks a different language than the therapist. That's why frameworks like PEO — Person, Environment, Occupation — matter as more than a clinical model: they're a schema. When a sensing system tags "bathing started 40 minutes late" as a Person-Environment-Occupation event rather than a raw timestamp, it becomes something a therapist can reason about in their own clinical language, not a data export they have to translate themselves.
The therapist's judgment doesn't get automated away in any of this. What gets automated is the retrieval, the pattern-matching across weeks of data, and the first draft of the paperwork — the parts of the job that were never really the clinical work in the first place. What's left is more time actually practising occupational therapy, which is presumably why most people went into the field to begin with.