An AI phone agent for the agency phone that never gets to stop ringing
A home-health or home-care agency runs on its phone in a way few businesses do. Referrals arrive by phone from discharge planners and physician offices that are calling several agencies at once — the agency that answers first and captures a complete referral is best placed to admit the patient; the one whose line goes to voicemail may never get the chance. Caregivers call off at 5:40 in the morning, and every call-off is a visit that must be re-staffed before it becomes a missed visit. Families call worried, clients call confused about today's schedule, and after five o'clock the whole load lands on an on-call coordinator carrying a phone that never stops being their problem.
ThunderPhone gives an agency an AI phone agent that works in both directions: inbound coverage for the calls you receive — referrals, call-outs, family inquiries — and outbound campaigns for the calls your coordinators never have time to place, like visit confirmations and shift-fill outreach. It follows your instructions, collects the fields your intake and scheduling teams require, and searches your approved knowledge base without inventing agency policy.
One boundary comes before everything else on this page. Home-health and home-care calls routinely involve protected health information, and the agent is a front-office worker, not a clinician: it never gives clinical guidance, never answers medication questions, and never discusses a client's condition. If a caller reports an emergency, the agent delivers the agency's approved instruction to hang up and dial 911 and escalates immediately. You define what it may collect, what it may never discuss, and where every call type routes — and it executes those rules exactly.
What a home-care phone agent can handle
Referral intake — the call that decides an admission
Referral sources rarely wait. When a discharge planner or case manager calls, the agent answers at any hour and runs your intake script conversationally: referral source and callback number, patient name and location, payer, services ordered, requested start of care, and whatever else your admissions checklist requires. The structured referral record is delivered to your intake team as soon as the call ends — by webhook, connected app, or an API your system exposes — and the agent can warm-transfer a referral source who wants a person, briefing your coordinator privately before connecting the call. A referral that arrives at 7:10 p.m. gets captured completely instead of becoming a voicemail your competitor never left them time to leave.
Caregiver call-outs and shift coverage
The early-morning call-off is the defining scramble of home care. The agent can run your call-off line around the clock: verify which caregiver is calling, capture the affected client, visit time, and reason in your required format, and deliver the record to your scheduling system and on-call coordinator the moment the call ends — flagged by urgency rules you define, such as visits starting within the next few hours. Outbound, a shift-fill campaign can then work your list of qualified, nearby caregivers one by one, present the open shift, capture an accept or decline on the call, and stop the campaign the moment the shift is filled. Your coordinator wakes up to a filled visit or a short, structured escalation — not a phone tree of their own making.
Visit confirmations and missed-visit follow-up
Agencies chase unconfirmed visits in both directions. Outbound campaigns can call caregivers with next-day shift reminders and call clients or family contacts to confirm scheduled visits, capture a change request on the spot, and repeat approved visit details. When your visit-verification workflow flags a missed clock-in, an immediate outbound call can attempt to reach the caregiver and, per your script, reassure the client that the office is on it — with every outcome logged and delivered to your systems by webhook. Voicemail detection can leave your configured message or hang up and retry on the policy you set.
Family inquiries and new-client intake
The first call from a family is often someone in a hard week — a parent being discharged, a fall, a spouse who cannot keep up alone. The agent can hold that conversation the way your best intake person would: capture who is calling and their relationship, who care is for, the kind of help they are looking for, the service area, timing, language preference, and how they found you, then schedule the assessment visit through a connected calendar and answer approved questions about licensing, caregiver screening, and backup coverage from your own policy material. Questions about cost or payer specifics beyond your approved answers route to your intake team rather than being guessed at.
Intake and assessment coordination
Between referral and start of care sits a stack of coordination calls the agent can carry: confirming the assessment appointment with the family, reminding them what paperwork to gather, following up when scheduling stalls, and logging every outcome for your admissions team. Communication gaps between intake, scheduling, and the field are where strong agencies still lose admissions — a phone agent that never forgets to place the follow-up call closes that gap.
