Healthcare AI is often discussed as though it were one market. It is not. A dental practice recovering unscheduled treatment, a gastroenterology clinic processing hundreds of referrals, a health system answering patient calls and a surgical practice monitoring recovery are solving fundamentally different operational problems. They use different systems, data models, billing rules, patient journeys and levels of clinical oversight.
Yet these markets are converging around one broader idea: healthcare organizations are moving from isolated automation tools toward a persistent coordination layer that helps patients progress through their entire care journey.
Where most vendors enter today
The destination is a continuous patient concierge: a system that understands where each patient is in their journey, knows what should happen next, communicates through the appropriate channel, completes administrative actions and brings in a human whenever clinical judgment or personal intervention is required. The industry is not there yet, but nearly every important category in healthcare automation is moving in that direction.
Executive summary
The healthcare automation market can be understood through four layers: systems of record (PMS and EHR), integration infrastructure (normalized data access), workflow applications (individual operational problems) and journey orchestration (workflows coordinated around the patient rather than around the software).
-
1
Systems of record
PMS and EHR hold authoritative schedule, clinical and financial data
-
2
Integration infrastructure
APIs and synchronizers normalize fragmented healthcare data
-
3
Workflow applications
Voice agents, referrals, prior auth, recall and intake automation
-
4
Journey orchestration
Multiple workflows coordinated around one patient over time
The market remains fragmented because each specialty has different clinical data, appointment structures, referral processes, insurance requirements, treatment pathways, documentation standards and patient communication needs. That explains why dental, behavioural health, physical therapy, chiropractic, ophthalmology and general medical practices frequently use different software platforms.
Point automation
Reminders, calls, forms and individual tasks
Workflow automation
Referrals, prior auth and treatment follow-up
Journey orchestration
Multiple workflows around one patient
Continuous concierge
Persistent coordination across the relationship
Healthcare’s digital foundation: PMS and EHR
Before understanding healthcare AI, it is necessary to understand the systems underneath it. New AI platforms usually do not replace the system of record. They sit above it and attempt to complete work inside or around it.
Practice-management system (PMS)
- Scheduling and provider availability
- Insurance, claims and patient balances
- Payments and appointment status
- Recall lists and operational reporting
Electronic health record (EHR)
- Diagnoses, notes and treatment plans
- Medications, allergies and test results
- Procedure history and clinical documents
- Longitudinal clinical record
According to the Office of the National Coordinator for Health IT, 95% of U.S. office-based physicians used some form of EHR in 2024 and 91% used a certified EHR.1 Adoption has increased substantially from 2008, when 42% reported using any electronic record system.2 Widespread adoption does not mean healthcare data has become easy to access. It means most information is now digital, but distributed across thousands of implementations, configurations, databases and workflows.
Why healthcare software remains fragmented
Healthcare specialties do not simply use different templates. They operate through different data models and care pathways. There is no legal rule requiring every specialty to use specialty-specific software. The decision is operational: a general platform can sometimes be configured for several specialties, but a specialty product frequently supports the relevant workflow more naturally.
Dental
Physical therapy
Behavioural health
Specialty medical
This has major implications for healthcare AI. A company cannot simply build one integration and assume it can serve the entire market.
The emerging healthcare automation stack
Layer 1: Systems of record
These are the PMS and EHR platforms where the practice’s authoritative information lives: Dentrix, Eaglesoft, Open Dental, athenahealth, eClinicalWorks, Epic, Oracle Health, NextGen, Jane, SimplePractice and many others. They hold the schedule, patient profile, treatment information, insurance data or clinical record. They are difficult to replace because years of operational and clinical history are stored inside them.
Layer 2: Integration and data infrastructure
Companies in this layer solve a different problem: how can a new healthcare application communicate with many different PMS and EHR platforms without building every connection independently?
NexHealth
Developer infrastructure and practice-facing applications. Capabilities vary by underlying PMS.
Sikka.ai
Universal translation layer across dental, veterinary, physician, chiropractic and optometry markets.
Layer 3: Workflow applications
These companies automate one operational problem or a related group of problems.
Layer 4: Journey orchestration
Earlier question
- What can we automate in the call centre?
- Which task can this agent complete?
- How do we reduce hold time?
Orchestration question
- Where is this patient in the journey?
- What is preventing progress?
- What should happen next?
Major workflow-automation markets
Patient access and voice agents
This category usually begins with inbound communication. The workflow is visible and easy to demonstrate: a practice can call the agent, test it and immediately understand the value of fewer missed calls, shorter hold times, after-hours coverage and more completed bookings.
Answer
Identify caller and intent
Resolve
Search availability and verify insurance
Complete
Book, reschedule or route escalation
Write back
Update PMS or EHR
Referral automation
Referral management is a different operational category. A single broken step can prevent the patient from receiving care.
