Skip to content
Zowork
All case studies
Flagship case studyBehavioral health AI

AI that writes the note, so clinicians can stay present.

Behavioral health clinicians can lose up to two days a week to paperwork. We built AI-assisted clinical documentation that listens to the visit, drafts a structured, evidence-backed note in near real time, and hands the clinician a draft to review and sign, built on AWS HealthScribe and Amazon Bedrock.

AI documentationAWS HealthScribeAmazon BedrockBehavioral HealthHIPAA-eligible
Listening
Passive capture
Draft note
ready to review
  • GoalImprove sleep and daily activation.
  • ProgressDifficulty sleeping; low motivation, ~2 weeks.
  • PlanSleep-hygiene routine; follow up in 2 weeks.
HIPAA-eligible · clinician signs off
Overview

The note should be a by-product of the conversation, not a second job

In behavioral health, the most valuable thing a clinician has is attention, and documentation quietly takes it away. Notes get written between sessions, after hours, or at home; many are still unfinished at the end of the day. The knowledge was never the bottleneck. The time and friction between a conversation and a finished, trustworthy note was.

So we built a system where the note is a by-product of the visit. It listens in the background, turns speech into a speaker-aware transcript, and drafts a structured clinical note grounded in what was actually said, then hands it to the clinician to review and sign. No headset, no dictation, no “start recording” ritual. The clinician stays in the room; the paperwork drafts itself.

Watch it work

See it in production

The same approach, running in a national behavioral health platform that serves providers caring for tens of millions of people.

In production: clinical documentation burden reduced with generative AI on AWS
The results

What changes when the note drafts itself

Reported outcomes from the AI-assisted documentation approach in a production behavioral-health deployment on AWS.

60%

Less time on documentation

68%

Faster session-to-sign

99.9%

Reimbursement success rate

30%

Higher monthly provider revenue

Source: reported production outcomes, behavioral-health providers on AWS.

The burden

Up to two days a week, lost to notes

Documentation is one of the biggest drivers of clinician burnout, and it lands hardest exactly where demand for care is highest.

~40% of the week

Clinicians can spend up to 40% of their time, nearly two full days, on documentation instead of care.

Burnout, not capacity

The administrative load drives burnout at the very moment demand for behavioral health is at record highs.

Tens of millions of lives

Behavioral health providers serve a vast population, and every hour saved compounds across the whole system.

The goal
If we can take that two days of documentation time down to two hours, then we succeed in giving that time back to providers and the clients they serve.
How it works

From the room to a reviewable note, in five steps

A pipeline that turns a live conversation into a structured draft, each stage doing one job well, with the clinician at the end.

Step 1 / 5Passive capture

The visit is captured from the room, no headset, no “start dictation.”

  • Streams securely from the clinician’s device in real time.
  • No ritual. The clinician just talks to their patient.
  • The same low-latency capture pipeline behind our AI-assisted scribe.

Tap a step to explore

The principle
The AI drafts; the clinician decides. Every stage is built to defer, never to overstep. The system surfaces evidence and confidence so a human can stay accountable for the chart.
Trust & evidence

A draft you can actually trust

A note clinicians won’t use is worse than no note at all. The whole design earns trust by showing its work: every line traces back to the transcript.

Draft note · GIRPP
evidence-mapped
Source transcript
  • Patient citedHonestly, I just haven't been able to sleep. I'm up till 3 most nights.
  • Patient citedAnd I can't get myself motivated to do anything. It's been about two weeks now.
  • ClinicianLet's try a wind-down routine before bed, and a short walk each morning to start.
  • ClinicianWe'll keep you on the same medication for now and check back in two weeks.

Select a note line to see the turns it was drawn from

Evidence-mapped

Every sentence links to the transcript turns it was drawn from, one click to verify.

Speaker-aware

The transcript knows clinician from patient, turn by turn, with word-level timestamps.

Structured terms

Conditions, medications and interventions are extracted from the dialogue, not guessed.

GIRPP, not generic

A goal-centric behavioral-health note, not a general-medicine SOAP template bolted on.

Human-in-the-loop

Nothing auto-files. The clinician reviews, edits and signs every note.

HIPAA-eligible

Encrypted in transit and at rest; patient data never trains the models.

Architecture

Managed AWS AI services, with the hard engineering around them

The models are the easy part to name and the hard part to operationalize. The work is everything around them: capture that survives real clinics, isolation that satisfies compliance, and a review surface clinicians actually trust.

AWS HealthScribe

Speaker-aware transcription plus clinical-note generation with built-in evidence mapping. HIPAA-eligible.

Amazon Bedrock

Frontier LLMs read the transcript and assemble the GIRPP note, grounded only in what was said.

Amazon Chime SDK

Secure, real-time passive capture from the clinician’s device, with no “start recording” ritual.

The engineering around the models
  • A low-latency passive-capture pipeline that survives the dead zones every clinic has, the same streaming foundation we detail in our conversation-capture deep-dive.
  • Tenant-isolated, encrypted storage with audit logging on every view and approval of a note.
  • A review surface that keeps evidence one tap away and keeps the clinician in control of the final note.
  • Write-back into the EHR, so the signed note lands where care actually happens, not in a separate tool.
Time reclaimed

Give the two days back

The point of all of it: shrink the documentation block in a clinician’s week so the time goes back to patients.

A clinician’s week
Before~2 days on notes
Documentation
Patient care
With AI-assisted documentationcloser to ~2 hours
Patient care

Up to 40% of a clinician’s time goes to documentation. Drafting the note from the conversation gives most of that time back, to patients, or to going home on time.

FAQ

Questions we get asked

What is AI-assisted clinical documentation?

It is AI that captures a patient visit in the background, with no headset or dictation, transcribes it with speaker roles, and drafts a structured clinical note from the conversation. The clinician reviews, edits and signs; the AI never files on its own.

Does the AI file notes into the chart automatically?

No. Every draft is reviewed and signed by the clinician before it reaches the chart. The system is built to defer, never to overstep: the AI drafts, the clinician decides.

How does it avoid inventing clinical details?

Every sentence in the draft is mapped back to the exact transcript turns it was drawn from, with a confidence signal, so anything the model wrote can be verified in one click. It is grounded strictly in what was said in the room.

Is it HIPAA-compliant?

It is built on AWS HealthScribe, a HIPAA-eligible service, with encryption in transit and at rest, tenant isolation, and audit logging. Patient data is not used to train the models.

Conclusion

The conversation does the documenting

AI-assisted documentation reframes the note from a chore into a by-product of good care. By listening to the visit, drafting from the transcript, and mapping every sentence back to its evidence, it behaves less like autopilot and more like a trusted colleague, one that drafts quickly, shows its work, and always leaves the final word to the clinician. The two days a week stop going to paperwork, and start going back to patients.

All case studies