CLINICAL DOCUMENTATION · A MULTI-COUNTRY HOSPITAL CHAIN
Clinical Scribe
The doctor talks to the patient; the agent writes the note, on-premise, HIPAA/GDPR, clinician-signed.
- Private
- Operationally cost-effective at scale
- Deterministic & controlled
- Subjective
- Dry cough ×10 days, no fever
- Objective
- Chest clear, SpO2 98%
- Assessment
- Acute bronchitis
- Plan
- Supportive care, review in 7d
- ICD-10
- J20.9 · checked
Illustrative example. The clinician reviews and signs every note; audio is processed on-premise and deleted after sync.
Illustrative
Executive summary
Clinical Scribe turns a normal doctor-patient consultation into a finished, coded clinical note while the visit is still happening. It listens in the room, writes the SOAP note in real time with the ICD-10 codes already filled in, and posts it into the medical record the moment the clinician approves it. The outcome: doctors get their evenings back, notes are fuller and more consistent because they are written while the visit is fresh, and billing matches what actually happened so claims go through cleanly. It runs entirely inside your own network, the audio never leaves the clinic, and every note is reviewed by a clinician before it is signed. It is in production today at a multi-country hospital chain.
Understanding the problem: what is a SOAP note?
A SOAP note is the standard record a clinician writes after seeing a patient. It is not a free-form diary entry; it is a structured document with four fixed parts, and every part has a job. It is what makes one patient visit legible to the next clinician, to the biller, to the insurer, and to an auditor years later. The acronym is the structure:
- S, Subjective. What the patient says, in their own words. The reason for the visit, the symptoms, when they started, what makes them better or worse, the relevant past history, current medications, and allergies. This is the patient's side of the story.
- O, Objective. What the clinician observes and measures. Vital signs (blood pressure, temperature, heart rate, oxygen saturation), the physical examination findings, and any test or scan results to hand. This is the measurable, third-party side.
- A, Assessment. The clinician's judgement: the diagnosis, or the working differential when it is not yet certain. Crucially, this is where the ICD-10 codes live, the internationally standard codes that turn a written diagnosis ("type 2 diabetes with diabetic neuropathy") into a machine-readable code (E11.40) that insurers, billing systems, and health authorities rely on.
- P, Plan. What happens next. The treatment, the prescription (drug, dose, duration), the tests ordered, the referrals, the patient instructions, and the follow-up date.
Here is how that note actually gets created today. A patient walks in. The clinician greets them, asks why they came, listens, examines, thinks, decides, prescribes, and explains the plan, usually inside a ten- to twenty-minute slot. During that window the clinician is doing two jobs at once: practising medicine, and trying to remember enough of it to write it down later. The data the visit produces is dense and easy to lose:
- The exact words the patient used to describe the symptom, and when it started.
- The history the patient volunteered, past episodes, family history, what they tried already.
- The vitals and examination findings, often called out aloud and not written down in the moment.
- The diagnosis reached, and the reasoning behind it.
- The precise drug, dose, and duration prescribed.
- The tests ordered and the follow-up agreed.
- The correct ICD-10 code (and there are tens of thousands of them) for every condition addressed.
Then the visit ends, the next patient is already waiting, and the note doesn't get written. It gets deferred. Most clinicians type their notes from memory hours later, between patients, at lunch, or after the clinic closes. That gap between the visit and the writing is where the damage is done. Doing this by hand is slow, costly, and error-prone in ways that compound:
- Doctors lose hours of every day to writing notes instead of seeing patients. The documentation burden is real clinical time, spent typing rather than treating.
- Notes written from memory at the end of the day miss details. A symptom's exact onset, a number from the examination, an instruction given to the patient, they fade. The record ends up thinner than the visit was.
- Missed or under-entered ICD-10 codes mean billing quietly leaks. If a condition addressed in the room never makes it into the Assessment as a coded diagnosis, the work was done but never billed. At clinic scale, that is real revenue lost to forgetfulness, not to policy.
- Paperwork is the number-one reason good clinicians leave the profession. Documentation burden is one of the most cited drivers of clinician burnout. The note is supposed to serve the patient; instead it costs the clinic the clinician.
The chart is the proof, the proof is incomplete, and the doctor is exhausted. The note is simultaneously the legal record of care, the basis for the bill, and the thing an audit will check first, and it is being reconstructed from tired memory after the fact.
SSubjective
What the patient says, in their own words: the symptoms, when they started, the relevant history, current medications and allergies.
