Home Care Is Broken

I spent the last couple of years building a startup in home care. I went in thinking the problem was documentation. Documentation turned out to be the part I could see from outside.

Almost 90% of adults 50 to 80 want to stay in their homes as long as possible. In 2025, around 11k Americans turned 65 every day, the most on record. The industry meant to support them splits into three sectors: personal care, home health, and hospice. Each has its own labor problem. All three run on the same software.

Personal care

Personal care is the most basic tier. Caregivers help seniors with daily activities: bathing, dressing, meal prep, companionship. These are the people who show up at someone's home every day to make sure they're safe and fed.

A caregiver makes about $35k a year, roughly $17 an hour. At the agencies I worked with, caregivers drove between clients on their own time, so the effective rate was even lower. The work is physically hard and emotionally draining, and there's almost no ladder to climb.

Caregiver turnover across the industry is nearly 80%, up more than 12 points in two years. At that rate an agency with 100 caregivers hires and trains roughly 80 replacements a year. For the senior on the receiving end, that's a rotating cast of strangers coming into their home, each one learning their preferences from scratch. Continuity is the thing that makes home care work at all, and it's close to impossible to maintain.

In the offices I sat in, administrative staff spent more of the week on recruiting and onboarding than on anything to do with care quality. You can't build a stable workforce at $17 an hour minus unpaid drive time. And agencies can't pay more by charging more, because Medicaid sets the reimbursement rate for a large share of these hours. I don't have margin figures for the agencies I worked with. The owners described their margins as thin, and none of them believed they could raise wages without a rate increase.

Home health

Home health is a step up in clinical complexity. Nurses, physical therapists, and occupational therapists visit patients at home after surgery, during recovery, or for chronic condition management. The work doesn't end when they leave the patient's home.

Every visit generates documentation. The largest single piece is the OASIS assessment, a 40+ page standardized document CMS requires for reimbursement. Miss a field and you don't get paid. Fill one out wrong and you get compliance problems or audit flags.

Clinicians drive from home to home all day and get paid only for the time spent with patients. So the clinicians at the agencies we worked with did their documentation at night, after the last visit. Several told me they were still writing OASIS assessments and SOAP notes near midnight.

I wrote about this in my post on browser agents for EMR automation. The documentation burden is what led us to build Northlight in the first place. But the longer I spent in the industry, the more documentation looked like a symptom of something that runs through all three sectors.

Hospice

Hospice is end-of-life care. Hospice clinicians and volunteers sit with people who are dying. They manage pain, provide comfort, and support families through the worst weeks of their lives.

The same documentation burden as home health sits on top of that work. And volunteers aren't optional: federal regulation requires hospices to staff volunteers for at least 5% of total patient care hours. The hospice coordinators I spoke to called recruiting and keeping those volunteers their hardest staffing problem. I don't have turnover or burnout numbers for hospice volunteers.

The staff who left described the same combination in every conversation I had: a day spent helping someone die with dignity, then regulatory paperwork in the evening.

The fax problem

Agencies get patients through referrals, and the referrals come by fax. When a patient is discharged from a hospital and needs home health or hospice, the hospital faxes over pages of clinical history, medication lists, and physician orders.

In 2026, agency staff still read each fax by hand, decide whether they can accept the patient, and respond. The intake teams I worked with couldn't get through the day's referrals in the day, because there weren't enough people to triage them. By the time someone reviewed a referral, another agency had accepted it or the patient had been placed elsewhere.

CMS has rules pointing the other way. A 2019 rule requires hospitals to run a real discharge planning process, give patients their post-acute options, and support "seamless exchange of patient information between health care settings." Seven years on, it hasn't changed how referrals arrive.

The technology problem

For every software team I watched try to sell into this market, including my own, the legacy EMRs were the thing that stopped them. The market figure everyone quotes is $130+ billion growing at double digits. I've never been able to trace it to a source I trust, so treat it as folklore.

A handful of vendors dominate home care EMRs, WellSky the biggest. These systems were built in the early 2000s, designed for a pre-API world, and scheduling, documentation, billing, and compliance all run through them.

The systems I looked at had no public APIs, no webhooks, and no integration layer worth mentioning. The vendors have captive customers who are afraid to switch, because migration means months of disruption and possible compliance gaps. Since switching is expensive, opening up buys the vendor nothing. The vendors we approached had no documented API and treated any integration as a paid custom engineering project priced case by case.

Every vendor we approached wanted a revenue share as the price of an integration partnership. We didn't sign one. After that, agency staff told us their EMR vendor had described outside tools like ours as a security risk and an unsupported integration, and our conversations with those agencies stopped. I can't prove the vendor caused it, but that was the sequence.

A better scheduling tool needs data out of the EMR. Automated documentation has to write back into it. Care quality analytics need the records that live inside it. Every improvement has to pass through a vendor that doesn't want you there.

That's why home care technology is a decade behind hospital technology. Acute care went through its EHR transition, forced by Meaningful Use regulations and billions in federal incentives. Post-acute care was left out, and the vendors who filled the gap built closed systems.

Where integration fails

The AI for three specific tasks here already exists. OCR and document intelligence can parse a faxed referral into structured fields, triage it, and rank it against an agency's capacity. Background agents can listen to a clinician's visit, structure the conversation into clinical language, and fill out the OASIS assessment, which is the approach I described with browser agents. Voice AI can handle caregiver scheduling calls, shift check-ins, and training modules.

Each one has to land its output somewhere. The referral data goes into the scheduling module. The OASIS assessment populates a specific patient record. The caregiver check-in updates the care plan. All of those live in WellSky or whatever legacy system the agency runs, and that's the step where the product stops.

You can build a genuinely good AI product for home care and lose at the integration layer. Of the teams I've watched try, the ones that got in did it through a revenue share or by being acquired. I wouldn't claim there's no third path, only that I haven't seen one work.

Three things could shift that: regulation forcing interoperability the way Meaningful Use did for hospitals, startups routing around the EMRs with tools like browser agents, or agencies collectively demanding more from their vendors. The current arrangement holds only as long as agencies keep paying for software they can't connect anything to.

Why this matters now

The number of Americans 65 and older is on track to nearly double to 95 million by 2060. Turnover near 80% and pay around $17 an hour describe a workforce that can't be staffed at the wages currently on offer, and demand is rising against that.

Technology is one piece of this. Labor economics, Medicaid reimbursement rates, and a handful of EMR vendors set the boundaries. Of those, the closed EMRs are the constraint a startup can attack directly, and that's where I'd spend the effort.