Intake
Fax or portal document
Review
Classify and collect missing information
Authorize
Verify insurance and obtain approval
Schedule
Outreach, booking and provider notification
Prior-authorization automation
Prior authorization remains one of healthcare’s most labour-intensive administrative workflows. The 2025 AMA Prior Authorization Physician Survey reported an average of 39 prior-authorization requests per physician per week and approximately 13 physician and staff hours spent on them; 40% of surveyed physicians reported staff who worked exclusively on prior authorization.8
Patient recall and reactivation
Patient recall is not one workflow. It contains several different markets, from routine preventive recall to conversational treatment coordination.
Routine recall
Inactive reactivation
No-show recovery
Unscheduled treatment
Case acceptance
Traditional recall
- Asks whether the patient would like to book
- Campaign-level messaging
- Limited objection handling
Treatment coordination
- Understands why the patient did not proceed
- Addresses insurance, fear and timing
- Closes the loop through scheduling
Administrative workflow automation
Companies such as TriFetch begin with high-value queues rather than a general patient concierge. TriFetch emerged from stealth in April 2026 and announced a $1.9 million pre-seed round on its website.16 Its approach illustrates an important market pattern: start with a queue that already exists, is already expensive and already delays revenue.
Queues worth owning first
Referral backlog
Prior-authorization queue
Unanswered-call queue
Unscheduled-treatment list
Cancellation list
Pre- and post-care coordination
Patient communication does not end when an appointment is booked. Commure’s acquisition of Memora Health brought a digital care-navigation platform into a broader healthcare AI portfolio.24 Hyro demonstrates proactive care-gap outreach, appointment offering and patient-record updates.23 These applications are closer to the long-term concierge model because they remain involved after the initial scheduling event.
Before care
- Preparation instructions and forms
- Transportation and fasting guidance
- Medication confirmation
- Financial and logistical questions
After care
- Recovery instructions and medication reminders
- Symptom check-ins and follow-up scheduling
- Rehabilitation adherence
- Escalation to a clinician when needed
Representative market landscape
The following table is illustrative rather than exhaustive. These companies may enter through different workflows, but they are gradually converging on the same position: the execution and orchestration layer above the EHR.
| Company | Initial market wedge | Direction of expansion |
|---|---|---|
| NexHealth | Scheduling, forms, messaging and PMS connectivity | Campaigns, recall, developer infrastructure and broader patient experience |
| Sikka.ai | Normalized dental and retail-healthcare data | Governed autonomous practice operations |
| Codent AI | Dental treatment-plan follow-up | Case acceptance and ongoing treatment coordination |
| TriFetch | Referrals, prior authorization and specialty-clinic administration | End-to-end front-, middle- and back-office automation |
| Tennr | Referral and document processing | Patient orchestration across payer and provider workflows |
| Assort Health | AI voice and patient access | Referrals, intake, scheduling and follow-up across the journey |
| Notable | Enterprise healthcare workflow automation | AI-agent workforce across access, revenue cycle and care operations |
| Artera | Patient communication | Cross-channel AI orchestration with human escalation |
| Luma Health | Patient access and engagement | Operational AI connecting access, intake, follow-up and payments |
| Commure/Memora | Care navigation and patient engagement | Broader clinical, operational and financial automation |
| Hyro | Conversational patient access | Proactive outreach, care-gap closure and post-visit workflows |
Company descriptions reflect each vendor’s published product positioning as of July 15, 2026. See company and product sources below.
Artera describes connecting phone, text, email and webchat into one patient conversation and coordinating AI agents with human teams when necessary.18 Luma Health frames its strategy as operational AI that moves from automation to orchestration across access, intake, engagement and payments.19 Notable documents AI agents across patient access, revenue cycle, care operations and contact centres.22
Why narrow workflows are winning today
The long-term concierge vision is broad, but current companies usually enter through a focused problem. There are four practical reasons.
-
1
The value is measurable
Calls answered, referrals processed, slots filled, treatment recovered
-
2
The workflow can be constrained
Allowed actions, escalation rules and success metrics are definable
-
3
Clinical risk stays limited
Administrative work first; FDA CDS guidance shapes clinical expansion
-
4
Integrations can be scoped
A referral product needs fewer objects than a universal concierge
Referral product scope
- Fax intake and document upload
- Patient creation
- Insurance verification
- Scheduling writeback
Universal concierge scope
- Clinical history and treatment plans
- Payments, referrals and prescriptions
- Recovery protocols and preferences
- Full communication history
Why the continuous concierge is still the likely destination
The fact that companies begin narrowly does not mean the broader vision is incorrect. It means the industry is building toward it incrementally. A patient does not experience healthcare as separate software modules. The patient experiences one relationship.
Problem
Find a provider and ask questions
Access
Book, prepare and attend
Decision
Receive recommendation and decide whether to continue
Completion
Complete treatment, recover and return when needed
The patient is the same, but the context is repeatedly lost between systems. The next generation of platforms will try to preserve that context. A continuous patient concierge would know why the patient contacted the practice, what care they are seeking, which appointment is relevant, what instructions they received, whether forms are missing, why treatment did not proceed, which channel they prefer and when they should not be contacted.
