Healthcare AI agent · Population health
Post-Discharge & Care-Gap Outreach AI Agent
Check in after discharge, confirm the follow-up and work open care gaps against the record — with every worrying reply escalated to a named clinician on a stated clock, and no clinical advice sent.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Clinical replies
- Clinician only
- After launch
- Agent Care
What this agent does
Carries the message, not the judgement
In
- 01
Take discharge summaries, ADT events, problem and medication lists, and open gaps from the EHR and registry.
- 02
Read the consent, contact preference, language, channel and quiet hours already recorded against the patient.
Reason
- 03
Check the patient is still contactable and still eligible — not recorded as deceased, not currently an inpatient.
- 04
Assemble the check-in the discharge plan calls for: the medicines dispensed, the instructions and the follow-up.
- 05
Check each care gap against what the record evidences, so nobody is chased for a test that was done somewhere else.
Decide
- 06
Detect a reply naming a symptom, a medication problem or distress and hand it to a named clinician on the clock.
- 07
Detect a gap it cannot evidence, a contact it cannot verify or a consent withdrawn, and stop rather than send.
Out
- 08
Send only wording a clinician approved, on the recorded channel and in the recorded language, inside the agreed hours.
- 09
Retain the attempt, the channel, the reply as the patient wrote it, the escalation and the clock it was met on.
Product statement
The agent contacts, listens and records inside rules a person set. Assessing a symptom, advising a patient and deciding what happens next stay with a clinician.
Example workflow
One discharge, end to end
- 1Discharge event receivedADT discharge message, discharge summary, medication list and the follow-up already booked
- 2Eligibility checkedDeath indicator, current admission, attribution, consent, language, channel and quiet hours
- 3Check-in sentApproved wording only, on the recorded channel, inside the hours the patient agreed to
- 4Reply read and classifiedSymptom, medication problem, appointment question or distress — matched against the client's escalation wording
No human action required
Stages 1 to 4 run without a person in the loop — eligibility, consent and the wording are settled before anyone is asked to look. A reply that names a clinical concern ends that stretch on the spot, at any hour.
Read by a clinician on the routine clock.
Paged to a named clinician on the stated clock.
Named clinician
Sees the reply as the patient wrote it, the discharge plan it was sent against and the clock it arrived on, then calls, advises or escalates further.
Call the patient · Advise · Escalate further
Handled — handed back
- 6Escalated or recordedThe escalation, the clinician's action and the outcome, written back to the record and the gap list
- 7Reach and outcome evaluatedWho was reached, who replied, how long escalations waited and what closed, by cohort
Never reached
Patients the agent could not reach at all are counted in the evaluation, not dropped.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Telling a patient what a symptom means.
- Advising whether to take, stop or change a medicine.
- Deciding that a reply is not clinically urgent.
- Reassuring a patient who reports something worrying.
Automation boundary
Agent acts unaided
- ✓Check in after discharge against the plan and the medicines a clinician set.
- ✓Confirm the follow-up appointment and chase what has to be in place first.
- ✓Offer the care gaps the record still evidences as open, and book them in.
- ✓Escalate every clinical reply to a named person inside a stated window.
Write actions run only inside the approval boundaries agreed during implementation. Judging a reply is not one of them.
- Closing a care gap the record does not evidence.
- Contacting a patient flagged deceased or readmitted.
- Disclosing clinical detail before identity is verified.
- Changing outreach cadence, wording or escalation rules.
Example output
One check-in reply, annotated
Everything the agent does with a reply is attached to the reply and the discharge plan it was sent against.
Outreach output · single reply
Illustrative example- Check-in sent
- Day 3, 19:40, SMS
- Patient replied
- More short of breath today
- Discharge
- Heart failure
- Escalated to
- Named clinician, paged
- Confidence
- 91%
- Clinical assessment
- Not made by the agent
As received
The message that went out, the hour it went out at and the patient's own words in the language they used.
Evidence used
- Discharge diagnosis
- Words in the reply
- Client escalation list
Why it did not wait for morning
Breathlessness after this discharge is on the client's escalation list, and that list does not stop at 5pm.