Client and family calls about care — route, don't discuss
Some callers want to know how their mother is doing, whether her medications changed, or what the nurse said yesterday. The agent never answers those questions. Per your configuration, it verifies nothing about any client's status, takes a structured message identifying the caller and the request, and routes it to the clinical or care-coordination team you designate. What the agent may acknowledge, what it must never discuss, and where each request goes are yours to define — and the safe default is narrow.
After-hours answering with agency-defined escalation
After hours, the small set of recurring call types — an emergency, a call-off, a new referral, a worried family member, routine admin — each gets the exact handling your protocol assigns. Emergencies get the 911 instruction and immediate escalation to your on-call clinician. Call-offs and referrals get full structured capture. Routine matters get queued for the morning instead of waking anyone. Nothing is improvised, and everything is logged.
Multilingual communication for clients, families, and caregivers
ThunderPhone supports 47 voice languages across 123 voices. Set a primary language and additional languages for your client and caregiver base, and the agent can switch mid-call when a caller changes to a configured language — the same intake, call-off, and escalation rules in every language, with a transcript for every call.
Sample home-care call flow
Agent: Thank you for calling Brightside Home Care. This is the agency's automated assistant. If this is a medical emergency, please hang up and dial 911. Are you calling about a shift, a referral, or something else?
Caller: It's Dana — I'm on the schedule for Mrs. Alvarez at 8 this morning and my car won't start. I can't make it.
Agent: Thank you for letting us know, Dana. To make sure I log this correctly: this is the 8:00 a.m. visit today for Mrs. Alvarez, and the reason is car trouble. Is that right?
Caller: Yes. I might be able to do my afternoon visit, but not this one.
Agent: Understood — I have recorded the call-off for the 8:00 a.m. visit only, with your afternoon visit unchanged. The on-call coordinator is being notified now because this visit starts soon. Is this the best number if they need to reach you?
Caller: Yes, this one.
Agent: You're set, Dana. The coordinator has the details and coverage outreach is starting.
Behind that call, the structured call-off lands with the on-call coordinator immediately, and — if the agency has configured it — an outbound campaign begins working the qualified-caregiver list for the 8:00 a.m. visit. Had the caller reported an emergency involving a client, the agent would have skipped everything else: the 911 instruction first, then immediate escalation to the on-call clinician.
Platform capabilities behind the agency workflows
GDPR and HIPAA compliance posture — and the agency's configuration responsibility
ThunderPhone's public compliance posture is GDPR and HIPAA compliant. That posture does not make any particular agency workflow compliant by itself, and in home care — where a caller's name tied to the fact that they receive care is already protected health information — this responsibility deserves the most attention on this page. The agency decides and configures: which information the agent may collect on each call type and which it must never request or discuss; whether calls are recorded and what notice callers receive; how long transcripts and recordings are retained; which integrations receive protected health information and under what agreements; where PHI-bearing calls route; and who can access call data. Limit the agent to the minimum information each workflow needs, attach only approved documents, and have your compliance owner review the prompt, recording settings, and data destinations before launch.
Inbound and outbound on your numbers
Agents answer inbound calls and place outbound campaigns. Bring your agency's existing numbers through a supported VoIP connection or manual SIP configuration; verified numbers handle both directions, so referral sources and families keep dialing the number they already know. A web voice widget can put the same agent on your website for families who prefer to start there.
Warm and cold transfer
Cold transfers route a caller immediately. On eligible production numbers that can place an outbound leg, warm transfers let the agent hold the caller, privately brief the on-call nurse or intake coordinator, and connect only when the recipient accepts — the right shape for an escalated referral source or an urgent family call.
Knowledge grounded in your documents
Attach service descriptions, service-area details, caregiver-screening and backup-coverage policies, and family FAQs — text files up to 5 MB, PDF or DOCX up to 50 MB — or import pages from your website by URL. The agent searches those approved sources mid-call instead of improvising, and routes anything they do not cover.
