AI use cases in healthcare: 12 systems teams build in 2026
Twelve AI systems providers, payers and digital health companies are building, with EHR integrations, HIPAA and FDA notes and illustrative team budgets.
By the Ryz Labs team · Updated October 2026
The AI use cases healthcare organizations build first are administrative: ambient clinical documentation, prior authorization, coding assistance, denial appeals, patient access agents and inbox triage. They take hours out of clinical and revenue cycle work without making a diagnosis, which keeps them under HIPAA's privacy and security requirements but generally outside FDA device regulation.
This guide covers twelve systems for providers, payers and digital health companies, with the data, the integrations that usually set the timeline, the hard parts and an illustrative team and budget. Ryz serves enterprises across industries, including healthcare. Our AI pod teams build in your cloud and repos, working on your team, and our engineers have experience working within HIPAA and SOC 2 controls.
About the numbers. Team shapes and build times are illustrative planning ranges, not quotes. Costs use only Ryz's public rates: mid-level $7,000 to $10,000, senior $10,000 to $15,000 and tech lead $15,000 and up per engineer per month, quoted per team, with ranges at the $15,000 lead floor.
Ground rules before you build
- HIPAA. Protected health information can only go to model providers and cloud services covered by a business associate agreement, under the minimum necessary standard, with Security Rule safeguards (access control, audit logs, encryption) in place.
- FDA. Software that diagnoses or drives treatment can be a medical device. The FDA's clinical decision support software guidance explains which decision support functions fall outside device regulation, largely those where a clinician can independently review the basis for a recommendation.
- Certified health IT and nondiscrimination. ONC's HTI-1 rule added transparency requirements for decision support interventions in certified EHRs, and HHS's Section 1557 rule requires covered entities to identify and reduce discrimination risks in patient care decision support tools.
Clinical workflow
1. Ambient clinical documentation
With patient consent, the system listens to the visit, produces a draft note in the clinician's preferred structure (HPI, exam, assessment and plan), proposes orders and codes for review, and files the signed note to the EHR.
- Data: visit audio, note templates per specialty, the clinician's prior notes and edits.
- Integrations: Epic or Oracle Health through their APIs and FHIR, mobile capture app, identity and consent records.
- What makes it hard: specialty vocabulary, multiple speakers, and recording consent rules that vary by state. The clinician signs every note, and edit rates by section are the core quality metric.
- Illustrative team and build: tech lead plus four senior engineers, six to eight months: $55,000 to $75,000 a month, $330,000 to $600,000 in total.
2. Patient portal message triage and draft replies
Clinician inboxes fill with portal messages. The system classifies each one (refill, results question, scheduling, new symptom), routes it to the right pool, drafts a reply for routine questions and escalates anything that reads as urgent.
- Data: message history, routing rules, approved reply templates, outcomes.
- Integrations: EHR inbox, nurse triage protocols, patient portal.
- What makes it hard: missing an urgent symptom is the failure that matters. Escalation recall is measured on a labeled set and monitored after launch.
- Illustrative team and build: tech lead plus three senior engineers, four to six months: $45,000 to $60,000 a month, $180,000 to $360,000 in total.
3. Chart abstraction for registries and trials
Quality registries and clinical trial screening require reading charts for specific criteria. The system extracts the elements each registry or protocol needs, cites the note and date, and leaves final abstraction or eligibility to a trained abstractor or coordinator.
- Data: clinical notes, labs, problem lists, registry definitions or inclusion criteria.
- Integrations: EHR data warehouse, registry submission tools, CTMS.
- What makes it hard: negation and timing ("no history of," "ruled out," "prior to 2024"). Evaluation must use criteria-level labels. See NLP development.
- Illustrative team and build: tech lead plus two senior engineers, four to five months: $35,000 to $45,000 a month, $140,000 to $225,000 in total.
4. Care management outreach prioritization
Care managers cannot call every member or patient. A risk model ranks who to contact this week using utilization, gaps in care and social needs data, and explains the top factors so care managers can judge the list.
- Data: claims, EHR data, care gap reports, social needs screenings, outreach outcomes.
- Integrations: care management platform, data warehouse, outreach tools.
