Maintenance request line
Maintenance requests often arrive without the information needed to act. “The heat is broken” does not identify the unit, the equipment, when the issue began, whether the whole property is affected, or whether anyone has observed a safety hazard. A voicemail may preserve the complaint but still leave staff with a round of follow-up calls.
An AI phone agent can turn that first report into a consistent intake record. It asks the approved questions, repeats critical details, creates a request through the configured system, and follows a written escalation rule when the caller reports a covered urgent condition. It should capture observations, not diagnose the cause or decide the repair.
A concrete maintenance call
Imagine a resident calling because there is no heat in the unit.
Agent: I can record the maintenance request and follow the property's escalation rules. Before we continue, do you smell gas, see smoke or fire, or believe anyone is in immediate danger?
Caller: No. The radiators are cold, and the thermostat display is on.
Agent: Thank you. I will record those as your observations. What is the property address and unit, and when did you first notice the problem?
Caller: It started this evening. The whole unit feels cold.
Agent: Is the maintenance team allowed to enter if you are away, or does someone need to be present? Please do not share a door code on this call unless the property's approved policy asks for it.
The agent confirms the caller's contact information, location, observed symptoms, affected area, access preference, and any pets or site constraints the maintenance policy requires. It submits the request to the system of record. If the rules classify the report for immediate review, the agent alerts or transfers to the on-call team; otherwise, it explains only the next step the organization has approved.
This workflow fits property operations, facilities desks, equipment fleets, campuses, and service organizations. Their safety and access rules differ, so use the industry examples for context rather than copying one industry's prompt into another.
How to configure the workflow on ThunderPhone
Build the prompt from the maintenance form and escalation policy already used by the team. Define required fields such as caller, callback details, precise location, affected asset or area, observed condition, start time, scope, access preference, and relevant site constraints. Require the agent to distinguish what the caller saw, heard, or smelled from what the caller thinks caused it.
Write explicit urgent-condition rules. Each rule should specify the approved safety language, the transfer or alert destination, and the fallback if nobody accepts. Do not ask the agent to make an open-ended safety judgment. Life-safety guidance must come from the organization's current policy, and immediate danger should be directed to local emergency services.
Attach approved troubleshooting boundaries, property information, access policies, and maintenance FAQs through the knowledge base when they are stable enough to use. ThunderPhone accepts selected text, Markdown, CSV, PDF, and DOCX documents and provides a built-in mid-call knowledge search tool. Documents must be attached to the agent; an organization upload does not apply automatically. Avoid putting a repair manual into the call flow if callers should not perform those steps themselves.
Use an inbound demo U.S. number for testing, not production. A live maintenance line should use a production number brought through a supported direct connection or manual SIP configuration. If attended transfer is part of the escalation, it requires an eligible production number that can place an outbound SIP leg. Demo numbers and browser calls can use cold transfer only.
Connect the workflow to the actual maintenance operation:
- Use a REST integration or signed webhook to create or update requests in a property, facility, or work-order system.
- Use Salesforce or HubSpot when service records live with customer accounts or cases.
- Use Google Sheets only when it is the approved request queue.
- Use Slack for on-call alerts and review requests.
- Use Google Calendar or Cal.com if the team exposes bookable visit availability.
- Use Zapier MCP for approved actions made available by the connected server.
The integration response should determine what the agent says next. It may confirm that a request was received if the write succeeded; it should not say that a technician is assigned or arriving unless an authoritative system confirms that fact. Field service dispatch covers the next operational step, while urgent-call routing goes deeper on escalation design.
Test ordinary issues, vague descriptions, duplicate requests, unsafe conditions, unsupported troubleshooting requests, access changes, unavailable staff, and failed writes. Browser mic tests, reusable simulations, graded call logs, regression suites with minimum pass-rate gates, and live-traffic A/B experiments are available. Simulations are billable real calls, with the charge shown before a run.
What to measure
Measure required-field completeness, successful request creation, caller corrections during read-back, duplicate rate, urgent escalation rate, transfer acceptance, and repeat calls about the same unresolved issue. Compare the original call record with the maintenance system to find lost or mis-mapped details.
Review whether the team changes the priority after reading the request. Frequent changes can mean the written rules are unclear or the prompt asks the wrong questions. Track successful human handoffs as their own outcome; an appropriately escalated call is not a containment failure.
Honest limits
The agent cannot inspect a property, determine the cause, authorize a repair, verify that an area is safe, or promise an arrival time without confirmed scheduling data. It should never encourage a caller to attempt hazardous troubleshooting.
Humans remain necessary for emergencies, uncertain hazards, access disputes, accommodation needs, tenant or customer complaints, cost authorization, and priority decisions that depend on context outside the call. The organization is also responsible for its caller notices, recording choices, data handling, integrations, and retention practices.