Home Health EHR Automation: Wellsky, HCHB, and Kinser

Home health is one of the fastest-growing segments in healthcare. Agencies coordinate care for patients at home, manage referrals from hospitals and physicians, schedule visits, and document everything for billing and compliance. The systems they use are desktop EHR applications: Wellsky, Home Care HomeBase (HCHB), Kinser, and others. None of them offer writable APIs for third-party integration.

AI companies building products for home health face the same integration wall. Care coordination platforms, referral management tools, intake automation, and documentation assistants all need to read and write data in these EHR systems. The data has to flow both ways. Referrals come in. Patient records get created. Visits get scheduled. Notes get documented. Without integration, the AI product is a standalone tool that creates more manual work instead of less.

The Core Workflows

Referral intake. A referral arrives from a hospital or physician. It contains patient demographics, diagnosis, orders, and sometimes clinical notes. The intake workflow: create a patient record in the EHR, enter the referral details, attach documents, route for clinical review. Manual intake takes 15 to 30 minutes per referral. High-volume agencies process dozens per day. The EHR has no API to create patients or upload referrals. The only path is through the GUI.

Scheduling. Once a patient is admitted, visits need to be scheduled. The workflow: find available caregivers, match them to the patient's needs and geography, create visit records in the EHR, avoid conflicts. The EHR holds the schedule. The AI might optimize assignments or suggest slots, but the actual creation of visits happens in the desktop application. Again, no API.

Documentation. After each visit, the clinician documents what happened. For AI-assisted documentation, the model generates a note from the visit data. The note needs to be entered into the EHR in the correct format, in the correct chart, with the correct billing codes. Manual entry is error-prone and time-consuming. Automated entry requires writing to an application that does not accept programmatic input.

Why Traditional Approaches Fail

API integration is not an option. These systems do not expose the endpoints needed for referral creation, scheduling, or documentation entry. Some offer read-only APIs for reporting. None offer the write access that AI products need.

Direct database access is theoretically possible but practically wrong. It bypasses application logic, breaks audit trails, and violates compliance. Home health is Medicare-regulated. Auditors expect data to flow through the application. Writing directly to the database is a compliance risk and a support nightmare.

Traditional RPA has been tried. Selector-based bots work until the vendor pushes an update. Home health EHR vendors update their applications regularly. Each update breaks bots. Agencies running multiple EHR instances (different versions, different configurations) multiply the maintenance burden. The maintenance cost of traditional RPA makes it unsustainable for AI companies deploying across many customers.

How Computer Use Agents Handle It

Computer use agents interact with the EHR through the visual interface. They see the screen, navigate menus, fill fields, and verify results. No selectors. No brittle scripts. When Wellsky, HCHB, or Kinser pushes an update that moves a button or reorganizes a form, the agent adapts by finding elements visually.

For referral intake, the agent receives structured referral data, logs into the EHR, creates the patient record, enters the referral details across the required screens, uploads documents, and routes for review. The workflow matches what a human would do, but at scale and without manual intervention.

For scheduling, the agent navigates to the scheduling module, finds available slots, creates visit records with the correct caregiver and patient, and verifies the entries. The logic for matching and optimization can live in the AI product. The write-back to the EHR happens through the agent.

For documentation, the agent takes the generated note, navigates to the correct patient chart and visit, enters the note in the required format, and ensures billing codes are applied correctly. HIPAA compliance for the automation infrastructure is non-negotiable when handling patient documentation.

The Vertical Reality

Home health agencies are not replacing their EHRs. These systems are entrenched. Migration timelines are years. AI companies that want to serve this vertical need to integrate with the systems that exist today. Computer use agents are the only approach that can read and write to Wellsky, HCHB, Kinser, and similar desktop EHRs at scale without requiring vendor cooperation or breaking on every update.

Configuration Variability

Home health EHRs are heavily customized. Each agency configures templates, fields, and workflows differently. A referral intake form at one agency may have different required fields than the same form at another. The scheduling module may present options differently. Documentation templates vary by agency and by payor.

Selector-based automation cannot handle this variability. A bot built for one configuration breaks on another. Computer use agents handle it by understanding the visual layout. The agent finds the "Patient Name" field whether it is in the first column or the third. It navigates to the scheduling section whether the menu structure is flat or nested. This visual flexibility is what allows a single automation to work across multiple customer deployments without building a custom bot per customer.

Latency and Throughput

Referral intake in home health is time-sensitive. A referral that sits for hours may mean a lost admission. The automation needs to complete in minutes, not hours. Computer use agents add per-action latency compared to selector-based clicks, but the difference is measured in seconds per action. For a typical referral intake of 20 to 40 actions, the total time is still well under what manual entry would take. The throughput requirement is met. The reliability requirement, which traditional RPA often fails, is the one that matters more.