Skip to content
StrandCalls
Get Support

Safety and escalation

An AI answering your phone is only worth having if it never embarrasses you. StrandCalls ships a set of hard guardrails that no setting, instruction, or caller pressure can turn off.

  • Invent a price. Quotes come from your service catalog as a structured lookup, read back. No catalog entry, no number.
  • Invent an ETA. It states arrival times only after you’ve given one. See Emergency dispatch.
  • Book an emergency for next Thursday. Urgent calls go through the dispatch loop, not the calendar.
  • Play doctor, electrician, or lawyer. For anything with a safety or legal dimension it follows the script and escalates rather than diagnosing.
  • Re-ask what it already knows. Returning callers aren’t made to repeat their address.

Your trade type in Settings → Business Profile selects the safety script the AI runs — exactly one, chosen for you. These are the “before anything else, do this” moments. Four trades have one today:

TradeWhat the script does
PlumbingCoaches the caller to the shutoff valve before anything else; gas smell means leave the house and call 911 from outside; sewage backup is always urgent
ElectricalAsks about fire or smoke first, then coaches only “turn off the breaker” — never wiring, never a diagnosis. A single dead outlet is explicitly not treated as an emergency
HVACA three-question vulnerability triage on any no-heat or no-cool call — who’s in the home, the indoor temperature, whether there’s a safe backup — so a no-AC call with an elderly resident isn’t handled like a routine one
Pest controlRoutine pests book normally; an active indoor termite swarm is escalated, and an allergic reaction in progress is 911 first

Scripts for locksmith / towing and garage door ship with the product and switch on automatically if those trades become selectable. The locksmith script carries the life-safety rule that matters most: a child or pet locked in a hot car gets 911 first, every time, before any conversation about service.

The common thread is that the script front-loads the safety step and then gets out of the way. None of them make the AI alarmist about routine work — several say so explicitly, because a receptionist that treats every call as an emergency is as useless as one that treats none of them that way.

The scripts aren’t editable — they’re the same for everyone in a trade, which is the point. Settings → Safety & Escalation carries a single Safety coaching scripts switch, on by default. Turning it off silences the script for hazardous situations while leaving everything else intact, which is a decision worth taking only on advice from your insurer or your lawyer.

Escalation re-alert timing and your dispatch alert phone are on the same screen — see Emergency dispatch.

When the AI is out of its depth, the correct behavior is to stop and get you — not to improvise. Escalation is a guaranteed path: the call produces a message, a notification, and a record, rather than a confident wrong answer.

This is why “I’ll have the owner call you back” is a feature. It is the honest ceiling of what a receptionist should decide alone.

For statements that carry legal or contractual weight — licensing, insurance, warranty terms, service-area claims — the AI reads from fact sheets you have explicitly verified, not from anything it inferred. Each is a topic and a short answer, and the AI quotes it as written.

  • Only verified sheets are used. Editing a sheet clears its verification until you re-verify it, so a half-finished edit can never reach a caller.
  • Up to five sheets are loaded on a call, yours taking precedence over any StrandCalls ships, most recently verified first.
  • If a caller asks something no fact sheet covers, the AI says it would rather have you confirm than guess, and takes a message.

Fact sheets are managed in Settings → Safety & Escalation.

Two different things are easy to conflate here.

Transcripts are always kept — they come from the conversation itself, and they’re what the summary, lead quality, and safety review are built from.

Audio recording is off until you turn it on, in Settings → AI Receptionist. Recording law varies by state, which is why it isn’t on for you by default. If your state or your customers require a specific spoken disclosure, raise it with support — this is not something to improvise in custom instructions.

Texting has its own, separate consent, captured on the call before the AI sends anything. See Texting consent and opt-outs.

Every safety script, the escalation guarantee, and emergency dispatch ship identically on every plan. Charging more for safety would be a moral hazard; it isn’t a lever we pull.

Read transcripts in Calls. Every quote the AI gave is logged, every escalation is recorded, and every dispatch has a state history. Calls are also scored for safety after they end, which is what suppresses a review request on a job that went badly. If you ever suspect the AI said something it shouldn’t have, the evidence is there — and it’s worth telling support about.