The operational backbone of patient care
Patient operations encompasses everything that happens around the clinical encounter: getting the patient to the right provider at the right time, collecting and processing their information, coordinating their care across providers and settings, managing the financial transaction, and ensuring the overall experience meets their expectations.
These workflows are overwhelmingly text-based, repetitive, and rule-driven — and they are where health systems lose the most operational efficiency. A patient who arrives for a new-patient visit fills out forms that duplicate information already in the EHR. A referral sits in a queue because the required clinical documentation is incomplete. A claim is denied because a registration error put the wrong insurance information on the encounter. A patient satisfaction score drops because a follow-up call was never made.
Each of these is a process failure, not a clinical failure. And each is addressable with AI — not by replacing the staff who manage these workflows, but by eliminating the manual, repetitive components so staff can focus on the exceptions and the human interactions that matter.
Which patient operations workflow creates the most friction in your organisation?
Patient intake form processing
Patient intake at most health systems still involves paper forms or PDF-based digital forms that are manually entered into the EHR. A new-patient visit might require: a demographic form, an insurance verification form, a medical history questionnaire, a medication list, an allergy list, a surgical history form, a family history form, and a review of systems.
The problems are well-known: patients leave fields blank, handwriting is illegible, insurance information is incomplete, medication names are misspelled, and allergy documentation lacks the specificity the EHR requires (drug allergy vs drug intolerance, specific reaction type).
AI can process intake forms by extracting structured data, validating it against known formats, and flagging issues before the data enters the EHR:
ROLE: You are a patient access specialist processing new patient intake forms
for data entry into the EHR.
SOURCE DATA: The following is a completed patient intake form.
[Paste or attach the form data]
TASK:
1. DEMOGRAPHIC EXTRACTION: Extract and structure:
- Full name, date of birth, sex, preferred language
- Address (verify format: street, city, state, ZIP)
- Phone numbers (home, mobile, work) with preferred contact method
- Emergency contact (name, relationship, phone)
- Employer information if applicable
2. INSURANCE VERIFICATION PREP: Extract:
- Primary insurance: carrier name, plan type, group number, member ID,
subscriber name and relationship to patient
- Secondary insurance (if applicable): same fields
- Flag any fields that are missing or potentially incorrect:
- Member ID format does not match known carrier formats
- Group number appears incomplete
- Subscriber relationship is unclear
3. MEDICAL HISTORY STRUCTURING: Extract and organise:
- Active medical conditions (map to common condition names)
- Surgical history with approximate dates
- Current medications: name, dose, frequency, route
- Flag any medication names that appear misspelled and suggest corrections
- Flag any medications missing dose or frequency
- Allergies: substance, reaction type, severity
- Distinguish drug allergies from drug intolerances
- Flag allergies listed without a documented reaction type
- Family history: condition, relationship, age of onset if noted
4. QUALITY FLAGS:
- List all blank required fields
- List all fields with potentially incorrect or incomplete data
- List all items that need clarification from the patient before entry
OUTPUT: Structured data ready for EHR entry, plus a flag list for front
desk staff to resolve with the patient before the visit begins.This workflow shifts intake staff from manual data entry to exception management. Instead of typing every field, they review the AI-extracted data, resolve the flagged items, and confirm the entry. For a health system processing 200 new patients per day, this can save 4-6 hours of registration staff time daily.
Appointment scheduling optimisation and referral management
Appointment scheduling at scale involves matching patient needs (urgency, provider preference, insurance network, visit type) with available capacity (provider schedules, room availability, equipment requirements, preparation time). Referral management adds another layer: ensuring the referring provider's clinical documentation is complete, the insurance authorisation is in place, and the patient is actually scheduled and follows through.
AI assists with the decision-support layer of scheduling and referrals — not the scheduling engine itself, but the intelligence that informs scheduling decisions:
ROLE: You are a care coordination specialist managing incoming referrals for
a [specialty department / multispecialty practice].
SOURCE DATA: The following is a batch of incoming referral requests with
attached clinical documentation.
[Paste referral data: referring provider, patient demographics, reason for
referral, clinical documentation, insurance information]
TASK:
1. TRIAGE: Classify each referral by urgency:
- URGENT (within 48-72 hours): Clinical indicators suggesting immediate
specialist evaluation needed (specify which indicators)
- ROUTINE (within 2-4 weeks): Standard specialist consultation
- FOLLOW-UP (within 4-8 weeks): Established patient, routine follow-up
- REDIRECT: Referral appears to be for the wrong specialty — suggest
the correct specialty based on the clinical documentation
2. DOCUMENTATION COMPLETENESS:
For each referral, check whether the required clinical documentation
is present:
- Reason for referral with specific clinical question
- Relevant history and physical findings
- Relevant diagnostic results (labs, imaging, pathology)
- Current medication list
- Insurance authorisation status
Flag any referral where documentation is incomplete and specify
what is missing.
3. SCHEDULING INTELLIGENCE:
- Suggested visit type (new consult, follow-up, procedure, etc.)
- Estimated visit duration based on complexity
- Required resources (specific equipment, interpreter services, etc.)
