A demo proves the agent can hold a controlled conversation. A launch needs more: approved knowledge, tested phone routing, fallback contacts, booking rules, integration behavior, caller disclosure, and someone responsible for review.
Before Havio answers live callers, the workflow should pass practical checks.
1. Approved knowledge
The agent should only answer from approved business information: services, hours, locations, pricing ranges, policies, FAQs, booking rules, and escalation boundaries.
If the answer is not approved, the agent should transfer, create a callback, or say that a person will follow up.
2. Call routing
The phone path should be tested from real devices, not only from an internal browser demo.
Test missed calls, busy calls, after-hours calls, office-hours calls, transfers, voicemail fallback, and pause behavior. The team should know how to stop forwarding if something is wrong.
3. Handoff rules
Define what must reach a person: urgent requests, sensitive topics, upset callers, high-value leads, account issues, medical/legal/financial questions, and anything outside the configured workflow.
Every handoff should include context, not just a phone transfer.
4. Integrations
Calendar, CRM, ticketing, Slack, Teams, sheets, and webhooks should be tested with real field names and failure paths.
The team should know what happens if a write fails. A useful call summary should still exist even when a destination system is unavailable.
5. Disclosure, recording, and retention
Decide the caller disclosure language, recording setting, transcript retention, deletion path, and sensitive-topic rules before launch.
The right answer depends on region, industry, and policy. Do not leave this to the agent to improvise.
6. Review ownership
Someone must review early calls. The first week should look for missing knowledge, confusing questions, weak qualification fields, failed handoffs, and calls that should not have been automated.
The review owner turns real call evidence into better rules.
A simple launch rule
Launch the smallest useful workflow first. Expand once the team can prove the agent answers correctly, hands off safely, updates the right systems, and leaves a record that humans trust.