An AI phone agent for the calls a restoration company cannot afford to miss
Water does not wait for business hours. A burst pipe at 2 a.m., a sewage backup on a holiday weekend, a storm that fills the lines with fifty simultaneous emergencies — restoration's highest-value calls arrive at exactly the moments a human cannot answer them. And the caller standing in an inch of water is not leaving a voicemail: they are working down a list, and the first company that answers live and gets a crew moving usually gets the job.
ThunderPhone gives a restoration company an AI phone agent that works in both directions: inbound coverage for the emergency line, the office line, and the adjuster calls that interrupt both, and outbound campaigns for the calls your team never has time to place — job-status updates, confirmations, post-completion follow-up. It follows your instructions and searches your approved knowledge base without inventing company policy.
The agent does not practice insurance and does not improvise triage. It never tells a caller whether a loss is covered, never promises what a carrier will pay, and never talks past an active safety emergency. You define what it may say, what it must collect, and when it must escalate — intake scripts and dispatch rules come from the company, and the agent executes them exactly.
What a restoration phone agent can handle
Emergency intake with a structured loss record
An emergency call is only useful to the crew if the right details were captured, and a panicked homeowner does not volunteer them in order. The agent can run your intake script conversationally, in whatever order the answers arrive: what happened and when it was discovered, whether the source is still active, whether water is near electrical, roughly which rooms are affected, whether the caller expects to file an insurance claim — and if so, the carrier and any existing claim number — plus the address and best callback number. Those questions are examples; the script is yours. The result is a structured loss record delivered to your systems by webhook or API the moment the call ends, not a half-legible relayed message.
Safety screening before anything else
Some calls are not restoration calls yet. If a caller describes an active fire, smoke, a gas odor, or any threat to life, the agent's first job is your configured safety language: call 911 from a safe location and do not re-enter the property. It does not continue intake while a danger is active, and it offers no safety advice beyond the wording you approved. Restoration begins after the scene is safe — your script can invite a callback or flag the lead for follow-up.
Dispatch escalation that does not stop at voicemail
Capturing the loss is half the job; waking the right person is the other half. When intake meets your emergency criteria, the agent follows the escalation path you configured: a warm transfer to the on-call technician — briefed privately before the caller is connected — an immediate dispatch message to your systems by webhook, or both. Triage tiers are yours to define: a genuine active loss pages the on-call crew, while a question about mold in a garage becomes a logged message for the morning team instead of a 3 a.m. phone call. The agent never invents its own definition of an emergency.
Adjuster and insurance-call routing
A restoration office fields constant traffic from adjusters and third-party administrators — scope questions, supplement discussions, document requests, re-inspection scheduling. The agent can recognize these callers by your script, collect the claim number, the insured's name, the reason for the call, and any deadline mentioned, then route by your rules: transfer to whoever owns the file, or file a structured callback task so the right person returns the call with the file open. What it never does is negotiate: scope, pricing, supplements, and coverage positions belong to your team.
Estimate and inspection scheduling
Not every call is an emergency. Callers who want an inspection, a moisture assessment, a rebuild estimate, or a mold evaluation can be booked directly into your calendar through a connected scheduling system — via HTTP tools, MCP servers, or webhooks. The agent gathers the fields your estimators need — property address, type of damage, whether a claim is involved — and books the right visit type. Reschedules and cancellations are captured the same way, so the morning schedule reflects reality.
Job-status and drying-progress updates — in both directions
During drying and demolition, customers want to know one thing every day: what is happening and how much longer. Outbound, a campaign can call each active job's contact with the update your project managers approve — who is coming tomorrow, what was completed today, what happens next. Inbound, when the agent is connected to a system that exposes job status, it can look up the caller's job mid-call and answer from the data your system returns — equipment still in place, next reading scheduled — instead of taking a message. Questions the data does not answer route to the project manager as a structured task. Fewer "just checking in" interruptions; fewer customers wondering whether the fans in their living room have been forgotten.