- Pre-visit instructions for the patient (fasting, medication holds, etc.)
4. PATIENT COMMUNICATION DRAFT:
For each scheduled referral, draft a patient notification that includes:
- Appointment details (provider, date/time placeholder, location)
- What to bring (insurance card, medication list, imaging CDs)
- Pre-visit preparation instructions
- How to reschedule if needed
OUTPUT: Referral triage list sorted by urgency, documentation gap report,
scheduling recommendations, and patient communication drafts.Referral leakage — patients who receive a referral but never complete the specialist visit — is estimated at 25-50% across health systems. AI-managed referral workflows that triage, flag documentation gaps, and generate patient communications can significantly reduce this leakage rate, improving both patient outcomes and downstream revenue.
What is your organisation's referral completion rate?
Patient communication drafting and satisfaction analysis
Patient communication has become a major operational burden since the adoption of patient portals. Health systems report that patient portal message volumes have increased 200-300% since 2020, with physicians receiving 50-100+ messages per day. Many of these are routine: medication refill requests, appointment questions, test result inquiries, and general health questions.
AI can draft responses for physician review, dramatically reducing the time physicians spend on inbox management:
ROLE: You are a patient communication specialist drafting responses to
patient portal messages for physician review and approval.
SOURCE DATA: The following is a patient portal message.
[Paste the patient message]
PATIENT CONTEXT (from EHR): [Recent diagnoses, medications, upcoming
appointments, recent lab results — de-identified or via HIPAA-compliant system]
TASK: Draft a response that:
1. Acknowledges the patient's question or concern
2. Provides the appropriate response based on the message type:
- MEDICATION REFILL: Confirm the medication, verify it is current,
note when the refill will be processed
- APPOINTMENT QUESTION: Provide scheduling information or instructions
for rescheduling
- TEST RESULT INQUIRY: Reference the result (if normal and within
standard communication protocols) or indicate that the provider
will review and follow up
- CLINICAL QUESTION: Draft a response based on the patient's documented
care plan — flag for physician review before sending
- ADMINISTRATIVE: Address billing, records, or referral questions
and route to the appropriate department if needed
3. Use patient-friendly language at a 6th-8th grade reading level
4. Include any relevant next steps or action items
CONSTRAINTS:
- Do not provide new medical advice or clinical recommendations
- Do not interpret test results outside of established communication
protocols (e.g., normal results with standard ranges)
- Flag any message that describes new symptoms, medication side effects,
or clinical deterioration — these require direct physician review,
not an AI-drafted response
- Flag any message with emotional distress indicators for priority
physician or care team review
CLASSIFICATION:
- AUTO-SEND ELIGIBLE: Routine administrative response that meets
established communication protocols
- PHYSICIAN REVIEW: Clinical content that requires physician approval
before sending
- PRIORITY ESCALATION: Urgent clinical or emotional content requiring
prompt physician attentionPatient satisfaction survey analysis is another high-value AI workflow. Most health systems collect thousands of patient satisfaction surveys (HCAHPS, Press Ganey, CG-CAHPS) but lack the capacity to analyse free-text comments at scale:
ROLE: You are a patient experience analyst reviewing patient satisfaction
survey free-text comments.
SOURCE DATA: The following are [N] free-text comments from [survey type]
collected during [time period].
[Paste the comment data — ensure patient identifiers are removed]
TASK:
1. THEME EXTRACTION: Identify the top themes across all comments,
ranked by frequency:
- Positive themes (what patients value most)
- Negative themes (what drives dissatisfaction)
- Suggestions (what patients want improved)
2. DEPARTMENT/SERVICE ATTRIBUTION: Where comments reference specific
departments, units, or service lines, attribute the feedback accordingly.
3. TREND IDENTIFICATION: If comments span multiple time periods,
identify trends — improving or declining themes.
4. ACTIONABLE INSIGHTS: For each negative theme, suggest specific
operational interventions and indicate which department should own
the improvement effort.
5. VERBATIM HIGHLIGHTS: Select 5-10 representative verbatim comments
for each major theme (positive and negative) suitable for inclusion
in a leadership report.Population health analytics and care coordination documentation
Population health management requires identifying patients at risk of adverse outcomes — hospital readmissions, emergency department visits, disease progression — and intervening before the adverse event occurs. This involves analysing patterns across large patient populations: demographic data, diagnosis histories, utilisation patterns, social determinants of health, and care gap indicators.
AI assists with the analytical and documentation components of population health — not the clinical decisions about individual patients:
ROLE: You are a population health analyst reviewing patient panel data for
a [primary care practice / accountable care organisation / value-based care programme].
SOURCE DATA: De-identified patient panel summary data including:
- Demographics (age, sex, geography)
- Active diagnoses (chronic conditions)
- Utilisation history (ED visits, hospitalisations, 30-day readmissions)
- Care gap indicators (overdue screenings, missed appointments, medication
non-adherence signals)
- HEDIS/quality measure compliance status
- Social determinant of health indicators (if available)
TASK:
1. RISK STRATIFICATION: Identify patients at highest risk for:
- 30-day hospital readmission (based on recent discharge, diagnosis
complexity, prior readmission history)
- ED utilisation for ambulatory-sensitive conditions
- Disease progression (e.g., diabetic patients with rising HbA1c trend,
hypertensive patients with uncontrolled BP readings)
- Care gap accumulation (multiple overdue preventive services)
2. PATTERN IDENTIFICATION: Across the panel, identify:
- Which chronic conditions are driving the most utilisation?