- What makes it hard: models trained on past spending can under-rank patients who had less access to care. Section 1557 obligations make bias testing a requirement, not an extra.
- Illustrative team and build: tech lead plus three senior engineers, four to six months: $45,000 to $60,000 a month, $180,000 to $360,000 in total.
Revenue cycle and payer operations
5. Prior authorization assembly
On the provider side, the system reads the order and chart, determines whether the payer requires authorization, assembles the clinical documentation that matches the payer's criteria, and submits it. On the payer side, it summarizes incoming requests against medical policy for the reviewing nurse or physician. The CMS Interoperability and Prior Authorization Final Rule (CMS-0057-F) requires impacted payers to support prior authorization through FHIR APIs, which makes this workflow more automatable.
- Data: orders, clinical notes, payer policies, past determinations.
- Integrations: EHR, payer portals and FHIR APIs, X12 278 transactions, utilization management systems.
- What makes it hard: payer criteria change often and differ by plan. A model can assemble and summarize, but denials on medical necessity must be made by qualified clinicians.
- Illustrative team and build: tech lead plus four senior engineers, five to seven months: $55,000 to $75,000 a month, $275,000 to $525,000 in total.
6. Coding assistance
The system reads the encounter documentation and suggests ICD-10-CM, CPT and HCPCS codes with the supporting text highlighted, for a certified coder to accept, edit or reject, and flags documentation gaps for clinician queries.
- Data: encounter notes, coded claims history, coding guidelines, coder edits.
- Integrations: EHR, encoder and coding workflow tools, billing system.
- What makes it hard: upcoding risk. Suggestions must be supported by documentation, and accuracy is measured against coder decisions, not volume.
- Illustrative team and build: tech lead plus three senior engineers, five to seven months: $45,000 to $60,000 a month, $225,000 to $420,000 in total.
7. Denial management and appeal letters
The system reads remittance data and denial reason codes, groups denials by root cause, and drafts appeal letters that cite the payer's policy and the chart evidence, for the appeals specialist to review and send.
- Data: X12 835 remittances, claims, payer policies, chart documentation, appeal outcomes.
- Integrations: billing system, clearinghouse, document management.
- What makes it hard: appeal deadlines vary by payer, and winning arguments differ by denial type. Outcome tracking feeds the next draft.
- Illustrative team and build: tech lead plus two senior engineers and one mid-level engineer, four to five months: $42,000 to $55,000 a month, $168,000 to $275,000 in total.
8. Eligibility and benefits verification
Before a visit, the system checks coverage, plan details and patient cost share through eligibility transactions and payer portals, reads benefit details that come back as free text, and flags problems for patient access staff.
- Data: X12 270/271 responses, payer portal content, scheduling data.
- Integrations: practice management, clearinghouse, scheduling.
- What makes it hard: free-text benefit notes and payer portals with no API. Browser automation needs monitoring because portals change.
- Illustrative team and build: tech lead plus two senior engineers, three to four months: $35,000 to $45,000 a month, $105,000 to $180,000 in total.
9. Payment integrity review
For payers, the system flags claims with likely coding errors, duplicate billing or patterns that suggest fraud, waste or abuse, and assembles the evidence for a reviewer before or after payment.
- Data: claims, provider history, medical policies, review outcomes.
- Integrations: claims platform, case management, SIU tools.
- What makes it hard: provider abrasion. False positives delay valid payments, so precision targets are set with network management. The invoice-scoring pattern from our fraud detection case study for a global fleet company applies here.
- Illustrative team and build: tech lead plus three senior engineers, five to seven months: $45,000 to $60,000 a month, $225,000 to $420,000 in total.
Patient access and operations
10. Scheduling and patient access voice agent
A voice agent answers scheduling calls, books, moves and cancels appointments against real provider templates, handles directions and preparation questions, and transfers clinical questions to staff. Outbound, it handles reminders and waitlist fills. A Ryz AI pod built an AI voice platform that has placed 1M+ outbound calls.
- Data: scheduling templates, visit types, call transcripts, FAQs.
- Integrations: EHR scheduling APIs, telephony, identity verification.
- What makes it hard: visit-type rules (new vs established, referral required, prep instructions) and recognizing callers who describe urgent symptoms.
