Use case · Inbound calls
Every call gets answered. The difficult ones reach a person who already knows why they rang.
A voice agent picks up when your front desk cannot: overnight, over lunch, during the Monday rush. It confirms who is calling, books or moves appointments in your real calendar, answers routine questions from approved content, and transfers anything sensitive to staff with a written summary.
For operations directors and practice managers at multi-site service businesses that take most of their bookings by phone.
One Tuesday, twice
The same phone line on two different days.
An illustrative day at a clinic with two receptionists and one shared number.
Without the agent
Two calls reach voicemail before opening. One caller leaves no message.
Both receptionists are checking people in. Three calls ring out.
Lunch cover is thin. A prospective new client waits on hold, then hangs up.
Voicemails are returned in the afternoon lull. Two callers have already gone elsewhere.
The out-of-hours message takes over until morning.
A caller wants to cancel tomorrow's slot and cannot. It goes unused.
With the agent
Both calls are answered. One books a first visit; the other's question is logged.
Unanswered calls overflow to the agent after a set number of rings. Reschedules are handled without interrupting the desk.
The new client answers intake questions, chooses from real free slots and is booked. Details land in the CRM.
Staff read call summaries instead of voicemails, each showing intent and what was already done.
Evening calls are answered. Urgent ones follow your route to the on-call number.
The cancellation is recorded and the slot returns to the calendar.
Inside a single call
From first ring to written record.
Three stages, each tied to a system you already run: the phone line, the calendar and the CRM.
Conceptual flow, not a live system or measured result.
Answer and identify
Your telephony platform routes the call to the agent over a SIP trunk or contact-centre integration, and speech is transcribed as it streams. The agent greets the caller with your business name, states that it is an automated assistant, and asks what the call is about. It matches the calling number to a CRM record where one exists, then confirms identity with two details you choose before discussing any existing booking or account.
Act in the real systems
The agent classifies intent: book, reschedule, cancel, question, complaint or something else. For bookings it queries the calendar or practice-management API for free slots by service, site and practitioner, offers a short list, reads the choice back and writes the appointment. A confirmation goes out by text or email. Questions are answered only from an approved knowledge base; if the answer is not there, the agent says so rather than guessing.
Hand over or close out
Handover triggers are configured, not improvised: the caller asks for a person, sounds distressed, raises a complaint, needs clinical or financial advice, or the agent misunderstands twice. It then makes a warm transfer and gives staff a short summary, or opens a callback task out of hours. Every call ends with a transcript, a summary and an outcome written to the CRM, and recordings follow your retention policy, so staff can see exactly what was said and done.
A call, annotated
What the agent is doing between the words.
An illustrative reschedule call to a fictional physiotherapy clinic, with the check behind each turn.
Agent: Good evening, Example Street Physio. I'm an automated assistant. I can book, move or cancel appointments, or put you through to the team.Caller: Hi, I need to move my appointment on Thursday.Agent: Could you confirm your date of birth and the first line of your address?Caller: Fourteenth of March, and it's Elm Road.Agent: Thanks. You're booked Thursday at 10:30 with Sam. Sam is free Friday at 09:00 or Monday at 14:15.Caller: Friday's fine. Actually, my knee's been worse since last time. Should I still come in?Agent: You're now booked for Friday at 09:00, and a text is on its way. I can't advise on your knee, so I'm putting you through to a clinician. If this is an emergency, please hang up and call 999.Transfer note: Moved Thu 10:30 to Fri 09:00. Knee pain worse since last session; asking whether to attend. Identity verified.Missed-call arithmetic
Size the calls that currently go unanswered.
Every default is illustrative; replace it with figures from your phone system. Results show value at risk today, not revenue the agent will win back.
Unanswered calls = calls × unanswered share. Bookings at risk = unanswered calls × share who wanted to book. Value at risk = bookings at risk × value per booking. Callers still decide whether to book, so none of this is guaranteed income; telephony, platform and implementation costs are excluded.
Illustrative model, not a quote or a measured result.
Guardrails on the line
Where staff keep the final word.
A person on request
Callers are told they are speaking to an automated assistant, and a request for a human is honoured immediately.
No advice beyond approved content
Clinical, legal, financial or pricing questions outside the knowledge base trigger a handover rather than a plausible guess.
Identity before detail
Bookings and account details are discussed only after your verification rules are met.
No card details by voice
Payments go to a secure link or a compliant payment line. Emergency and complaint wording uses reviewed scripts.
Not every phone line needs this
Situations where we would advise against it.
- Most calls are emotional or advisory, such as bereavement or crisis support. Those callers should reach a trained person first time.
- Volume is low and staff already answer nearly everything. Better call forwarding or an answering service will cost less.
- Your diary lives on paper or in a spreadsheet with no API. The agent could only take messages until scheduling moves into a system it can use.
- You want payments taken or regulated advice given on the call. Those stay with qualified staff or a certified payment line.
Connects to
Telephony, calendars and CRM
- Twilio
- Amazon Connect
- Genesys Cloud CX
- Microsoft Teams Phone
- Google Calendar and Microsoft Outlook calendars
- Salesforce
- HubSpot
- n8n and Zapier for other booking systems
Asked on discovery calls
What operations leads want settled first.
Can we keep our existing phone numbers?
Yes. Numbers stay with your carrier or contact-centre platform, and routing rules decide when the agent answers: overflow after a set number of rings, out of hours, or every call on one line. Starting with out-of-hours calls lets you evaluate it on real callers without changing daytime handling.
Where are recordings and transcripts stored?
Storage follows your residency and retention requirements, including your own cloud tenancy where needed. We choose speech and model providers that can process in the regions you require, redact sensitive fields from transcripts, and match recording retention to your existing policy rather than a vendor default.
What happens when it mishears a name or a date?
Dates, times, names and numbers are read back before anything is written, and names can be spelled out. If confidence stays low or the caller corrects the agent twice, it transfers or books a callback. Pilot transcripts are reviewed to find recurring misrecognitions.
What drives the running cost?
Mostly call volume and average call length, since speech recognition, voice synthesis, model usage and telephony are billed per minute or per request. Integration upkeep adds a smaller fixed element. We model costs from your real call data before any build, so the business case rests on your numbers.
How does the front-desk team fit in?
Reception staff take transfers with summaries, work the callback list and own the knowledge base the agent answers from. They know which calls go wrong, so they help write the handover rules, and their feedback on transcripts drives tuning after launch.
Have you built voice agents before?
Yes. ColdAI built CallBookIt, an AI voice concierge platform; the case study is linked on this page. Client projects are scoped from your own call recordings and booking rules, and latency, interruptions, read-backs and handover are tested before any caller hears the agent.
Go deeper
The pieces behind this workflow.
What would you like to move forward?
Start with the outcome that feels closest. We'll help give the next step a useful shape.
One conversation. A clearer direction.
Tell us what you want to improve. A few sentences are enough to start.
shayan@coldai.orgLast reviewed 25 September 2026.