How we saved over $30k/year for a Home Care Agency with a $20/month Claude Subscription

Small businesses usually assume the fix for a scheduling problem is more software or more staff. For one home care agency in Orange County, the fix was a $20 Claude subscription and a week of design work.

The agency provides in-home non-medical care and respite for older adults. Around 40 caregivers, a few hundred visits a week, and one care coordinator who owned the schedule. She was good at it, and she was drowning. The owner had budgeted a part-time scheduling assistant to absorb the overflow before they called us.

What We Evaluated First

We did not start with the tool. We started with three questions: where the hours were going, what the agency could spend, and what the data rules had to be.

The workflows

We sat with the coordinator for two mornings and logged what she did minute by minute. The pattern was consistent:

• Visit confirmations. Every client visit for the next day got a confirmation call or text, and every caregiver got a shift reminder. Most went unanswered the first time, so each one became two or three touches.
• Reschedules. A caregiver calling in sick at 6:30am set off a chain: find coverage, notify the family, update the schedule, and confirm the replacement. Each one took 20 to 40 minutes of phone tag.
• Inbound email. New inquiry forms, family members asking about next week, caregivers asking about hours, referral partners checking on a placement. The same fifteen or so emails, in slightly different words, every day.
• Follow-ups that never happened. Referral sources and past inquiries were supposed to get a check-in. They did not, because confirmations always came first.

Roughly three hours a morning went to confirmations and routine email alone. None of it required judgment. All of it required a person, because the agency management software could post a schedule but could not write a warm message, notice a non-reply, or answer a question.

The budget

The owner had priced two options: a scheduling platform add-on with messaging, at several hundred dollars a month plus per-client fees, or a part-time assistant at about $30,000 a year with benefits and turnover risk. Both were real money for a business with thin home care margins. The constraint we were given was simple: keep the existing agency management software, do not add a second system of record, and keep the recurring cost near zero.

The compliance needs

Whether a home care agency is a HIPAA covered entity turns on the definition in 45 CFR 160.103, and non-medical agencies often sit outside it. We did not build on that hope. A consumer Claude plan does not come with a Business Associate Agreement, so the design rule was that protected health information never enters the model at all. That decision shaped everything that followed, and it is the reason the $20 plan was viable.

What We Built With Claude Cowork

Claude Cowork is the part of the Claude Pro plan that runs multi-step tasks on a schedule, in the cloud, with connectors to Google Workspace. The agency already ran on Gmail and Google Calendar, and their scheduling software could export the next day's visits as a spreadsheet. That was enough.

Appointment setting and confirmation

Each evening the coordinator drops the next day's visit export into a shared Drive folder. The export was redesigned first: it carries client first name, visit window, caregiver first name, and preferred contact channel. Nothing else.

A scheduled Cowork task runs at 6am and does four things:

1. Drafts every client confirmation in the agency's voice, with the visit window and caregiver first name, into a review queue in Gmail.
2. Drafts every caregiver shift reminder with the address pulled from Calendar, where the caregiver already had access.
3. Flags non-replies from the previous day and drafts a second, shorter touch for each.
4. Prepares a reschedule packet for any visit marked open: a ranked list of caregivers with a matching availability block, and a pre-written message to each.

The coordinator opens the queue at 7:30am, reads, edits the handful that need a personal note, and sends. The morning that used to take three hours now takes about 20 minutes. When a caregiver calls in sick, she picks from the ranked list and sends messages that are already written, instead of starting from a blank phone screen.

Email automation

The second set of tasks lives in the inbox. Cowork watches a labeled queue and produces a drafted reply for the routine categories:

• New inquiries get a same-day response with the agency's intake steps and a link to book an assessment call, plus a follow-up draft three days later if there is no reply.
• Family scheduling questions get an answer drawn from the same minimal export, never from care notes.
• Caregiver hours and availability questions get a reply plus a Calendar hold.
• Referral partner check-ins get a status update and a thank-you, on a cadence the agency had never managed to keep by hand.

Anything that mentions a complaint, a change in a client's condition, medication, billing, or a health question is routed to a person with no draft at all. That routing rule is part of the written policy, not a setting someone can toggle off on a busy morning.

Writing the Rules Before Switching Anything On

The policy took an afternoon and it is one page. It names the fields allowed in the export, the message categories a person must handle, who approves the morning queue, and what happens when the coordinator is out. It is the same approach we describe in how to write an AI usage policy for a small business, and the reason it matters is the same: people move quickly when they have a clear answer instead of a judgment call.

For the broader question of what belongs in a general-purpose assistant, is ChatGPT safe for business data covers the reasoning we applied here.

The Result

The part-time scheduling role was never posted. The math the owner used is below, and it is deliberately conservative: it counts only the hire that did not happen.

Beyond the number, three things changed. Confirmations go out every day without exception, so no-shows and missed shifts dropped. Reschedules are handled in minutes instead of a morning. And referral partners hear from the agency on a schedule, which is the follow-up work that had always been pushed aside.

Our own fee was a one-time AI Strategy & Assessment plus two working sessions with the coordinator and owner. There is no ongoing charge to us built into the numbers above.

What to Take From This

If an owner is about to hire someone to keep up with confirmations and routine email, the job is probably a workflow problem dressed as a staffing problem. The steps that worked here transfer to almost any appointment-driven small business:

1. Log where the hours go before choosing a tool.
2. Keep the existing system of record and automate around it.
3. Decide what data the model may see, and make the export enforce it.
4. Draft everything, auto-send nothing, and route the sensitive categories to a person.

That sequence is the substance of our assessment work, and for teams that want the automations kept running as software and rules change, On-Going Support covers it. If you are weighing this against hiring, AI consultant vs. hiring an in-house AI engineer lays out the comparison.