- Illustrative team and build: tech lead plus four senior engineers, five to seven months: $55,000 to $75,000 a month, $275,000 to $525,000 in total.
11. Fax and referral intake
Referrals, records and orders still arrive by fax. The system classifies each document, matches it to the patient, extracts the referral details and creates the work item, so nothing sits in a fax queue.
- Data: inbound faxes, patient index, referral requirements.
- Integrations: fax server, EHR document import, referral management.
- What makes it hard: patient matching. A document filed to the wrong chart is a privacy incident, so low-confidence matches go to a person.
- Illustrative team and build: tech lead plus one senior and one mid-level engineer, two to three months: $32,000 to $40,000 a month, $64,000 to $120,000 in total.
12. Clinician knowledge assistant
Clinicians and nurses ask about internal protocols, order sets and formulary rules. A retrieval system answers from approved internal content only, cites the source and its review date, and declines questions outside that content.
- Data: policies, protocols, formulary, order set documentation.
- Integrations: intranet and document management, EHR launch points, identity provider.
- What makes it hard: outdated protocols. Content ownership and review dates matter as much as the model. See RAG development.
- Illustrative team and build: tech lead plus two senior engineers, three to four months: $35,000 to $45,000 a month, $105,000 to $180,000 in total.
How to pick your first use case
- Start administrative. Intake, coding assistance, denials and scheduling stay clear of device questions and show results in revenue cycle metrics your CFO already tracks.
- Sign the BAAs first. Confirm which model endpoints and cloud services are covered before engineering starts, not after a prototype works.
- Get clinical owners on the hook. Every clinical-adjacent system needs a named physician or nursing lead who signs off on the evaluation set.
- Decide what data the model really needs. Many workflows run on a narrow slice of the chart. Where you are building evaluation sets or analytics, de-identify under HIPAA's Safe Harbor or Expert Determination methods so engineers and reviewers can work without full PHI, and keep identified data in the production path only.
- Integrate early. EHR integration and security review usually take longer than the model work. Start that path in week one.
What it costs
Cost is team size times months times monthly rate. Ryz rates: mid-level engineers $7,000 to $10,000 per month, senior engineers $10,000 to $15,000, tech leads $15,000 and up, quoted per team.
- Revenue cycle pilot (tech lead plus two senior engineers) for four months: $35,000 to $45,000 a month, $140,000 to $180,000 in total.
- Clinical documentation pod (tech lead plus four senior engineers) for eight months: $55,000 to $75,000 a month, $440,000 to $600,000.
- Full pod of about seven senior engineers including a tech lead and an ML engineer: $75,000 to $105,000 a month.
Cloud and model usage run in your accounts and are not included. You meet the named engineers first, and a pod can usually start in under two weeks once you say go. Ryz is not the right fit if you want a packaged EHR add-on with its own models; we build in your stack and you own all code and IP.
Related
FAQ
What are the most practical AI use cases in healthcare today?
Ambient documentation, portal message triage, prior authorization assembly, coding assistance, denial appeals, eligibility checks, scheduling agents and fax intake. They are administrative or clinician-reviewed, which makes them faster to approve and easier to measure.
Is it HIPAA compliant to use large language models with patient data?
It can be, when the model provider and cloud services are covered by a business associate agreement, data use follows the minimum necessary standard and Security Rule safeguards are in place. HIPAA compliance is a property of how you build and operate the system, not of the model alone.
When does healthcare AI need FDA clearance?
Software intended to diagnose, treat or drive clinical management can be a medical device. FDA's clinical decision support guidance describes which functions are excluded, mainly those where a clinician can independently review the basis for a recommendation. Get regulatory advice for anything clinical.
How long does an EHR-integrated AI system take to build?
The illustrative ranges here run from two to three months for fax intake to six to eight months for ambient documentation. EHR integration approvals and security review often set the schedule.
Can Ryz engineers work in our HIPAA environment?
Yes. Ryz engineers have experience working within HIPAA and SOC 2 controls, work in your cloud and repos under your access policies, and you can revoke access at any time. Your compliance program and certifications remain yours.
Questions we didn't answer? Email info@ryzlabs.com.