OObjective
What the clinician observes and measures: vital signs, examination findings, and any test or scan results to hand.
AAssessment
The clinician's judgement: the diagnosis or working differential, and where the standard ICD-10 codes live.
PPlan
What happens next: the treatment, the prescription, the tests ordered, the referrals, and the follow-up date.
What Clinical Scribe does
Clinical Scribe sits in the room (or on the call) for the consultation and produces the finished SOAP note as the visit happens, rather than waiting for the clinician to reconstruct it afterwards. Across a full schedule, dozens of consultations per clinician per day, every working day, it converts each spoken conversation into a structured, coded clinical record that is ready for the chart the moment the clinician signs it. It is not a dictation tool that hands back a transcript; it produces the actual document the workflow needs. From a single consultation it surfaces and decides:
- Which parts of the conversation belong in Subjective (the patient's account) versus Objective (the clinician's observations and measurements), keeping the two halves cleanly separated.
- The structured SOAP fields themselves, filled in rather than left as loose paragraphs.
- The ICD-10 codes for the conditions addressed, suggested and validated against the standard code set so the Assessment is coded at the moment of writing, not retro-fitted by a coder days later.
- The specialty shape of the note, because a paediatrics note reads differently from a cardiology note, using templates tuned per department.
- A confidence label on every part of the draft, so a section the agent is unsure about is surfaced for the clinician rather than slipped silently into the chart.
- A source link from every line of the note back to the moment in the conversation it came from, so nothing in the record is unsupported.
- The patient's prior history, on demand, answered in plain English and grounded in the record.
Illustrative. The agent fills the set fields, validates every code against the ICD-10 set, and runs entirely inside the clinic, but nothing enters the medical record until a clinician reads, adjusts, and signs.
Questions it can answer
Beyond writing the note, Clinical Scribe lets a clinician query the patient's record in plain English and get an answer grounded in what is actually on file, and lets clinics adapt it to how their own consultations work. Grounded in its real Q&A examples and the variations teams ask for:
- "What did we prescribe last visit?"
- "Has this patient had this symptom before?"
- "What were the results of the tests we ordered last time?"
- "When is this patient's follow-up due, and for what?"
- "Our consultations mix Hindi and English, can the note still come out clean?"
- "Can it produce the note in our own EMR template, with ICD codes suggested?"
- "Can it also draft the patient-friendly summary we send after the visit?"
How it works
It listens in the room, and the audio never leaves the clinic
It captures the normal conversation between doctor and patient, in any of the languages your clinics use, and processes it inside your own network. The recording is deleted afterwards. It is HIPAA-compliant for the US and GDPR-compliant for Europe.
It writes the SOAP note in real time
The ICD-10 codes are already filled in, with templates tuned to each specialty, because paediatrics reads differently from cardiology.
One click sends the note into the medical record
The moment the doctor approves it. The doctor stays in charge: they read, adjust, and sign.
You can ask a patient's history in plain English
Like "what did we prescribe last visit?" or "has this patient had this symptom before?" Every answer is grounded in the record.
Under the hood (for your technical team)
Voice Transcription
Turns speech into accurate text, live or from a recording.
Conversation-to-Notes
Turns a conversation into a structured record, like a SOAP clinical note.
Rule & Tolerance Checks
Business rules that gate every answer before it is final.
System Posting & Actions
Writes results into your systems and takes the next step, whether that's to book, reschedule, route, or draft the reply.
Source Citation & Audit Trail
Links every answer to its exact source, with a tamper-evident log.
Sovereign Deployment
Runs on your own servers or private cloud; your data never leaves.
The building blocks it is composed from. Clinical Scribe is assembled from proven capability-agents, each doing one job:
- Voice Transcription, turns speech into accurate text, live or from a recording. This is the front door: it captures the consultation as it happens.
- Speaker Identification, separates who said what, keeping doctor and patient cleanly attributed so the "Subjective" (what the patient said) and "Objective" (what the doctor observed) halves of the note are not muddled.
- Conversation-to-Notes, turns the conversation into a structured record in the SOAP shape, rather than a loose transcript. This is what makes the output a clinical note and not just a wall of text.
- Rule & Tolerance Checks (ICD-10), business rules that gate the note before it is final, validating diagnosis codes against the ICD-10 set and flagging anything that does not resolve.
- Confidence Scoring, labels every part of the draft as solid or shaky, so a low-confidence section is surfaced to the clinician rather than slipped silently into the chart.
