Blog / Owner setup
Set up an AI receptionist without filling out every possible rule at once
Start with the business essentials, let the assistant ask focused follow-up questions, and review a short draft before activation. Leave unknown prices and exceptions unresolved for human review instead of inventing defaults.
Tell the assistant about a normal job
Describe what your business does, where you work, when you take appointments, and what usually happens on a straightforward job. Give one example of work you decline and one example that needs inspection. This supplies context without asking you to predict every edge case.
The onboarding assistant should extract a draft and show which answer supports each rule. You should be able to correct a service, time, price, or location directly. If the assistant infers something you did not say, it should ask a question rather than present that inference as your policy.
Ask only for the missing decision
If a fixed price is missing, ask whether that job uses a fixed price, a range, or an inspection. If after-hours handling is missing, ask whether to offer a callback, contact an on-call person, or decline new appointments. The setup should permit “I do not know yet” without creating a fake answer.
Some omissions block activation for a particular flow. For example, the assistant cannot promise emergency dispatch without an approved contact and fallback. Explain the missing item and let the owner complete other setup work. Avoid turning an unresolved emergency rule into a surprise live behavior.
Make pricing easy to review
Group prices by service, show included work beside exclusions, and make additional charges visible. Do not require an owner to enter a detailed price for a job that always needs inspection. A clearly stated inspection policy can be the accurate answer.
Keep customer-facing language separate from the underlying rule. The owner should review both: the price decision and the words callers will hear. A friendly sentence cannot authorize a discount or override a fee that the owner did not approve.
Practice before activating
Use a routine request, an out-of-area caller, an unavailable slot, a disputed price, and an urgent after-hours call. Include a caller who refuses recording and one who asks an unknown question. Review both the spoken answers and the resulting records.
Approval should name the exact rule version. Later feedback can propose an update, but should not silently change the live policy. If a new version fails a scheduling or pricing check, keep it as a draft and explain what must be resolved.
Keep troubleshooting useful and bounded
A support assistant can ask what happened, collect a request identifier, explain known configuration issues, and prepare an escalation with the relevant facts. It should not ask customers to paste passwords or API keys, disclose another business’s records, or claim a repair succeeded without checking.
Zephaux’s adaptive AI onboarding and troubleshooting agents are planned, not currently live. The existing workspace provides configuration screens and a Test Lab. Read the Trust Hub for current availability and the Privacy Policy before providing account information.