Post-completion follow-up campaigns
The follow-up call after a job closes — checking that the customer is satisfied, inviting a review, staying visible for the referral — is the outbound workflow every busy company intends to run and almost never does. An agent does not get tired of it. A campaign can work the completed-jobs list on your script and cadence, log every outcome, and flag unresolved concerns for a human callback. Voicemail detection can leave your configured message or hang up and retry on the policy you set.
Storm-surge overflow
After a major weather event, call volume can multiply overnight, and every caller is comparing you to the next number on their list. The agent answers in parallel, runs the same intake on every call, and escalates only the ones your rules flag. Your team works a clean, prioritized queue of structured loss records instead of a voicemail box.
Multilingual callers
ThunderPhone supports 47 voice languages with 123 voices. Set a primary language and additional languages for your service area, and the agent can switch mid-call when a caller changes to one of the selected languages — no separate line, no waiting for the one bilingual coordinator to be free. The same intake, safety, and escalation rules apply in every configured language, and every call produces a transcript the team can review.
Calls that hit someone else's phone tree
Outbound calls sometimes reach the other side's automation. The agent can wait on hold at a flat reduced rate and navigate keypad menus — useful for reaching a carrier's claims line or a TPA queue before handing off to your coordinator.
Sample restoration call flow
Agent: Thank you for calling Meridian Restoration. This line is answered around the clock. Are you calling about new property damage, an existing job, or something else?
Caller: Our water heater let go — there's water all over the basement and it's still coming.
Agent: I'm sorry — let's get help moving. First, is anyone in danger, and is there any fire, smoke, or gas odor? If so, please call 911 from a safe location first.
Caller: No, nothing like that. Just a lot of water.
Agent: Understood. Is the water still flowing, and have you been able to shut off the supply valve?
Caller: I just found the valve and shut it off.
Agent: Good. Is the water near any electrical outlets or panels?
Caller: It's close to the panel, but it's only about an inch deep.
Agent: Thank you. Can I get the property address, and roughly how many rooms are affected?
Caller: 42 Alder Court — the whole basement, maybe two rooms' worth.
Agent: Got it. Do you expect this to go through your homeowner's insurance? If you know the carrier, I'll note it for the technician — you don't need a claim number yet.
Caller: Probably, yes. I'd have to check which carrier.
Agent: No problem — the technician can walk you through that on site. I'm notifying our on-call team now with everything you've told me, and they will call you back at this number to confirm an arrival time. Is this the best number to reach you?
Every branch — the safety questions, the intake fields, what triggers a page versus a morning message, and what the agent may say about insurance — comes from the company's script. Whether the loss is covered is a question the agent never answers.
Features that matter in a restoration company
GDPR and HIPAA compliance posture
ThunderPhone's public compliance posture is GDPR and HIPAA compliant. The company still controls the workflow: limit the agent to necessary information, attach only approved documents, choose recording settings deliberately, and define where call data may be sent.
Dispatch and job-system connections
Agents can use connected apps, HTTP APIs, or MCP tools during a call — to create a loss record, page the on-call rotation, look up a job's status, or book an inspection when the receiving system exposes the required action. Webhook deliveries are HMAC-signed and retried automatically, so a briefly unavailable endpoint does not lose an emergency dispatch.
Knowledge grounded in your documents
Attach service descriptions, service-area boundaries, response-window policy, what-to-expect guides, and insurance-process FAQs — as text or PDF/DOCX uploads (up to 5 MB per text file and 50 MB per PDF or DOCX), or imported directly from a page on your website. The agent can search those documents mid-call instead of relying on a generic answer, and each agent uses only the documents you scope to it. Anything the material does not cover — any price included — is a callback or transfer, not a guess.
Your numbers, your telephony
Bring your existing emergency and office numbers through a supported VoIP connection or SIP configuration; verified numbers handle inbound and outbound calling. A web voice widget can put the same agent on your website, so a visitor comparing companies at midnight can start intake by voice without dialing.
Human handoff
Cold transfers are available for immediate routing. Warm transfers are available on eligible production VoIP or SIP numbers that can place an outbound leg: the caller waits while the agent privately briefs the recipient, then connects only if the recipient accepts — the right shape for waking an on-call technician with the loss details already summarized.