- System Posting & Actions, writes the approved note into your record system and takes the next step, with no rip-and-replace of the systems you already run.
- Source Citation & Audit Trail, links every line of the note back to the moment in the conversation it came from, with a tamper-evident log of who reviewed and signed.
The defined recipe for this agent combines Voice Transcription, Speaker Identification, Conversation-to-Notes, Confidence Scoring, System Posting & Actions, and Source Citation & Audit Trail: it captures the visit, separates speakers, turns the conversation into a clinical note, scores confidence, and posts after review. One visit in, one finished, reviewable note out.
Inputs and modalities. The primary input is live audio of a spoken consultation, handled across the languages your clinics use (the production deployment spans a multi-country hospital chain, and the agent supports mixed-language consultations such as Hindi and English in one visit). It also works from a recording rather than only live. Output is a structured SOAP note with coded fields rather than free text, with specialty templates shaping the note per department. The same engine can drive an optional second output, such as a patient-friendly summary drafted after the visit.
What it integrates with. It posts finished notes into your existing medical record / EMR through System Posting & Actions, and the same Connector Framework approach that lets Attentions agents plug into the tools you already run (EMR, practice software, databases) applies here, with new sources added in days rather than quarters. The plain-English history lookup queries your record with permissions intact, so a clinician sees only the patients they are entitled to see.
Data-flow and deployment topology. Audio is captured and transcribed inside the clinic's own network; transcription, speaker separation, and note generation all happen on the customer's infrastructure, so the conversation never leaves the building. The recording is deleted after the note is synced. Only the clinician-approved, structured note is written into the medical record, with its citation trail attached. At the highest tier the model itself runs on dedicated hardware inside your walls, so nothing, not the audio, not the transcript, not the note, ever leaves your control.
Built for production
Private
Audio is captured and processed inside the clinic and deleted after sync; HIPAA- and GDPR-compliant.
Operationally cost-effective at scale
Sized to a full clinic schedule, with cost ~flat per day instead of per-minute or per-token billing.
Deterministic & controlled
Structured SOAP fields and ICD-10 rule checks, and a clinician reviews and signs every note.
Private
Patient data is the most sensitive data there is, so sovereignty is not a feature here, it is the whole premise. Clinical Scribe runs inside your own network: the audio is captured, transcribed, and turned into a note on your own infrastructure, and the recording is deleted after the note is synced. At the highest tier the model itself runs on dedicated hardware inside your walls, so nothing, not the audio, not the transcript, not the note, ever leaves. Sovereign Deployment is the block that enforces this: agents run on your own servers or private cloud and your data never leaves. Access & Permission Inheritance means the plain-English history lookup respects who is allowed to see which patient; if a clinician cannot see a record in the source system, they cannot see it through the agent (RBAC and SSO). Source Citation & Audit Trail logs every note, review, and sign-off in a tamper-evident record. This is what lets the deployment stand up to HIPAA in the US and GDPR in Europe, because the controls are built in, not bolted on.
Operationally cost-effective at scale
A full clinic schedule is a high-volume workflow: dozens of consultations per doctor per day, every working day. Billed per token against a giant general-purpose model, that cost scales with the volume of visits, and the busier your clinics get, the worse the maths. Clinical Scribe instead runs a small, job-specific model fine-tuned for one task, turning a clinical conversation into a coded note, on your own infrastructure. A model that only has to do this needs far less computing power than one that also has to write poetry, which is why a focused agent runs roughly ten times cheaper in production than the general-purpose approach, with the running cost staying nearly flat as visit volume grows. The hardware is sized to the workload, not to the brochure, and the running cost is watched from the first day in production and managed as volumes grow. FinOps for AI, handled for you.
Deterministic and controlled
In a clinical record, a confident wrong answer is worse than no answer. Three mechanisms keep Clinical Scribe honest. First, structured fields, not free text: Conversation-to-Notes fills the set SOAP fields rather than writing loose paragraphs, which leaves far less room for mistakes to hide and makes the result easy to check. Second, Rule & Tolerance Checks gate the note before it is final, validating ICD-10 codes against the standard set so a code either resolves or gets flagged. Third, Source Citation ties every line back to the moment in the conversation it came from, so there is no claim in the note without something behind it. Confidence Scoring routes anything shaky to the clinician rather than guessing, and the clinician reads, adjusts, and signs every note before it enters the record. The agent never writes to the chart on its own. The same conversation produces the same structured note, and a human always has the last word.
