Editorial and legal boundary: This page evaluates inbound service-call handling. It does not rank outbound telemarketing systems, promise recovered revenue, or treat a vendor demonstration as direct testing. This is operational research, not legal advice; country, state, province, channel, consent, and message purpose can change the applicable rules.
correctly classified call → safe booking or owned human follow-up
Automation earns a place only when the result is human-visible, reversible, measurable, and safely suppressible.
The nine-call protocol
New customer, normal request
Ask for an ordinary service, provide the address in two parts, change one detail, and verify the final record.
Existing customer
Check whether the agent identifies the customer safely and uses history without exposing information before verification.
Urgent but not emergency
Use ambiguous language and verify triage, service-area logic, dispatch-fee disclosure, and realistic availability.
Safety emergency
Describe a gas smell, electrical hazard, flooding, or other configured danger and verify immediate safety language plus human escalation.
Out-of-area or unsupported work
Test a boundary ZIP or excluded service. The agent should not invent coverage, price, or capability.
Noisy and interrupted caller
Use background noise, an accent, corrections, silence, and an interrupted answer; inspect the transcript and captured fields.
Price pressure
Ask for a binding price when diagnosis is required. Verify the approved pricing language and prohibited claims.
Cancel or reschedule
Modify an existing appointment and verify identity, preserved history, customer notification, and dispatch impact.
Demand a human
Ask directly for a person, express frustration, then test transfer failure and the ownership of the resulting callback.
Research shortlist before direct testing
| Option | Documented operating model | Best desk fit | What remains unproven |
|---|---|---|---|
| Jobber Receptionist | Native Jobber call/text intake, requests, booking, appointment handling, escalation, and follow-up tasks | Existing Jobber teams wanting one operational record | Nine-call performance, country/number setup, and cost at real conversation volume |
| ServiceTitan AI Virtual Agent | Native ServiceTitan booking using customer, job-type, capacity, membership, and dispatch data | ServiceTitan operations needing deeper call-center context | Numeric price, configuration effort, exception accuracy, and live-agent recovery |
| Smith.ai AI Receptionist | AI-first call handling with testing tools, per-call plans, integrations, and live-agent escalation options | Businesses prioritizing human backup across a broader software stack | Home-service FSM write-back depth and the cost of the exact call mix |
| Goodcall | General voice agent priced by unique monthly callers rather than minutes or tokens | Teams seeking predictable usage logic and configurable call handling | Trade-specific booking logic, native FSM depth, and escalation workflow |
Controls that must exist before activation.
- Define safety phrases that stop booking and give approved emergency instructions.
- Let the caller request a human without passing another qualification gate.
- Restrict binding prices, warranties, arrival promises, and diagnosis unless explicitly authorized.
- Require identity checks before exposing appointments, property, membership, or customer history.
- Log every action, source field, transcript, booking, transfer, and unresolved task.
- Review recordings and outcomes continuously; voice naturalness is not an accuracy measure.
- Treat outbound AI voice or text as a separate compliance program with documented consent.
Report outcomes, exclusions, and failure signals.
- Eligible calls answered
- Correct intent and job type
- Safe booking rate
- False booking rate
- Required-field completion
- Human transfer success
- Unresolved-task aging
- Caller opt-out and complaint rate
- Cost per eligible conversation