Appointment confirmations
Calls to confirm, remind, or reschedule: and writes the result back to your calendar. Cuts no-shows without tying up reception.
It makes and takes phone calls in natural Greek or English: confirming appointments, qualifying inbound leads, running feedback calls. It understands replies, handles interruptions, and hands the call to a person the moment it should.
A voice agent holds a real phone conversation. It speaks in natural Greek or English, understands the caller's replies in real time, follows a goal (confirm an appointment, qualify a lead, collect feedback), and acts on the outcome, booking, logging, or transferring to a human with context. It runs outbound (proactive calls) and inbound (answering). Two things govern the outbound side in Europe: it calls only numbers you hold consent for, because an automated call is a stricter legal category than a human one, and it says it is an AI in its opening line, as Article 50 of the EU AI Act has required since August 2026. Operational tier, built on Claude with a low-latency speech layer. For pure inbound reception, see the Call Answering Agent; for building the consent list in the first place, see the AI Sales Agent.
theagency47 · Updated June 2026| Job-to-be-done | Hold goal-directed phone conversations at volume, in Greek or English, and act on the outcome |
| Tier | Operational (departmental flow) |
| Underlying model | Anthropic Claude + low-latency speech-to-text / text-to-speech |
| Trigger | Outbound: scheduled call list, filtered to numbers with recorded consent · Inbound: incoming call to a number |
| Inputs | Call objective + script guardrails, consented contact list (outbound), knowledge base, calendar/CRM access |
| Consent gate | Fail-closed. No outbound number is dialled without a consent record attached, with its source and timestamp. Unverifiable records are blocked, not assumed. |
| AI disclosure | Discloses that it is an AI in the opening line, per EU AI Act Article 50 (applicable since 2 August 2026). Not configurable off. |
| Tools available | Telephony (SIP), calendar, CRM read+write, live transfer, SMS follow-up |
| Autonomous decisions | Conversation flow, when to confirm/repeat, when to transfer, follow-up channel |
| Escalation rules | Caller asks for a human → immediate transfer · Out-of-scope request → transfer · Low confidence / distress detected → transfer with context |
| KPIs measured | Calls handled, completion rate, transfer rate, average handle time, booking/qualification rate, CSAT |
| Eval suite | 22 test cases incl. accent/interruption handling and transfer triggers. Greek + English transcripts reviewed. |
Calls to confirm, remind, or reschedule: and writes the result back to your calendar. Cuts no-shows without tying up reception.
Calls inbound enquiries, asks your qualifying questions, scores the lead, and books the fit ones onto a human's calendar.
Runs post-service or post-purchase calls, captures structured feedback and sentiment, flags unhappy customers for follow-up.
Speaks fluent, natural Greek (and English): handles interruptions, clarifications, and informal phrasing, not a rigid IVR tree.
Transfers to a person the moment the caller asks or the request goes out of scope: with a spoken summary so the human starts informed.
Every call transcribed, summarised, and logged to your CRM with outcome and next action. Full audit trail.
Outbound dials only where a consent record exists, with source and timestamp. An automated call is a stricter legal category than a human one, so the gate fails closed rather than guessing.
Says it is an AI in its opening line, as the EU AI Act has required since August 2026. On a call the person agreed to receive, saying so disarms people faster than pretending would.
Custom integrations with proprietary systems are quoted as add-ons. Not sure if yours fits? Describe your stack and we'll confirm.
Yes, with the current generation of speech models, Greek is fluent and natural, not robotic. We review real call transcripts in both languages during the eval phase and tune until tone and handling meet the bar before go-live.
It transfers to a human immediately (on request, on any out-of-scope topic, or on low confidence) and gives the person a spoken summary so the caller never has to repeat themselves.
They overlap but differ in focus. The voice agent is built for proactive, goal-directed conversations (outbound confirmations, qualification, feedback). The Call Answering Agent is built for inbound reception, answering your main line, triaging, and routing. Many clients run both.
No, and that is a deliberate limit rather than a missing feature. An automated call, where a system dials and holds the conversation without a human, is treated more strictly in European law than a person picking up a phone: it needs express prior consent, and in Germany there is no business-to-business exception at all, with the existing-customer exception covering email but not calls. So the outbound list is filtered to numbers with a consent record before anything is dialled. If you need to reach businesses that have not consented yet, that first approach is made by a person, and the AI Sales Agent is the system that researches them, prepares the briefing and builds the consent list the voice agent then works.
Yes. Article 50 of the EU AI Act, applicable since 2 August 2026, requires that a person be told they are interacting with an AI system unless that is already obvious. Our agent discloses it in the opening line and we do not offer a version that hides it. In practice it costs less than people expect: on a call the recipient has already agreed to receive, saying so plainly tends to lower their guard rather than raise it.
Build as a Spark engagement; running costs are telephony minutes plus Claude + speech usage, billed at cost. We size it on call volume during discovery.