What the score decides
Confidence decides which clock the reply lands on, never whether a person sees it. Every reply is read.
Action
Call the patientRecord and closeEscalate furtherValue
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 discharge and open gap
ADT feed, discharge summary, gap list
Check the record before the phone
Death status, current admission, attribution, consent and what the record already evidences are read before anyone is contacted.
Reach the patients a call list never gets to
Check-ins and gap outreach run on the channel each patient agreed to, in their language, without someone working down a printout.
Put every clinical reply in front of a person
A reply naming a symptom, a medication problem or distress reaches a named clinician on a stated clock, including out of hours.
Build an evidence trail
Retain the attempt, the channel, the reply as written, the escalation, the clock it was met on and the clinician's action — 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 & discharge data
- Epic
- Oracle Health
- MEDITECH
- Discharge summary
- medication list
ADT & eligibility feeds
- ADT events
- HIE alerts
- Death indicator
- current-admission check
Population health & measures
- Registry and gap lists
- measure engine
- Care management
- supplemental data
Agent
Post-discharge & care-gap outreach
- Checks eligibility
- Reaches and listens
- Escalates on a clock
Outreach channels
- Outreach CRM
- Telephony
- SMS
- Patient portal
- interpreter line
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 your patients
Each control wraps the one inside it. A message clears every layer before it is sent, and no layer lets the agent answer a clinical question.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Decide whether the patient may be contacted
- L3
- Decides what the message is allowed to say
- L4 – L5
- Put clinical replies with a person, on the record
- L6
- Pulls outreach back when signals degrade
How Nestack evaluates it
Evaluate who was reached — not only how many were messaged.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the message the patient receives
- E1Final-output evaluationWas a reply that named a clinical concern classified as one?
- E2Step-level evaluationDid it read the right discharge plan, medication list and gap?
- E3Tool evaluationDid it contact the right patient, on a number still theirs?
- E4Escalation timingHow long did a clinical reply wait for a named person?
- E5Slice evaluationHow do reach and escalation change across patient cohorts?
- E6Business outcomeWas the concern handled, and did the gap close in the record?
Floor — the concern that reached a clinician in time
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 · Eligibility
1 mode
- PD-01
Outreach after death
A reminder reaches a family after a death.
Stage gathers
Discharge data, ADT status, consent and the gap list
02 · Targeting
2 modes
- PD-02
Gap closed elsewhere
The test was done elsewhere, outside the feed.
- PD-03
Reach skews digital
The largest gaps sit with the least contactable.
Stage selects
Who is contacted, on which channel, in what language
03 · Send / contact
2 modes
- PD-04
Stale number reached
Another person receives the patient's detail.
- PD-05
Advice in the message
Wording drifts into telling a patient what to do.
Stage sends
Only where consent, channel and quiet hours allow it
04 · Reply / escalation
1 mode
- PD-06
Logged, not escalated
An out-of-hours symptom reply waits for morning.
Stage routes
The reply, the clock it started and who it reached
05 · Change / Version
1 mode
- PD-07
Outreach fatigue
More messages, fewer patients still reachable.
Stage tracks
Model, prompt, gap-logic and escalation-rule changes
- Sev-1 · the agent said what it must not
- Sev-2 · a clinical concern waits for a person
- Sev-3 · outreach misses who needed it
Affected slices
Messages sent is not the same as patients reached
Aggregate outreach volume can look strong while a small number of patient cohorts carry most of the unreached contacts and most of the late escalations. 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 |
|---|---|---|---|---|
| Patients without SMS or portal | 5.9% | 3.7× | Review | |
| Interpreter-assisted outreach | 4.3% | 2.7× | Review | |
| Symptom replies out of hours | 3.0% | 1.9× | Watch | |
| Routine screening reminders | 1.0% | 0.6× | Normal |
Bar: lift vs. routine-reminder baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
Every late reply changes the clock for the next one
A concern that waited too long is not closed when that patient is finally called. The fix lands in the escalation list, the clock or the rota.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Escalation latency, unreached patients or wasted chases move in a cohort.