Who benefits
Doctors
The doctor is the person this is built for. They keep doing exactly what they have always done, talk to the patient, examine, decide, but the after-hours typing comes off their plate. The note is drafted live and waiting for review the moment the visit ends, so the chart is closed before the next patient walks in instead of at 9pm. The note is fuller too, because it captures what was actually said and done rather than what could be recalled hours later. Less typing, more patients, and the documentation backlog that follows them home simply stops following them home.
Clinic admin and the CFO
For the people who run the clinic as a business, the win is twofold. Billing is captured properly: every condition addressed in the room is coded into the Assessment at the moment of writing, so the work that was done is the work that gets billed, and the quiet leak from missed or under-entered ICD-10 codes closes. And because the documentation load no longer caps how many patients a clinician can see in a day, the schedule can carry more appointments per doctor without adding clinicians. On top of that, paperwork stops being the thing that drives good clinicians out, which protects the most expensive asset the clinic has.
Coding and compliance
Coding teams normally inherit notes after the fact and reconstruct the codes from a thin write-up, days late. Here, every note is coded at the moment of writing, with the ICD-10 codes validated against the standard set as the note is produced, and anything that does not resolve is flagged rather than guessed. Compliance gets a full, tamper-evident audit trail: who saw the note, what was changed, and who signed it, with every line traceable back to the moment in the conversation it came from. When an auditor asks for the basis of a coded diagnosis, the answer is one citation away.
IT and compliance (security)
For the team accountable for patient-data security, the deployment topology is the whole point. The audio never leaves the clinic, capture, transcription, speaker separation, and note generation all run on the clinic's own infrastructure, and the recording is deleted after the note is synced. Only the clinician-approved structured note is written into the record. Access follows the source system's own permissions through RBAC and SSO, and the standards their regulators care about (HIPAA in the US, GDPR in Europe) are designed in rather than bolted on. There is no rip-and-replace of the record system to certify, because the agent posts into the EMR they already run.
Patients
The patient gets the most human thing back: the doctor's attention. Instead of watching their clinician type into a screen and split their focus between the conversation and the keyboard, they get a clinician who looks at them and listens. Nothing about the visit changes for the patient except that the person treating them is fully present.
In short
“In simple terms, the doctor talks to the patient and the agent writes the note, finished, coded, and ready for one click into the record, so the only thing that disappears is the after-hours typing.”
Core business value
Clinical Scribe transforms disconnected consultations, after-hours typing, half-remembered details, and retro-fitted billing codes into a single, structured, coded clinical record produced at the point of care. It helps organisations:
- Return clinical hours to clinicians by removing the documentation burden.
- Produce fuller, more consistent notes, because they are written while the visit is fresh.
- Capture billing accurately by coding every addressed condition at the moment of writing.
- Reduce clinician burnout, the leading cause of attrition, by taking away the paperwork.
- See more patients per clinician without adding headcount, because documentation no longer caps the schedule.
- Keep patient data sovereign, audio never leaves the building, and the recording is deleted after sync.
- Stay audit-ready, with every line of every note traceable to the conversation behind it and every review and sign-off logged.
- Keep the clinician in charge, because every note is read, adjusted, and signed by a human before it enters the chart.
In simple terms, the doctor talks to the patient and the agent writes the note, finished, coded, and ready for one click into the record, so the only thing that disappears is the after-hours typing.
The return (illustrative)
The return on Clinical Scribe stacks four sources, applied to this workflow:
Hours returned
the after-hours typing moves off the doctor's plate. If a clinician currently spends 1.5 to 2 hours a day writing notes (illustrative), most of that comes back as either patient capacity or time at home.
Error cost avoided
notes written live, with ICD-10 codes validated at the point of writing, capture billing that today leaks through missed or under-entered codes; recovering even 3 to 8 percent of coded revenue that was previously dropped (illustrative) is meaningful at clinic scale.
Speed
the chart is closed before the next patient walks in instead of at 9pm, so documentation stops being a backlog and claims submit cleanly.
Scale without headcount
more appointments per doctor without adding clinicians, because the documentation load no longer caps how many patients each doctor can see in a day.
Documentation time per visit
Illustrative.
Why teams adopt it
Nothing about the visit changes for the patient or the doctor. The doctor still talks to the patient the way they always have. The only thing that disappears is the after-hours typing, which is exactly the thing everyone wants gone. There is no rip-and-replace of your record system and nothing new for clinicians to learn, which is why people actually want to use it.
Start with an assessment.
We scope the right first workflow on your own data and give you an honest go or no-go before you commit to anything bigger.