- Which HEDIS measures have the lowest compliance rates?
- Are there demographic or geographic clusters of high-risk patients?
- What are the most common care transitions (e.g., ED to inpatient,
inpatient to SNF) and where do breakdowns occur?
3. INTERVENTION PRIORITISATION: Rank the top 10 opportunities to
improve population health outcomes, based on:
- Number of patients affected
- Potential clinical impact
- Feasibility of intervention
- Financial impact under the current payment model
4. OUTREACH LIST: Generate a prioritised outreach list for care
coordinators, with:
- Patient identifier (de-identified)
- Risk factors present
- Recommended outreach type (phone call, home visit, telehealth,
pharmacy consultation)
- Specific care gaps to address during outreachThis workflow turns raw data into actionable care coordination priorities. A care coordinator who previously spent hours reviewing charts to build their outreach list can instead review an AI-generated prioritised list and focus their time on the patient interactions.
Is your organisation participating in value-based care contracts?
Revenue cycle management — claims scrubbing and denial management
Revenue cycle management (RCM) is where every upstream documentation decision becomes a financial outcome. A clean claim — one that is submitted correctly the first time and paid without intervention — represents the ideal. Every deviation from that ideal costs money: a denied claim costs $25-118 to rework, a delayed claim delays cash flow, and a written-off claim is revenue lost permanently.
AI addresses the revenue cycle at multiple points:
Claims scrubbing — reviewing claims before submission to catch errors that would trigger denials:
ROLE: You are a revenue cycle analyst conducting pre-submission claims scrubbing.
SOURCE DATA: The following is a batch of claims ready for submission.
[Claim data: patient demographics, insurance, encounter details, diagnosis
codes, procedure codes, modifiers, charges]
TASK: Review each claim for:
1. CODING VALIDATION:
- Do the diagnosis codes support the procedure codes per LCD/NCD requirements?
- Are modifiers correct and necessary (e.g., modifier 25 for significant
E/M with a procedure)?
- Is the principal diagnosis sequencing correct?
- Are the diagnosis codes coded to the highest specificity available
in the documentation?
2. INSURANCE VERIFICATION:
- Is the insurance information complete (payer ID, group, member ID)?
- Is the claim being submitted to the correct primary/secondary payer?
- Is prior authorisation required and documented for the services billed?
- Is the rendering provider in-network for this payer?
3. TECHNICAL VALIDATION:
- Do date fields match (date of service, admission/discharge dates)?
- Are place of service codes correct?
- Are units billed consistent with the documentation?
- Do charges fall within expected ranges for the procedure codes?
4. DENIAL RISK SCORING:
For each claim, assign a denial risk score:
- LOW: All elements verified, clean claim expected
- MEDIUM: Minor issues that could trigger a request for information
- HIGH: Issues identified that will likely result in denial — hold
for correction before submission
OUTPUT: Claims sorted by denial risk score, with specific issues identified
and recommended corrections for MEDIUM and HIGH risk claims.Denial management — analysing denials, identifying root causes, and generating appeal strategies:
ROLE: You are a denials management analyst reviewing a batch of denied claims.
SOURCE DATA: [Denied claims data with denial reason codes, original claim
details, and available clinical documentation]
TASK:
1. DENIAL CATEGORISATION: Classify each denial by root cause:
- REGISTRATION: Insurance verification, demographic errors, eligibility
- AUTHORISATION: Missing or expired prior auth
- CODING: Code not supported by documentation, specificity issues,
bundling/unbundling errors
- MEDICAL NECESSITY: Payer determined service not medically necessary
- TIMELY FILING: Claim submitted after payer filing deadline
- DUPLICATE: Claim identified as duplicate submission
2. ROOT CAUSE ANALYSIS: Across the denial batch, identify:
- Top 5 denial reasons by volume and dollar amount
- Departments or service lines with highest denial rates
- Payers with highest denial rates
- Patterns suggesting systemic issues (e.g., all denials for a specific
CPT code from a specific payer)
3. APPEAL STRATEGY: For each denial where appeal is warranted:
- Recommended appeal approach (written appeal, peer-to-peer, second-level)
- Key evidence to include in the appeal
- Expected overturn probability based on denial reason and payer
- Priority ranking based on dollar amount and overturn probability
4. PREVENTION RECOMMENDATIONS: For the top denial root causes,
recommend process changes to prevent future denials:
- Registration workflow changes
- Pre-service authorisation checkpoints
- Documentation improvement targets for clinical staff
- Coding education needsModule 5 — Final Assessment
Why is patient intake form processing a high-value AI use case?
What is referral leakage and why does it matter?
In population health analytics, what role does AI play vs clinical staff?
What is the purpose of pre-submission claims scrubbing with AI?
