Healthcare AI agent · CDI query
Clinical-Documentation-Integrity Query AI Agent (CDI)
Read the chart after the encounter, surface the indicators that make it ambiguous, and draft a query in the health system's own templates for the treating clinician to answer — or not.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Clinician-led
- Post-encounter
- After launch
- Agent Care
What this agent does
Drafts the question, never the answer
In
- 01
The chart is read after the encounter closes, against the facility's configured query triggers and template library.
- 02
Normalise indicator values, their date and their exact location in the chart across sources.
Reason
- 03
Identify where documentation is ambiguous, conflicting or silent against a diagnosis the chart may or may not support.
- 04
Apply the approved query templates and compliant-query rules configured for the facility, never assumed.
- 05
Bind each cited indicator to its location in the record, and mark what remains clinically undetermined.
Decide
- 06
Build multiple-choice options as clinically reasonable alternatives, 'clinically undetermined' included.
- 07
Route every drafted query to the treating clinician who provided care in that encounter, for their answer alone.
Out
- 08
Retain the indicators cited, their chart location, the draft sent and the clinician's answer against the record.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent drafts the query; only the treating clinician may answer it, and that answer must reach the permanent record before anything downstream can use it.
Example workflow
One query, chart to clinician
- 1Record reviewedEHR chart, closed encounter note, labs and prior query history
- 2Indicators gatheredLabs, vitals, orders, prior notes and each one's exact location in the chart
- 3Query draftedDraft query, cited indicators and confidence
- 4Controls appliedLeading-language checks, template compliance checks, indicator-support checks and confidence threshold
No human action required
Stages 1 to 4 run unaided, and no query reaches a clinician at any of them — the agent is drafting, and the CDI specialist's lane opens at the confidence gate.
Goes to the CDI specialist to approve.
Adds a second compliance read first.
Specialist approval
The draft is held with its cited indicators, its chart locations and the confidence.
Approve · Edit · Send for compliance review
Approved — released to clinician
- 6Query routed to clinicianOnly where write access and approval policy allow it
- 7Outcome evaluatedResponse rate, edit distance, indicator-support accuracy and compliance-audit findings
Specialist edits
Every specialist edit made at approval is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Naming or implying the diagnosis a query is meant to produce.
- Recording an answer on the clinician's behalf.
- Sending a query to anyone but the treating provider from that encounter.
- Disclosing reimbursement, DRG weight or severity impact in a query.
Automation boundary
Agent acts unaided
- ✓Read the chart after the encounter and cite.
- ✓Draft the query in the facility's own approved templates.
- ✓Build multiple-choice options as reasonable alternatives, undetermined included.
- ✓Flag what the record leaves unclear, and hold the draft for the CDI specialist.
Any write happens inside the boundaries agreed at implementation, never ahead of approval.
- Resending a query for the same diagnosis without a new indicator.
- Setting or enforcing a response deadline with a consequence attached.
- Using a yes/no query to originate an undocumented diagnosis.
- Assigning a code, or documenting the note during the encounter.
Example output
One query, annotated
Everything the agent drafts is attached to the indicators it was drawn from.
Query output · single encounter
Illustrative example- Patient
- Respiratory-status indicator review
- Drafted query
- Sepsis, another documented process or clinically undetermined — based on the charted vitals and labs
- Query type
- Multiple-choice
- Indicator source
- Nursing flowsheet, day 2
- Confidence
- 88%
- Clinician status
- Sent to the treating hospitalist
As received
Taken from the chart and the locations cited — nothing on this side is interpreted by the agent.
Indicators cited
- Vital-sign flowsheet
- Lab result with time
- Prior query this stay
Why this reads as ambiguous
The options are clinically reasonable alternatives, undetermined included.
What the score decides
Below the configured threshold the draft picks up a second compliance read before.
Action
ApproveEditSend for compliance reviewValue
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 closed encounter
From the chart and the query history
Read for ambiguity
Weigh the indicators already charted against the diagnosis they do, and do not, support.
A query may not suggest its answer
An indicator that supports a diagnosis and one that undercuts it move through the same read, not two.
Point the specialist at what needs
Indicators with no clear chart location and low-confidence drafts are marked, so the specialist's read starts where ambiguity concentrates.
Build an evidence trail
The query, the indicators behind it and the physician who answered stay in the record.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
EHR and chart systems
- Epic
- Cerner
- MEDITECH
- HL7/FHIR chart feeds
CDI and query platforms
- 3M 360 Encompass
- Iodine
- Optum CDI
- nThrive
Coding and DRG systems
- 3M encoder
- TruCode
- MS-DRG grouper feeds
Agent
Clinical-documentation-integrity query
- Reads the chart
- Drafts the query
- Holds for approval
Query and HIM systems
- CDI worklist tools
- HIM and audit-trail systems
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 clinician
Each layer contains the next. What escapes all of them is named in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a drafted query may stand
- L3
- Puts the release in a specialist's hands
- L4 – L5
- Keep the query and the indicators behind it
- L6
- Falls back to indicator surfacing when signals degrade
How Nestack evaluates it
Evaluate the query workflow — not only the finished draft.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the query the clinician sees
- E1Final-output evaluationDid the drafted query cite indicators the chart actually documents?
- E2Step-level evaluationDid the agent use the right chart, template rules and trigger configuration?
- E3Tool evaluationDid it read the correct chart and route to the correct treating clinician?
- E4Confidence calibrationDo low-confidence drafts actually draw more specialist edits?
- E5Slice evaluationHow does performance change across specific query types?
- E6Business outcomeHow many drafted queries needed a specialist edit or were found non-compliant on audit?
