Follow one request through the workflow
An HVAC customer calls after hours to request an estimate for a replacement system. The workflow captures the caller’s contact details, service address and requested work, checks the approved service area, and offers only appointment slots the company has made available. A booking is complete only after the calendar returns an event ID and the CRM stores that ID against the correct contact. If the calendar write fails, the caller is told the appointment is not confirmed and a staff callback task is created.
- 01Capture caller and request
- 02Check service area and approved rules
- 03Read current appointment availability
- 04Save appointment and CRM reference
If the normal path fails: Urgent or unclear requests go to the company’s approved human escalation path. A failed calendar write becomes a callback task, never a claimed booking.
Where this approach stops
- The agent cannot diagnose equipment, promise a technician arrival time or approve a price outside explicit rules.
- Available API access, calendar permissions and reliable contact matching must be confirmed before quoting.
What changes the cost
Voice usage, call length, number of locations, transfer rules, and calendar/CRM write access drive the quote. Voice-agent projects are quoted separately from the $1,500 simple workflow entry price; telephony and model usage are recurring client costs.
Acceptance checks before launch
| Test input | Expected result |
|---|---|
| Slot taken while caller decides | Offer refreshed availability; create no duplicate event. |
| CRM unavailable after booking | Retain event ID and queue reconciliation for a named owner. |
| Unsupported or urgent request | Use the approved escalation; do not improvise advice. |
Choose the calls you want the AI to handle
An AI receptionist project needs a defined job: capture missed calls, answer approved questions, or book a specific appointment type. For a home-service business, these are different scopes. An estimate visit may be simple to schedule; a repair needing a particular technician, service zone, or equipment may require dispatch review.
Start by reviewing a small sample of real calls with permission. Separate new inquiries, existing-job questions, scheduling changes, complaints, and urgent requests. Choose the repeatable category with clear answers and a clear next step. Keep ambiguous requests in a human-owned queue.
Define what counts as a booked appointment
A caller asking for Tuesday morning is a request. It becomes a booking only when the scheduling system accepts an available slot and returns a confirmation. The receptionist should not promise a technician or arrival time based on an empty calendar alone.
Document service area, appointment duration, travel constraints, opening hours, and who may override the schedule. If your scheduling system cannot expose those rules reliably, capture a preferred time and arrange confirmation from your team. That is still a useful improvement over an untracked voicemail.
An example: an after-hours estimate request
Illustrative workflow: a homeowner calls about replacing an air-conditioning system. The receptionist captures their name, callback number, service address, and request. It checks the approved service area, creates or matches the CRM contact, and adds an estimate request for the next business day.
If the CRM write fails, the request stays visible in an exception queue and the assigned person receives an alert. The caller hears that the team will confirm the next step. The system does not claim the appointment is booked. ServiceTitan, JobNimbus, or another CRM can be assessed for this handoff; available actions depend on account access and the specific integration.
Test the boundaries before routing customer calls
Use test calls for a wrong service area, unclear address, repeat customer, noisy connection, unavailable appointment, failed transfer, and unavailable CRM. Confirm the record and task in the destination system, not just the call summary.
For urgent or uncertain situations, follow a business-approved escalation path. The AI should not diagnose hazards or improvise emergency advice. Give callers a clear route to a person and define what happens when that person does not answer.
- Approved answers: services, hours, and pricing statements the team has authorized.
- Human handoff: complaints, uncertain requests, and exceptions to normal scheduling.
- Record quality: correct contact, request details, owner, and next action.
- Failure handling: an observable fallback when booking, transfer, or saving fails.
Budget for the complete handoff
The voice subscription is one part of the cost. Implementation also includes call design, access setup, CRM mapping, tests, and handover. Ask whether number configuration, usage charges, monitoring, and ongoing changes are included.
GoodHandoff workflow projects start at $1,500 for a narrow existing-tools workflow. A multi-step AI receptionist with scheduling and CRM work needs its own written scope; do not assume it fits the entry price. Bring your phone setup, CRM name, and one call type to the first conversation.
Put this into practice.
Bring your workflow to a short conversation. We’ll review the problem and identify whether a scoped project makes sense.
Talk through your workflow →