Automation for the home health office
The work that eats a home health office’s week is repetitive, rule-bound and easy to get wrong. We built eight systems for it in a working California agency. A person still makes every decision; the systems do the reading, matching and chasing.
Eight office tasks, and what now does them
Each card says whether the system is running in production or built and held. Try six of them in the demo, with invented data.
Checking every caregiver’s credentials
A weekly audit compares each caregiver’s requirements with the personnel files and rebuilds a color-coded board. When a caregiver e-mails a photo of a renewed card, it is read, and filed only after the office confirms it.
Running in production Read the case studyRoute sheets to payroll
Paper and photographed route sheets are read, matched to the visits in the EMR and filed. Payroll gets a short list of exceptions instead of a stack of paper.
Running in production Read the case studyTherapy vendor invoices
Each billed visit is matched to a visit note. The office gets one report: ready to pay, wait, already paid, or needs a person.
Running in production Read the case studyVoicemail to task
A caller picks who they are from the phone menu. An unanswered call becomes a transcript and a task with a due time.
Running in production Read the case studyPolicy and HR questions
Two assistants in Teams answer only from the agency’s own documents and cite the source; a weekly watcher flags when a saved regulation changes.
Running in production Read the case studyProving it ran
Recurring jobs are buttons on a Teams page, and every run leaves a record nobody can edit, ready to show an auditor.
Running in production Read the case studyPhysician orders
Loose orders awaiting signature are identified by patient and filed under the physician of record from the EMR.
Built and verified, switched off on purpose Read the case studyAdmission paperwork
A start-of-care packet scanned at the office printer is matched to the right patient and split into one document per form.
Built and verified, switched off on purpose Read the case studyMost of this turned out to be less of an AI problem than it looked. Credential tracking is arithmetic against a folder listing. Invoice checking is tables and matching. A model earns its place only where a person used to read handwriting or a photo, and even there it only proposes.
Before you automate anything
Which one should we start with?
The one your office spends the most hours on, or the one an auditor would ask about first. For most agencies that is credentials or route sheets.
What stays with a person?
Every decision. The systems read, match and list; a person approves a filing, decides an exception and chooses what is paid. Two of the systems above are built and verified but switched off on purpose until the agency is ready to hand them that work.
Do we have to change our EMR?
No. Everything here works alongside the agency’s existing EMR and Microsoft 365.
Is AI reading our patient records?
Not in public AI tools, ever. Where AI touches PHI, it runs inside your Microsoft 365 tenant under your agreement, or on encrypted hardware you control, and a person confirms the result. Most of what we build uses no AI at all.