Healthcare AI agent · Patient access
Patient-Intake & Triage-Support AI Agent
Collect registration, coverage and symptom detail, retrieve the prior record and hand the nurse a structured intake — with identity checks, red-flag escalation, audit trails and clinician-assigned acuity.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Acuity decision
- Clinician only
- After launch
- Agent Care
What this agent does
Does the intake, not the triage
In
- 01
Take registration, demographics, coverage and reason-for-visit detail from the portal, phone line, kiosk or schedule.
- 02
Retrieve the existing chart — problem list, medications, allergies and prior encounters — once identity is verified.
Reason
- 03
Normalise what the patient wrote into coded problems, medications and allergies, keeping the original words beside it.
- 04
Work the client's approved protocol question set, branching on the answers already given rather than asking everything.
- 05
Check the payer eligibility response against the coverage already on file and mark where the two disagree.
Decide
- 06
Detect a stated red flag, stop the question set and hand the session to a person immediately.
- 07
Detect thin, contradictory or low-confidence capture and route it to the intake queue for a person to complete.
Out
- 08
Present a structured intake — the patient's own words, the coded capture, the retrieved record and what is still missing.
- 09
Retain the transcript, the sources read, the protocol path taken and every correction a reviewer makes.
Product statement
The agent gathers, structures, retrieves and presents. Acuity, care setting and any clinical instruction to the patient stay with a qualified clinician.
Example workflow
One intake, end to end
- 1Intake openedPatient portal, phone line, kiosk, secure message or a scheduled-visit prompt
- 2Identity verifiedChecked against the agreed fields before any record is retrieved or read back
- 3Reason and symptoms capturedThe patient's own words first, then the approved question set in the branch those words open
- 4Record and coverage retrievedProblem list, medications, allergies, prior encounters and the payer eligibility response
No human action required
Stages 1 to 4 run without a person in the loop — capture and retrieval finish before anyone is asked to read anything. A stated red flag ends that stretch on the spot.
Reaches the nurse's queue ready to read.
Goes to a person immediately, ahead of the queue.
Nurse or clinician
Reads the patient's own words, the coded capture and the retrieved record, then assigns acuity and the care setting.
Triage · Correct capture · Call the patient
Triaged — handed back
- 6Intake filed and queuedWritten to the encounter only where write access and policy allow; the acuity field stays empty
- 7Outcome evaluatedCapture completeness, red-flag escalation, agreement with the nurse's decision and corrections by cohort
Corrections
Capture that the nurse rewrites is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Assigning an acuity level or triage score.
- Choosing the care setting or level of care.
- Telling a patient what to do clinically.
- Deciding a red-flag session can wait in a queue.
Automation boundary
Agent acts unaided
- ✓Capture registration, coverage and reason-for-visit detail from the patient.
- ✓Work the approved protocol question set and record the answers verbatim.
- ✓Retrieve the chart, problem list, medications and allergies once identity is verified.
- ✓Escalate a stated red flag to a person and route thin capture to the queue.
Write actions run only inside the approval boundaries agreed during implementation. The acuity field is not one of them.
- Confirming a patient match on weak identifiers.
- Merging, unmerging or amending an existing chart.
- Recording a coverage or eligibility decision as final.
- Changing protocol question sets or red-flag rules.
Example output
One intake session, annotated
Everything the agent presents is attached to the session and the record it came from.
Intake output · single pre-visit session
Illustrative example- Patient's words
- Chest tightness since this morning
- Channel
- Pre-visit portal, English
- Identity
- Verified, 3 fields
- Handed to
- Nurse — now, not queued
- Confidence
- 91%
- Acuity
- Not assigned by the agent
As received
The patient's own words and the channel they arrived on, kept verbatim beside the coded capture — nothing on this side is rewritten.
Evidence used
- Three-field identity match
- Problem list and medications
- Chest-symptom branch
Why it went straight to a person
The wording matched the client's red-flag list, so the agent stopped the question set and paged a nurse.
What the score decides
Confidence decides how much the nurse should re-ask, not how urgent the patient is. A clinician assigns acuity.
Action
TriageCorrect captureCall the patientValue
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value lands
Value 01 – 04
Every arriving patient
Portal, phone, kiosk or scheduled visit
Apply the client's own protocol
Use the approved question sets, red-flag list, identity rules and the record already on file.
