AI Agent for Home Care Agencies: The Enquiry That Arrives on Discharge Day

An enquiry to a home care agency almost never originates at the agency. It originates in a hospital ward on the day a family is told that dad is discharged the day after tomorrow and cannot be left alone, or in a conversation between siblings after another fall at home. The person searching is not the patient but a son or daughter in their forties or fifties, at work, who got three phone numbers from a social worker or a neighbourhood WhatsApp group and sent all three the same short message. The mechanism that decides who wins the client is brutally simple: who answered, how fast, and how much order there was in the answer. None of those three has anything to do with the quality of care you actually provide.

The enquirer is the adult child, the channel is WhatsApp, and the hour is not office hours

Three facts about that enquirer explain nearly everything that follows. First, they cannot talk. They are standing in a ward corridor, sitting in a meeting or driving, so they write. An enquiry answered only with a callback gets an unanswered call, and there is no second attempt.

Second, they write when the office is closed. Ten at night, after the kids are asleep and the decision can no longer be postponed, or on Saturday morning when the siblings finally talk to each other. This is not a discipline problem in your team, it is a structural mismatch between enquiry hours and answering hours.

Third, and most important: they do not know the sector. They do not know the difference between a live-in caregiver and an hourly worker, and they are not sure whether the long-term care benefit is money coming to them or hours of care. So the first answer has to do two things at once: collect information, and explain in two lines which system they are even standing in. The agency that does that first turns from one of three suppliers into the party leading the process.

Above all, a very Israeli layer: a substantial share of the elderly population and of their children are Russian speakers, and the enquiry arrives in whichever language it is easiest to ask for help in. An agent that answers in Hebrew, Russian and English at equal quality removes a real barrier here, not just an inconvenience. We covered that behaviour in our guide to a multilingual AI agent in the Israeli market.

What actually has to be established on the first enquiry

Intake in home care is nearly identical every time, which makes it an obvious candidate for leaving the coordinator's personal phone. This is the list that determines whether anything can be offered at all:

• Who the patient is and their functional state. Mobility, whether they stand up unaided, whether lifting is required, whether there is cognitive decline or dementia.
• What happened now. Discharge from hospital, a fall, gradual deterioration, or a previous caregiver who left suddenly.
• When it needs to start. Tomorrow morning, in a week, or planning ahead for next month.
• Scope. Hours per day, days per week, night shift, sleep-in, or a live-in caregiver.
• Where. City and neighbourhood, floor, whether there is a lift. That determines who will agree to work there at all.
• The language of the household. Hebrew, Russian or English. In home care this is not a preference, it is a condition for daily functioning.
• Caregiver gender preference. Asked in almost every case, and better asked upfront than at the home visit.
• How the household lives. Kashrut, smoking, pets, cooking, other family members present.
• The funding picture. Whether a long-term care benefit application has been submitted, whether there is private insurance, whether the family is also preparing for supplementary private funding.
• Who decides. Whether there is another sibling who has to be in the conversation before anything is signed.

An AI secretary for scheduling and first-line response collects that list in a conversation rather than a form: one question at a time, branching on the answer, on WhatsApp and in the website chat, at ten at night and on Saturday. At the end one card opens in the WhaleBiz CRM with a short summary, and the coordinator enters a conversation that already has direction. The less obvious gain is that once every enquiry is collected in the same structure, you can finally see which source produces the enquiries that actually become placements.

The red line: what the agent never says

In a sector dealing with vulnerable people, what the agent refuses to do matters as much as what it does. Four boundaries are defined during implementation and are not left to judgement in real time.

No medical answer. The agent does not diagnose, does not assess a condition, does not advise on medication and does not say whether a doctor is needed. It says so explicitly and hands off to a person, rather than phrasing something that sounds like advice.

No commitment on eligibility. The long-term care benefit and its scope are determined in a National Insurance process that includes a functional assessment. A number stated in a chat is heard as a promise, and when the outcome differs it becomes a complaint. The agent collects the funding picture and passes it to the coordinator.

No promising a specific caregiver. "We have exactly the person you need" before availability is checked is the sentence that starts placements which break in week two.

No final price before the scope is known. It can explain how pricing is built and what affects it, but the number is set once hours, sleep-in and funding are known.

There is also a privacy layer here that must not be glossed over: the conversation contains health information about an identified person, usually a person who is not the one writing. So it needs to live on a card in a system with permissions, not on the private phone of a coordinator who leaves in six months. We set out these duties in our guide to Israel's privacy law and AI agents.

The match is the product, and it starts at the enquiry

A home care agency does not sell hours, it sells a match. Two families with identical hours and identical budgets need two completely different people, and the difference between a placement that lasts two years and one that breaks after two weeks is decided by details that look marginal: whether the caregiver can lift, whether they speak the language of the household, whether they will sleep over, how they handle dementia at night, whether there is a dog.

In most agencies those details are gathered at the home visit, meaning after a drive and an hour of coordinator time have already been spent. When they are gathered at the first enquiry, the order of work inverts: the coordinator arrives at the visit with two or three relevant candidates instead of going out to find them afterwards. That shortens the time between enquiry and placement, and that time is the only thing genuinely competed over in this sector, because a family that waits a week has already signed somewhere else.

