AI
AI scheduling for home health: what it can do, and what it must never decide.
Almost every scheduling product now says it uses AI. Three different things hide under that word, each fails in its own way, and some decisions belong to a person whatever the software can do.
Almost every scheduling product for home health now says it uses AI. For a nurse, therapist, aide, hospice clinician or agency scheduler whose day is a list of addresses, that word covers very different things: a genuine help, a relabelled feature that has existed for decades, or a system making decisions that a person should be making. This is a plain guide to what AI scheduling for home health can actually do today, what it must never be allowed to decide, and how to judge a vendor's claim.
Three different things are all called "AI"
| What it is | What it does well | Where it goes wrong |
|---|---|---|
| Route and schedule optimization | Orders a day or a week around time windows, visit frequency and the drive. This is operations research, decades old and well understood; calling it AI is marketing, but it is the part that saves hours. | It optimizes what it is told. A window nobody entered, or a preference it treats as a rule, produces a confident, wrong day. |
| Language models | Understand a spoken or typed request ("move Rosa to Thursday"), and read a messy referral fax into fields. | They can produce text that sounds right and is not: a name that almost matches, an address that was never on the page. |
| Prediction | Estimates things like drive time, visit length or the chance of a cancellation from past data. | A prediction is an average. It is useful for planning and wrong for any single Tuesday. |
Knowing which of the three a vendor means is the first question, because each fails in its own way and needs its own safeguard.
What it can genuinely do
- Order the day. Given addresses, windows and visit lengths, a solver finds an order that drives less and keeps the windows better than most hand-planned days. The method it improves on is in how to plan a home health route by hand.
- Re-plan after a change. A cancellation at 8:40 is a new problem, and a solver can answer it in seconds while you are parked. What to do with the freed hour is covered in cancellations in home health.
- Turn a referral into a draft. Reading a faxed or photographed referral into name, address, phone and frequency saves typing, if every field is shown for review. The cost it attacks is set out in what a referral fax actually costs you.
- Take a request by voice. "Call my next patient", "what time is Beatriz", "push the 2 o'clock to tomorrow" — asked from the driver's seat without typing. More in the voice scheduling app.
What it must never decide
Anything clinical
Visit frequency, which discipline goes, whether a patient is seen today or can wait: these belong to the plan of care and the clinician. A scheduler that quietly drops a visit to make the day fit is making a clinical decision without the training to make it.
A time a patient has been told
An agreed time is a promise to a person. Software may suggest times, but once one has been confirmed it should never move without a warning and a human choice. Why a suggested time and a confirmed one must look different is in confirming home health visit times.
Which patient a request is about
Acting on the wrong patient is the worst failure a voice or text assistant can have. When two people answer to one name, or a heard name is only close to one, the right answer is a question, never a guess.
Anything that commits without you
Every change to the schedule should arrive as a proposal — this is what I will do — and happen only after you confirm it. A rule in the software, not an instruction to the model, is what makes that true.
How to judge a vendor's AI claim
- Which kind of AI is it? Optimization, language model or prediction. "All of them" is fine, but each needs its own answer below.
- Does it propose or does it act? Ask to see what happens when you say no.
- What happens when it is wrong? Ask for a demo with a smudged fax and a mumbled name, not the clean example.
- Is every extracted value on the page? A field the model filled in from nothing should be empty, not plausible.
- Where do audio, photos and patient data go? Ask which vendors see them, whether a business associate agreement covers each one, and whether audio leaves the phone at all. HIPAA obligations and state privacy law vary; your compliance team, not the sales deck, is the authority.
- Why did it do that? A dropped visit should come with a reason you can act on, such as a window that cannot be reached after the visit before it.
If you are comparing tools, our honest comparison of home health scheduling apps covers the agency systems, and what home health scheduling software has to do makes the wider case. The Windshield Tax calculator puts a figure on the driving any of them should be reducing.
Questions people actually ask
What is AI scheduling in home health?
It usually means one or more of three things: an optimizer that orders visits around time windows and the drive, a language model that understands spoken or typed requests and reads referrals, and predictions of drive time or cancellations. They fail differently, so ask a vendor which one it means.
Can AI replace a home health scheduler?
Not the judgement. Software can order a day, re-plan after a cancellation and take requests by voice, but clinical priorities, promises made to patients and the question of which patient a request is about need a person. The useful version proposes and a person confirms.
Is route optimization the same as AI?
Not strictly. Route optimization is operations research that predates today's AI; vendors often call it AI anyway. It is still the part that saves the most driving, so the label matters less than whether it holds your time windows.
Is AI scheduling software HIPAA compliant?
It depends on the vendor, how data flows and whether business associate agreements cover every service that sees patient data. Ask where audio, photos and records go. HIPAA and state rules vary; check with your compliance team rather than relying on marketing. Obillions makes no HIPAA compliance claim.
Can AI read a referral fax accurately?
It can read most typed referrals into fields and saves a lot of typing, but it makes mistakes on poor scans and handwriting. Every field should be shown for review, and a value that is not on the page should be left empty rather than guessed.
What should an AI scheduling assistant never do on its own?
Change anything without your confirmation, move a time a patient has agreed to, guess between two patients with similar names, or drop a visit to make a day fit without telling you why. Those are decisions for a person.
Obillions
A voice agent for a day of visits. You say what changed and the day re-plans itself: the order, the drive times, and who nearby could take the hour a cancellation just freed. Nothing changes until you confirm, and your voice and photos never leave your phone.
In private testing now. A public TestFlight comes next, then the App Store.