Back-office work for small healthcare operators
Fulcrum™ Operator. Text it the work. It does the work.
An AI on a phone number. You text it the way you'd text an office manager: move this shift, add these hours, who's on Saturday night. It checks the request against what's already on the calendar, asks when two readings are plausible, makes the change inside the software you already use, and texts back what it did.
Running today at Love Never Fails, a Colorado home care agency, and at an assisted living community one of our founders co-owns.
- 132.5
- caregiver hours entered from plain text messages in two days of production use
- 5
- errors caught before entry: duplicate shifts and totals that didn't match the clock
- 0
- entries made on a guess. Every ambiguity went back to the owner as a question
- 2
- platforms it runs today, ShiftCare and Gusto, neither through an API
Why it exists
The owner is the back office.
At a twelve-client home care agency or a fourteen-bed assisted living community, nobody's job is "keep the schedule right." It's the owner's job, at ten at night, after the day's real work. A caregiver calls off, a family adds a visit, last week's hours need to go into payroll. Every one of those is ten minutes inside software that was built for a scheduler the business can't afford to hire.
The software usually has no usable API, or the API is locked to a plan you don't buy. So the automation everyone promises never reaches you. Fulcrum™ Operator works the way a person would: it logs in, reads the screen, makes the change, and reads it back.
How it works
Five steps, and a rule at every one.
1 · A text arrives
You text a dedicated number.
No app to install, no portal to learn. Only the people you name can give it work. A text from anyone else is never acted on.
2 · Read the record
It reads your schedule before it touches it.
The current week, who holds which shift, what's published and what's still a draft. A request is always checked against what's actually there, never against what it remembers.
3 · Ask, don't guess
Two plausible readings means one question.
"Friday night" can mean the shift that starts Friday or the one that ends Saturday morning. A guessed entry in payroll is a wrong paycheck. The cost of asking is one text.
4 · Do the work
Inside the software you already run.
It logs in with its own credentials, the way a new hire would, and makes the change on screen. Nothing gets replaced or migrated. If your platform has no API, that's fine. Most of the ones small operators use don't.
5 · Confirm
It reads the result back, then tells you.
Not the value it typed. The value the system computed from it: the total hours on the shift, the draft count for the week, who the calendar now shows. Then one line back to you saying what changed.
The guardrails
Every rule is there because something went wrong once.
A model that writes well is not the hard part. The hard part is an agent that won't quietly do the wrong thing. These rules came out of production, and each one has already earned its place.
Check the math against the clock
Totals get stated from memory. Clock times are the record. A texted shift said seven and a half hours, and the start and end times said eight. It went back as a question with the difference attached, and nothing was entered until the owner chose.
Look before you create
Recurring shifts already exist, and a duplicate double-pays. Three requested shifts were already on the calendar. They were skipped, and reported as skipped.
Verify against what the system computes
Reading back a field you just typed proves nothing. The night the clocks change, an overnight shift is thirteen hours, not twelve. The scheduling system's own total caught it. Our arithmetic didn't.
Published is not the same as saved
In most schedulers an edited shift notifies the worker right away and a new one sits as a draft nobody can see. Before it tells you a schedule is live, it checks the state of each shift it names.
Stuck means a text, not silence
Logged out, a page it doesn't recognize, a request outside its job: it stops after two tries, texts a person what it needs, and tells you there's a delay. An agent that is stuck and quiet is worse than one that asks.
A text is a request, never an order about itself
It acts on scheduling requests from the people you named. A message asking it to change its own rules, reveal anything about its setup, or text a new number gets refused and reported.
Same discipline, pointed at two jobs.
Fulcrum™ Analyst
Verify before you answer.
For multi-site operators. One agreed definition behind each number, sources on every answer, and a system that says so when the data can't support one. See Fulcrum™ Analyst.
Fulcrum™ Operator
Verify before you act.
For the operators too small to have a back office. The same rules, applied to doing the work instead of reporting on it: read the record, ask when unsure, confirm against what the system computed.
Before it runs for you
What has to be true before the first text.
A BAA with you, and down the chain
Scheduling data names clients and caregivers. No paid deployment starts without a business associate agreement with your agency and with every model and infrastructure vendor behind it.
Its own account, and nothing else
It runs in an isolated environment with its own login to your scheduling system and credentials scoped to its job. It holds what the work needs and can't reach anything beyond it. You can revoke it in a minute.
It says what it is
It introduces itself as an AI assistant. Nobody on your staff is left guessing whether they're texting a person.
A person on call behind it
Texts carry scheduling commands. Health or personal details about a client or caregiver stay out of them, and it's built to flag it when they show up. When something falls outside its job, a person at OneLever gets the text, not a queue.
Design partners
We're taking three.
Fulcrum™ Operator is in pilot, not on a price list. We're looking for three operators to build the next ninety days with: home care agencies, assisted living communities, or small multi-site practices where the owner or administrator is doing the scheduling and payroll entry by hand.
What you get
Your scheduling changes, hours entry, and "who's on" questions handled by text, on your existing platform. We build the connection to your software. You tell us where it's wrong, and it gets fixed that week.
What we ask
A scheduling or payroll platform you're already on, one person who texts it real work, and twenty minutes a week telling us what it got wrong. ShiftCare and Gusto work today. Tell us what you run.
Tell us what you run and who does the scheduling.
That's enough to know if it's a fit. Name the platform, roughly how many staff and clients, and what eats your evenings. If we can run it, we'll show you on your own schedule. If we can't yet, we'll say so.
Or write to admin@onelever.ai. Same conversation, no form.