Integrations that fit your stack
Connected apps, HTTP APIs, and remote MCP servers let the agent act during a call — file a referral, log a call-off, book an assessment — when your scheduling or intake system exposes the action. Webhook deliveries are signed with HMAC-SHA256 and retried automatically, so a briefly unavailable endpoint never loses a referral.
Testing before a family or referral source ever hears it
Rehearse the calls that scare you: a call-off for a visit starting in one hour, a family member pressing for clinical details, a referral source in a hurry, an emergency mentioned mid-call. AI-caller simulations create graded transcripts, passing scenarios become a regression suite so a prompt edit cannot quietly weaken the escalation script, and A/B experiments settle wording questions on live traffic. Production calls produce transcripts, recordings where enabled, and grading.
Guardrails an agency should write into the agent
- No clinical guidance, ever. The agent never assesses a symptom, answers a medication question, or discusses a client's condition, diagnosis, or care status. Those calls route to the clinical team you designate.
- Emergency script first. A caller-reported emergency gets the agency's approved 911 instruction and immediate escalation — before and instead of everything else.
- Minimum necessary collection. Each call type collects only the fields the agency requires; the agent does not probe beyond them.
- Escalation is configured, not improvised. On-call paths, urgency thresholds, and transfer targets are set by the agency and proven in simulation before launch.
- A person is always reachable. Any caller who asks for a human gets a transfer path or a structured callback — the agent never stands between a family and the agency.
What would this cost each month?
ThunderPhone is pay as you go — per-minute engine rates with no subscription. Suppose an agency handles 900 AI-assisted calls per month, averaging 4 minutes each:
900 calls × 4 minutes = 3,600 minutes per month
- Spark:
3,600 × $0.02 = $72/month - Bolt:
3,600 × $0.05 = $180/month - Storm:
3,600 × $0.09 = $324/month
That is illustrative base engine usage, not a quote. Premium voice or language paths add 2¢ or 3¢ per minute depending on the voice, and phone-number or carrier charges depend on the telephony setup. Hold time bills a flat 2¢ per minute; voicemail and phone-menu minutes bill at the engine rate. Optional add-ons carry their own published surcharges. The agent builder shows the current all-in rate before calls are placed.
Frequently asked questions
Is ThunderPhone HIPAA compliant for a home-health agency?
ThunderPhone's public posture is GDPR and HIPAA compliant. The agency remains responsible for configuring its own compliant workflow: what the agent collects on each call type, what it may never discuss, recording and notice settings, retention, which integrations receive protected health information, and where PHI-bearing calls route. Review the configuration with your compliance owner before launch.
Can the agent capture referrals from hospitals and physician offices?
Yes. It answers immediately, runs your referral-intake script — source, patient, payer, services ordered, start-of-care timing — and delivers the structured record to your intake team by webhook or integration the moment the call ends. It can also warm-transfer a referral source who wants a person.
Can it handle caregiver call-outs at 5 a.m.?
Yes. It logs the call-off in your required format, notifies the on-call coordinator instantly with urgency flags you define, and — where configured — starts an outbound campaign through your qualified-caregiver list to fill the open shift, stopping when someone accepts.
Will it answer a family's questions about their parent's condition or medications?
No. The agent never discusses clinical status, diagnoses, or medications. It takes a structured message identifying the caller and the request and routes it to the team the agency designates.
What happens if a caller reports an emergency?
The agent follows the agency's emergency script: the approved instruction to hang up and dial 911, followed by immediate escalation to the agency's on-call path. It never triages or assesses the situation itself.
Can it confirm visits with clients and remind caregivers of shifts?
Yes, as outbound campaigns in both directions — client visit confirmations with change capture, and caregiver shift reminders — with voicemail detection, retry policies, and every outcome delivered to your systems by webhook.
Can one agent serve families and caregivers in different languages?
Yes. Configure a primary language and additional languages from the 47 supported, and the agent switches mid-call when a caller changes language, provided the selected voice supports the configured set.