- 02
Diagnose
Traced to the eligibility feed, the target list, the wording, the classifier or the rota.
- 03
Improve
The escalation list, clock, rota or suppression rule is re-approved and version-linked.
- 04
Verify
Re-run against held-out replies from the affected cohort, including the ones that were missed.
- 05
Learn
That reply becomes a regression case and the changed rule enters the outreach runbook.
Learn → Detect
The return edge. A cadence or targeting change decides who hears from you next, so the next cycle re-reads the reach slices, not only the escalation total.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, feeds and eligibility, content and routing, escalation and evaluation, channels, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Workflow discovery and automation-boundary definition.Week 1
- 02ADT, discharge and death-status feeds.Week 1, Week 2
- 03Consent, contact and quiet-hours rules.Week 1, Week 2
- 04Eligibility and suppression logic.Week 2
- 05Check-in content and approved wording.Week 2, Week 3
- 06Care-gap definitions and evidence checks.Week 3
- 07Reply classification and escalation routing.Week 3, Week 4
- 08Escalation clocks, rota and out-of-hours path.Week 4
- 09Evaluation suite, reach slices and regression cases.Week 4, Week 5
- 10Telephony, SMS and portal channel integration.Week 5
- 11Write-back to the record and care queue.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 discharge cohort | ProductionProduction EHR integration | AdvancedMulti-site / payer programmes |
|---|---|---|---|
| Introduced at Pilot | |||
| Check-ins on clinician-approved wording | ✓ | ✓ | ✓ |
| Eligibility and consent suppression | ✓ | ✓ | ✓ |
| Named-clinician escalation on a clock | ✓ | ✓ | ✓ |
| Baseline evaluation and reach reporting | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Care gaps evidenced against the record | — | ✓ | ✓ |
| Follow-up confirmation and prep chasing | — | ✓ | ✓ |
| Two-way SMS, telephony and portal | — | ✓ | ✓ |
| Recorded-language and interpreter routing | — | ✓ | ✓ |
| Observability and evaluation | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Care-management and registry write-back | — | — | ✓ |
| Multi-site and payer-programme controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on ADT and EHR integrations, outreach channels, care-gap scope, escalation cover, patient volume, 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 consent, contact-preference, language and opt-out rules
→Consent, contact-preference and quiet-hours rules
Week 102ADT, discharge and death-status feeds, and how current they are
→Feed assessment, then eligibility and suppression logic
Week 203The discharge pathways and what each one asks patients about
→Check-in content and clinician-approved wording
Week 204The care-gap definitions your measure engine actually uses
→Care-gap definitions and evidence checks against the record
Week 305The wording that must escalate, and the clock it must meet
→Reply classification and escalation routing
Week 306Named clinicians on the rota, including out of hours
→Escalation clocks, rota and out-of-hours path
Week 407Replies that were missed, and patients who were never reached
→Evaluation suite, reach slices and regression cases
Weeks 4–5
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. No message reaches a patient before the escalation rota is standing, which is why week 5 is the first week anything is sent.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Consent, quiet hours and the wording a clinician has approved
W2ADT and discharge feeds, then eligibility and suppression
W3Check-in content, care-gap definitions and reply routing
W4Escalation clocks and rota, then the evaluation suite
W5Channel integration, first supervised outreach and corrections
W6Real replies escalated on the clock, then Agent Care picks up
- Reading the band
- Escalation sits in week 4, ahead of outreach — the rota and the clock exist before anything is sent, not after.
- At the end of W6
- Outreach has run alongside the existing calls and real replies have been escalated on the clock, 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 outreach agent around your discharge and gap lists.
Show us your discharge pathways, your open care gaps and who picks up a worrying reply at two in the morning. We'll write the escalation list and the clock first, then scope what the agent may send.
- Typical delivery
- 4–6 weeks
- Deployment
- Your EHR and outreach stack
- Clinical replies
- To a named clinician, on a clock
- After launch
- Agent Care available
Nestack Agents · Post-discharge & care-gap outreach
AGT-HC-07 · Agent Care available after launch