Floor — the outcome the health system answers for
Failure modes
Where each failure originates in the agent
Seven failure modes, placed at the stage in the query workflow each one begins.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- DQ-03
Superseded value cited
A preliminary lab or vital is queried after a corrected result posts to the chart.
Stage gathers
Chart notes, labs and the template library
02 · Reasoning
2 modes
- DQ-04
Leading-query construction
Option order or wording points toward one diagnosis before the query sends.
- DQ-06
Indicator interpreted, not quoted
A term such as malnourished replaces the value the guideline requires quoted.
Stage proposes
Draft query, cited indicators and confidence
03 · Tool / write
2 modes
- DQ-02
Response inferred, not entered
A likely answer is treated as documented before the clinician enters one.
- DQ-05
Query resent without new indicators
The same diagnosis is queried again after an unfavourable answer, indicators unchanged.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- DQ-01
Financial rationale attached
Reimbursement or severity impact is visible in the query or its surrounding screen.
Stage returns
The query the clinician answers
05 · Change / Version
1 mode
- DQ-07
Silent trigger-point drift
A coding-guideline or DRG-table update ships without the trigger points recalibrated.
Stage tracks
Model, prompt, triggers and templates
- Sev-1 · a query sends outside the boundary
- Sev-2 · a leading query reaches the clinician
- Sev-3 · indicators degrade, goes to review
Affected slices
One query type can carry the rest
A service line's query-compliance rate can look acceptable while one indicator or query type accounts for most of what a compliance audit would flag. Nestack reports the rate by slice, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Queries raised on a single indicator | 7.6% | 3.9× | Review | |
| Sepsis and respiratory-failure queries | 5.5% | 2.9× | Review | |
| Post-discharge retrospective queries | 3.4% | 1.8× | Watch | |
| Concurrent queries with full indicators | 1.9% | 0.8× | Normal |
Bar: non-compliant-query rate lift vs. the concurrent full-indicator baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
The loop closes on the query, not the volume
A cycle is closed when the leading query is a case the next release must survive. That suite is what the next query raised is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Non-compliant-query rate rises in a service-line slice.
- 02
Diagnose
The query that named the diagnosis it wanted is traced to the template, the trigger rule or the training data that produced it.
- 03
Improve
The change ships against a version, with the queries that exposed it attached.
- 04
Verify
Release is blocked until the affected query cases pass again.
- 05
Learn
The case joins the standing suite and the query templates move with it.
Learn → Detect
The return edge. The next query drafted runs against a suite one case longer.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, sources, query workflow, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Query-workflow discovery and boundary definition.Week 1
- 02EHR and chart-source assessment.Week 1, Week 2
- 03Template and trigger-rule mapping and rule mapping.Week 1, Week 2
- 04Chart ingestion and indicator normalising.Week 2
- 05Indicator retrieval and drafting logic.Week 2, Week 3
- 06Confidence scoring and flag routing.Week 3
- 07Specialist approval workflow.Week 3, Week 4
- 08Chart-source and coding-system integration.Week 5
- 09Query-construction cases.Week 4, Week 5
- 10Guardrails and compliance controls.Week 4
- 11Query-trail instrumentation.Week 5, Week 6
- 12Deployment, documentation 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 facility, one service line | ProductionProduction chart integration | AdvancedMultiple facilities / lines |
|---|---|---|---|
| Introduced at Pilot | |||
| Query drafting from your chart | ✓ | ✓ | ✓ |
| Specialist approval | ✓ | ✓ | ✓ |
| Query-compliance baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by service line | — | ✓ | ✓ |
| Approval workflow in your systems | — | ✓ | ✓ |
| Approved query write-back | — | ✓ | ✓ |
| Record-system integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-facility query rules | — | — | ✓ |
| Multi-stage specialist approvals | — | — | ✓ |
| High query volume | — | — | ✓ |
| Multi-facility query controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, transaction 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 EHR access, chart sources and query history
→Chart ingestion and prior-query mapping
Week 102Representative charts across query-type cohorts
→Draft baseline and indicator-citation extraction
Week 203Your approved query templates and trigger rules
→Template and trigger-rule mapping
Week 104Access to relevant APIs, feeds or exports
→Chart-source and coding-system assessment, then integration setup
Week 205Queries you would not want sent
→Leading-query cases and failure-mode testing
Week 406What no query may imply
→Confidence scoring, flag routing, guardrails and approval controls
Week 307Named CDI specialists to approve drafted queries
→Specialist approval workflow, then pilot 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
Each phase sits over the weeks it really occupies, and week 5 carries evaluation alongside the pilot.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Query workflow discovery, template mapping and the boundary
W2Chart-source integration and the drafting baseline
W3Drafting logic, confidence scoring and approval controls
W4Evaluation suite, guardrails and failure-mode testing
W5Query-system integration, pilot charts and corrections
W6One coding cycle queried under the CDI lead, then Agent Care handover
- Reading the band
- A bar covers the weeks its work is named in, and nothing else. The week 5 overlap is real, not padding.
- At the end of W6
- The last checks clear on live queries and Agent Care picks up monitoring.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Healthcare AI agent
Build a CDI-query agent around your compliant-query workflow.
Show us your chart sources, your approved templates and who answers a query. The answer to every query stays with the treating clinician, entered in the record.
- Typical delivery
- 4–6 weeks
- Deployment
- Your EHR and chart sources
- Clinician-led
- Post-encounter
- After launch
- Agent Care available
Nestack Agents · CDI query
AGT-HC-16 · Agent Care available after launch