Its purpose would not be to communicate constantly. Its purpose would be to ensure continuity.
What the future concierge will actually do
The future product is unlikely to be one chatbot with unrestricted access to every part of healthcare. It will be an orchestrated system with multiple specialized capabilities across the care journey.
Before the appointment
After the consultation
After a missed visit
Before a procedure
After a procedure
During ongoing care
The human does not disappear
A recurring mistake in healthcare AI is assuming that escalation is an edge case. In reality, many high-value workflows depend on human judgment. Some patient questions are administrative. Some require a trained treatment coordinator. Some require a clinician.
Questions patients actually ask
“Why do I need this treatment?”
“Is this definitely covered?”
“Why is my cost different from the estimate?”
“Is this symptom normal?”
“Should I stop taking this medication?”
AI handles volume and continuity. Humans handle judgment, empathy and exceptions.
After the human responds, the agent can continue coordinating the journey.
Compliance and safety become product architecture
Healthcare automation cannot treat compliance as a certification badge added after development. When an AI vendor creates, receives, maintains or transmits protected health information on behalf of a covered healthcare organization, it may operate as a HIPAA business associate.11 Cloud and subcontractor arrangements generally require appropriate business-associate agreements, safeguards and risk-management processes.5 Automated calls and texts also create separate communication-law obligations; FCC rules strengthen patients’ ability to revoke consent through reasonable methods.7 This section presents a governance framework, not legal advice.
-
1
Administrative actions
Scheduling, forms, directions, reminders and approved educational materials
-
2
Protocol-controlled actions
Structured check-ins, warning-sign detection and escalation routing
-
3
Human or clinical judgment
Diagnosis, treatment changes, emergencies and complex disputes
The real challenge is not conversation, it is orchestration
Language models have made it much easier to produce natural conversations. That does not mean the healthcare workflow is solved. A reliable healthcare agent must coordinate patient identity, data permissions, appointment rules, provider availability, insurance requirements, communication history, human availability, escalation deadlines, PMS or EHR writeback and outcome tracking.
Easier problem
- Can an AI send a convincing message?
- Can it answer a general question?
Hard problem
- Take the correct action in the correct system
- For the correct patient under the correct rules
- And detect when it should stop
Why integrations still fail in practice
Incomplete or outdated data
Inconsistent entry between locations
Readable but not writable objects
Significant synchronization latency
The market will remain niche-specific
The underlying models may become general, but the workflows will remain specialized. The likely winners will combine horizontal infrastructure with deep vertical knowledge.
Dental coordinator
GI referral agent
Ophthalmology agent
How founders should enter the market
Weak starting point
- Build the AI operating system for all healthcare
- Promise to automate the entire relationship
- Start without measurable workflow ownership
Strong starting point
- One specialty and one patient cohort
- One broken workflow and one measurable next step
- Human escalation and reliable writeback first
The best early products do not merely report that a problem exists. They take responsibility for progressing the workflow.
How healthcare practices should evaluate AI automation
Practices should look beyond a polished demo. A serious evaluation should examine integration depth, workflow completion, human involvement, clinical boundaries, compliance and economics.
Data and integration
Workflow completion
Human involvement
Clinical boundaries
Compliance
Economics
What the industry is likely to look like next
-
1
Voice becomes a feature
Combined with text, email, forms, scheduling and human handoff
-
2
Response → proactive action
Monitor referrals, no-shows, care gaps and authorization status
-
3
Patient-level memory
Each patient treated as an evolving case, not a campaign segment
-
4
Human-in-the-loop remains central
Safe escalation, full context and workflow resumption
-
5
Outcome pricing grows
Appointments recovered, referrals completed, treatment finished
-
6
Infrastructure and apps overlap
NexHealth, Sikka and EHR vendors compete and enable simultaneously
-
7
Concierge emerges from connection
Independent workflows share context until the patient feels one relationship
The central opportunity
The largest opportunity in healthcare automation is not simply reducing labour. It is preventing patients from disappearing between steps.
Where patients are lost
A call is missed
A referral is not processed
An authorization stalls
Treatment is not followed up
A cancellation is never recovered
Responsibility moves between staff and systems
Every patient should feel that someone from the practice is looking after their next step, even when they are not physically inside the practice.
Conclusion
Healthcare automation is moving through a predictable progression. The first generation digitized records. The second automated individual tasks. The third is automating complete workflows. The next generation will connect those workflows around the patient.
The broad concierge vision should not be dismissed because it is difficult to deliver today. Difficulty is not evidence that the direction is wrong.
The continuous patient concierge is not the best starting point for most healthcare AI companies, but it is increasingly becoming the destination.
The companies most likely to reach it will begin by solving one broken, measurable and specialty-specific workflow extremely well, then connect the next workflow until the practice no longer sees a collection of automations and the patient no longer experiences a collection of disconnected systems. Instead, the patient experiences one continuous relationship with the organization responsible for their care.