Decision week: follow-up that is not pestering

Between the first enquiry and a signature there are usually a few days to two weeks, and inside them a family conversation happens that you are not present at. The drop-off is silent: nobody tells you they chose another agency, they simply stop replying. Four follow-up sequences cover most cases:

• Enquiry with no introductory meeting. A reminder referring to a real detail from the conversation, not generic wording.
• After the home visit, before signing. A short check for open questions that need answering before the family convenes.
• A family that asked for time. "We will speak to my brother and come back" is a defined state, with an agreed return date rather than a hope that someone remembers.
• Those who chose another agency. One polite check a few weeks later. Placements break often, and that family already knows you.

Two rules keep this from becoming pursuit: a stop point after which the case moves to a cold list, and any reply from the family immediately halting the sequence and handing off to a person. The number of automations and follow-ups is unlimited and there are no trigger meters, so a long follow-up sequence does not make your plan more expensive.

After the placement: where coordinator time actually burns

Most of the cost in a home care agency does not sit in acquiring the client but in maintaining open cases. A coordinator's day fragments into shards: a family asking who is coming tomorrow, a caregiver reporting sick at six in the morning, a request for a three-day temporary replacement, a question about extra hours over a holiday, and a complaint that has to reach the right person today rather than tomorrow.

A large share of that traffic repeats and can be answered immediately through an AI agent for service and support: confirming a shift, giving status, capturing a replacement request, and logging a complaint with routing and an alert to a person. What remains for a human is what requires judgement, which is also what justifies their salary.

Here it is important not to gloss over one thing: the agent can answer "who is coming tomorrow" only if the roster exists as a system and not in the coordinator's head. An agency whose roster is updated by phone will get organised enquiries and correct routing, but the answer naming the caregiver stays human until the source is organised. In practice, organising that source of truth is usually the first gain from the project.

And a last practical detail: older family members record instead of typing, and their children photograph a discharge letter. The agent receives both, continues from that point and attaches the file to the card. One agent reply counts as one message, whether it answered text, voice or an image. There is no media multiplier and no double counting.

Where to start and what it costs

The entry point for this sector is the Standard plan, and the current figures are on the price list. It is the first plan that includes both WhatsApp through the official Meta API and calendar booking, and without both there is no product here: enquiries arrive on WhatsApp, and their destination is a home visit or an introductory meeting in the calendar. The Start plan does not fit this sector because it has neither WhatsApp nor appointment booking.

An agency with several coordinators needing separate permissions, several calendars, Facebook and Instagram coverage or an API connection to the scheduling system moves to Business. Prices exclude VAT, the setup fee is paid in one payment before work starts, the subscription begins on the day the agent goes live, and business days are counted from receipt of all materials rather than from payment.

The comparison worth making is not against "how much does it cost" but against a single placement. If an average placement is worth a few thousand shekels a month and lasts months, one enquiry a month that would otherwise have gone unanswered in time covers both the subscription and the setup. Substitute your own numbers, but do the calculation before the call rather than after it.

Frequently Asked Questions

Can the agent tell a family whether they qualify for the long-term care benefit and how many hours they will get? No, and this is one of the most important definitions set during implementation. Eligibility and its scope are determined by National Insurance in a process that includes a functional assessment, and no private party commits to them in advance. A number stated over WhatsApp is heard as a promise, and when the outcome differs it becomes a complaint. The agent maps the funding picture and passes it on: whether an application was submitted, whether an active benefit exists, whether there is private insurance and whether the family is also preparing for supplementary private funding.

Our families send voice notes and photos of documents. Can the agent handle that, and what does it do to the message quota? Yes, and this is a critical channel here: a family member leaving a ward or driving does not type, they record. The agent receives voice notes and images, continues the intake from that point and attaches the file to the card. On billing the rule is simple: one agent reply is one message, whether it answered text, voice, an image or a document. There is no media multiplier and no double counting.

What happens when a family asks a medical question, or when it is an emergency? The agent does not diagnose, does not assess a condition and does not advise on medication, and it says so explicitly rather than working around the limit. Distress signals are defined in advance as a list, for example a described fall, difficulty breathing or a suspicion of neglect, and each triggers immediate routing with an alert to the person on duty instead of waiting in the normal queue. During hours when no person is available the agent points explicitly to the official emergency channels.

We manage shifts in a spreadsheet and in the coordinator's head. Can the agent answer who is coming tomorrow? Only if the roster exists as one current source of truth. An agency managing it in a system or an organised file can connect it, and the agent answers with the caregiver name and arrival time from that source. An agency where the roster is updated by phone gets different, deliberate behaviour: the agent will not guess a name or a time, it confirms the question was received, marks it urgent and passes it to the coordinator with the context. That is a decision, not a limitation, because a wrong message about who is coming tomorrow burns trust that does not come back.

Which plan does a home care agency start on and how long does implementation take? On Standard. It is the first plan with both official WhatsApp and calendar booking, and without them there is no product in this sector. An agency with several coordinators and permissions, several calendars or an API connection to the scheduling system needs Business. Prices exclude VAT, and the subscription begins on the day the agent goes live.