Take the keying off the front desk
Registration, coverage and history arrive already structured, so the nurse starts from a complete summary rather than a blank form.
Put a person in front of the patient sooner
Stated red flags and thin capture reach a nurse immediately instead of waiting their turn behind routine intake.
Build an evidence trail
Retain the patient's words, the coded capture, the sources read, the protocol path, the confidence and the nurse's correction — on both paths.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
EHR & registration
- Epic
- Oracle Health
- athenahealth
- MEDITECH
- registration APIs
Eligibility & coverage
- Payer eligibility
- clearinghouse
- Coverage discovery
- card capture
Patient channels
- Patient portal
- secure messaging
- SMS
- Telephony and IVR
- interpreter services
Agent
Patient intake & triage support
- Verifies identity
- Structures capture
- Escalates to a person
Prior records
- HIE query
- record exchange
- C-CDA import
- document repositories
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six layers between the model and the patient
Each control wraps the one inside it. An intake clears every layer before it reaches the chart, and the acuity decision sits outside all six.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Decide whether intake may proceed
- L3
- Decides what may be retrieved or read back
- L4 – L5
- Keep acuity with a person and the trail intact
- L6
- Pulls automation back when signals degrade
How Nestack evaluates it
Evaluate the whole intake — not only the summary the nurse reads.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the intake summary the nurse opens
- E1Final-output evaluationWas the capture complete and faithful to what the patient said?
- E2Step-level evaluationDid it code the symptom to the right concept and take the right protocol branch?
- E3Tool evaluationDid it read and write the correct patient, chart and coverage record?
- E4Red-flag recallWas every stated red flag escalated to a person, and how quickly?
- E5Slice evaluationHow does capture quality change across specific patient cohorts?
- E6Business outcomeHow often did the nurse's decision match the routing, and how much was re-asked?
Floor — the triage decision a clinician makes
Failure modes
Where each failure originates in the agent
Seven failure modes plotted against the five stages of the agent lifecycle.
Agent lifecycle
Direction of processing →
01 · Identity / retrieval
2 modes
- PT-01
Wrong or duplicate chart
A common name and birth date match another chart.
- PT-02
Stale external record
A discontinued drug arrives from an outdated external record.
Stage matches
The patient, the chart, the coverage and prior records
02 · Intake capture
1 mode
- PT-03
Non-English capture thins
Detail is lost and the summary reads thin.
Stage asks
The patient's own words and the approved question set
03 · Structuring
2 modes
- PT-04
Red flag not surfaced
A stated warning symptom does not escalate.
- PT-05
Wrong symptom code
A neighbouring code opens the wrong question set.
Stage codes
Problems, medications and capture confidence
04 · Handoff / write
1 mode
- PT-06
Over-escalation floods review
So much escalates that the queue stops being read.
Stage presents
The summary the nurse reads and the queue it enters
05 · Change / Version
1 mode
- PT-07
Silent protocol regression
A question-set change moves the escalation line unnoticed.
Stage tracks
Model, prompt, question-set and red-flag changes
- Sev-1 · a person should have been called
- Sev-2 · wrong information reaches the chart
- Sev-3 · capture degrades, more is re-asked
Affected slices
Capture quality is not spread evenly across patients
Aggregate capture can look acceptable while a small number of patient cohorts carry most of the re-asking, most of the nurse corrections and most of the escalation risk. Nestack reports performance by slice, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Non-English intake sessions | 5.4% | 3.4× | Review | |
| Older adults, atypical presentation | 4.3% | 2.7× | Review | |
| First-time patients, no record | 3.0% | 1.9× | Watch | |
| Routine scheduled-visit intake | 1.0% | 0.6× | Normal |
Bar: lift vs. scheduled-visit baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
Every miss becomes a question the next patient is asked
A missed escalation is not closed when it is fixed for one patient. It changes the question set, the red-flag list or the retrieval rule.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Red-flag recall, re-asking or nurse corrections move in a cohort.
- 02
Diagnose
Traced to identity match, retrieval, symptom coding, the branch taken or the escalation rule.
- 03
Improve
The question set, red-flag list or retrieval rule is re-approved clinically and version-linked.