Testing before a real emergency hears it
Use a browser mic test for quick iteration, then run AI-caller simulations of the calls that worry you: a panicked homeowner with an active leak, a caller describing smoke, an adjuster with a supplement question, a customer frustrated about equipment still in the house. Simulations create transcripts and grading so the team can tighten the prompt before deployment, and graded scenarios can be kept as a regression suite so a prompt change never quietly breaks the safety script. When two intake scripts seem equally good, an A/B experiment on live traffic settles it. Every production call leaves a transcript, and a recording when recording is enabled.
Guardrails the company controls
- No coverage opinions, ever. The agent never tells a caller whether a loss is covered, never predicts what a carrier will approve, and never promises a claim outcome. Coverage determinations belong to the carrier and adjuster; the agent collects claim details and routes the conversation.
- Active danger means 911. Fire, smoke, gas odor, or any threat to life triggers your configured safety language — call 911 from a safe location, do not re-enter — and intake pauses. The agent never substitutes its own safety advice.
- Approved answers only. Pricing, response windows, and service-scope answers come from the prompt and scoped knowledge base; the agent quotes nothing you have not configured. Anything unanswered routes to a person instead of a guess.
- Escalation is configured, not improvised. What counts as an emergency, who gets paged, and the after-hours path are set by the company and exercised in simulation before launch.
What would this cost each month?
Suppose a restoration company handles 500 AI-assisted calls per month, averaging 4 minutes each:
500 calls × 4 minutes = 2,000 minutes per month
- Spark:
2,000 × $0.02 = $40/month - Bolt:
2,000 × $0.05 = $100/month - Storm:
2,000 × $0.09 = $180/month
That is base engine usage, not a quote — pay as you go, with no subscription. Premium voice or language paths can add up to 3¢ per minute, and phone-number or carrier charges depend on the telephony setup. Hold time bills a flat 2¢ per minute; voicemail and phone-menu minutes bill at the engine rate. Optional add-ons such as verbal acknowledgements, call supervision, and long prompts carry their own published surcharges. The agent builder shows the current all-in rate before calls are placed.
Frequently asked questions
Can a ThunderPhone agent answer our emergency line around the clock?
Yes — 24/7 inbound answering is the core use case. The agent runs your intake script on every call, captures a structured loss record, and follows your configured escalation path — a warm transfer to the on-call technician, an immediate dispatch message by webhook, or both. Non-urgent calls become categorized messages for the morning team.
How does the agent know what counts as an emergency?
You tell it. The company writes the classification rules — which loss types and caller answers trigger a page versus a logged message — and the agent applies them verbatim. Anything the script does not cover routes to a person.
Will it give safety advice during an active emergency?
Only your approved wording. If a caller describes fire, smoke, a gas odor, or a threat to life, the agent gives your configured 911 language — call from a safe location, do not re-enter — and does not continue intake while danger is active. It never improvises safety instructions.
Can it tell callers whether insurance will cover their damage?
No, by design. Coverage decisions belong to the caller's carrier and adjuster. The agent can collect the carrier, claim number, and adjuster details, explain your claims-assistance process in the general terms you approve, and route coverage questions to your team — but it never predicts or promises a claim outcome.
Can it handle calls from adjusters and TPAs?
Yes. The agent can identify the caller and claim, capture the reason for the call and any deadlines, and either transfer to whoever owns the file or create a structured callback task. It does not discuss scope, pricing, or supplements on its own.
Can it quote prices for restoration work?
Only what you configure. If your knowledge base includes approved figures — a service-call fee, for example — the agent can state them. Anything else, including job estimates, routes to your team: restoration pricing depends on inspection, and the agent is never allowed to guess.
What happens when fifty calls come in at once after a storm?
The agent answers them in parallel rather than one at a time. Every caller gets live intake instead of voicemail, and your team receives a prioritized set of structured loss records, with only the calls your rules flag escalated to the on-call crew.