- 04
Verify
Re-run against held-out sessions from the affected cohort, including the ones that failed.
- 05
Learn
That session becomes a regression case and the changed rule enters the intake runbook.
Learn → Detect
The return edge. A change to the red-flag list is a clinical change — it is signed off before it ships, not after.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, capture, retrieval and coding, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Workflow discovery and automation-boundary definition.Week 1
- 02EHR, scheduling and eligibility API assessment.Week 1, Week 2
- 03Protocol question sets and red-flag list, clinically approved.Week 1, Week 2
- 04Identity verification and consent rules.Week 2
- 05Intake capture on the agreed channels.Week 2, Week 3
- 06Symptom normalisation and coding.Week 3
- 07Prior-record and coverage retrieval.Week 3, Week 4
- 08Escalation routing and nurse handoff workflow.Week 4
- 09Red-flag recall and subgroup evaluation.Week 4, Week 5
- 10Language and accessibility testing.Week 5
- 11EHR write-back and triage-queue integration.Week 5, Week 6
- 12Observability, deployment and Agent Care handover.Week 6
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel
Final scope and sequence confirmed in discovery
Engagement tiers
What each tier includes
Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.
| Capability✓ in scope · — not at this tier | PilotOne intake channel | ProductionProduction EHR integration | AdvancedMulti-site / multi-channel |
|---|---|---|---|
| Introduced at Pilot | |||
| Structured intake capture | ✓ | ✓ | ✓ |
| Identity verification and audit logging | ✓ | ✓ | ✓ |
| Red-flag escalation to a person | ✓ | ✓ | ✓ |
| Acuity assigned by a clinician | ✓ | ✓ | ✓ |
| Baseline evaluation | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Client protocol question sets | — | ✓ | ✓ |
| Prior-record and coverage retrieval | — | ✓ | ✓ |
| Nurse handoff and EHR write-back | — | ✓ | ✓ |
| Observability and evaluation | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-language and accessibility capture | — | — | ✓ |
| Multi-site and enterprise controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on intake channels, EHR and eligibility integrations, patient volume, protocol complexity, approval controls and deployment requirements.
- Separate from build
- Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.
What we need from you
What you bring, and what we build with it
Each input maps to a piece of build scope and a week in the delivery timeline.
You bring
We build with it
01Your approved triage protocol and red-flag list
→Protocol question sets and red-flag list, clinically approved
Week 102Registration, demographic and coverage field requirements
→Intake capture and eligibility field mapping
Week 203Access to EHR, scheduling and eligibility APIs
→Epic and athenahealth registration, schedule and eligibility feeds
Week 204Identity-verification, consent and minimum-necessary rules
→Identity gate, disclosure scope and audit logging
Week 205Real intakes that went wrong, including the near misses
→Evaluation suite, regression cases and failure-mode testing
Week 406The languages, reading levels and channels your patients use
→Language, accessibility and low-literacy capture testing
Week 507Named triage nurses and clinical reviewers
→Nurse handoff workflow, then pilot sessions and production validation
Weeks 5–6
Nothing else is required
Deployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
Phases are drawn over the weeks they actually occupy. Week 5 carries both the subgroup evaluation and the first supervised intake sessions.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Triage protocol and red-flag list approved by your clinicians
W2Identity, consent and capture on the first intake channel
W3Symptom coding, record retrieval and escalation routing
W4Evaluation suite, red-flag recall testing and subgroup slices
W5EHR integration, supervised intake sessions and targeted corrections
W6Nurses triage from live intake sessions, then Agent Care starts
- Reading the band
- Red-flag recall is tested in week 4, before any supervised session runs in week 5. Nothing reaches a patient until it has.
- At the end of W6
- Intake has run alongside the existing process and been read by the nurses who triage from it, then Agent Care takes over monitoring.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Healthcare AI agent
Build an intake agent around your triage process.
Show us your intake channels, your approved protocol and red-flag list, and how a patient reaches a nurse today. We'll run one intake against your own protocol and test the red-flag list on your own near misses.
- Typical delivery
- 4–6 weeks
- Deployment
- Your EHR and access stack
- Acuity decision
- Nurse or clinician, always
- After launch
- Agent Care available
Nestack Agents · Patient intake & triage support
AGT-HC-01 · Agent Care available after launch