Sort every enquiry into four lanes within the hour, active leaks first, then reported damage on owner-occupied roofs. AI can draft the sort and the replies; a person takes leaks and anything about insurance. Contact only people whose consent covers texts or AI calls, offer two inspection slots, and stop after five touches in fourteen days.
Most pages on this topic sell lists. The harder work starts after the hail: eighty enquiries in three days, six inspection slots a day, and a caller asking whether insurance will pay. Three rules handle that, and they are best written before the storm: who is called first, who you may text or ring with an AI voice, and what the assistant must refuse to discuss.
The first hour: what the AI does and what a person does
The first hours after hail contain three jobs: sorting the enquiries, replying to them and booking inspections. AI handles the first two well once your rules are written down. A person has to keep everything involving danger, money or insurance.
| Task | The AI | A person |
|---|---|---|
| Read each form, voicemail transcript or message and score it | Scores it against your rules and explains why | Checks every top-lane lead and anything unclear |
| Acknowledge the enquiry | Drafts a reply that says it is an AI and offers an opt-out | Phones the top lanes personally |
| Ask the triage questions | Asks them in a fixed order, by text or on an inbound call | Takes over on any handover trigger |
| Offer inspection slots | Offers two slots from a read-only view of your calendar | Confirms exceptions and double-bookings |
| Anything about insurance, price or a dispute | Declines politely and hands over | Handles it personally |
| Safety | Tells the customer to keep clear of damaged ceilings and not to go on the roof, then hands over | Sends an emergency crew where it is safe to do so |
Write the handover triggers as a short list the assistant can't argue with: water coming in now; a bulging or sagging ceiling; water near light fittings or sockets; any mention of an insurer, adjuster, excess or claim number; a contract already signed with another firm; a question about price; an angry or distressed customer; and the question "am I talking to a robot?", which always gets a truthful yes. The tutorial on AI receptionists for storm-day calls covers the phone side in more depth.
Consent before cleverness: who you may contact, and on which channel
Rules on automated texts and calls differ by region. They differ between marketing messages and messages about an enquiry the customer started, and they are still being tested in court. At least one regulator has ruled that calls made with AI-generated voices count as artificial voices, which pulls them into the stricter rules for recorded calls. I can't tell you what the law is where you work, and an adviser should, but I can tell you which starting position is safest for each kind of lead.
| Where the lead came from | Starting position | What to do |
|---|---|---|
| The customer rang you | They expect a reply about that enquiry | Ask permission to text during the call and record it |
| Your web form, consent box ticked by them | Consent as worded on the form that day | Use only the channels the wording names |
| The customer texted you first | A reply about that enquiry is usually expected, but what a reply may cover varies | Reply about the enquiry only; ask before anything more |
| A bought or shared list | No consent to you unless the seller can prove it | Hold, and ask for the consent record per lead |
| A door-knock card or flyer | Nothing, unless they filled in a form | Invite them to book through your own form |
| A past customer | Whatever they agreed at the time | Check the record; refresh consent where it is thin |
Some software enforces part of this for you. Roofr's help centre says that to send a proposal by text you must be registered for Roofr Texting and the customer contact must have opted in to text communication. If your tools have an equivalent opt-in field, treat it as the consent flag your assistant must check before it writes a text. Until your adviser says otherwise, treat written, specific consent that names your company and the channel as the safe standard.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowPrior-consent wording to take to your adviser
Three places collect consent: the web form, the phone call and the first text. Here is wording for each (illustrative, not legal advice; an adviser who knows your rules should settle the final text). The form box must start unticked.
WEB FORM (unticked box)
[ ] I agree that [COMPANY NAME] may contact me about my roof enquiry by phone call and text message, including messages and calls that use automated or AI-assisted systems, at the number I have given. I can withdraw this at any time by replying STOP or telling you. Agreeing is not a condition of getting an inspection.
PHONE (read aloud, then record the answer)
"May we text you inspection details and updates about your roof? Some of our messages are written with an AI assistant, and you can say stop at any time."
FIRST TEXT (only after a yes)
Hi [first name], this is the assistant for [COMPANY NAME] (I am an AI). Thanks for getting in touch about your roof. May I ask a few quick questions so we can book your inspection? Reply STOP to opt out.Wording is half the job. The other half is a record you can find in thirty seconds, because a complaint arrives weeks later and the question is always "when did they agree, and to what?". A simple sheet will do, one row per contact (illustrative):
| Contact | Source | Date and time | Wording version | Channels agreed | Opt-out |
|---|---|---|---|---|---|
| 0412 | Web form | Day 1, 14:07 | Form v2 | Call, text, AI-assisted | None |
| 0413 | Inbound call | Day 1, 14:22 | Phone script v1 | Text only | None |
| 0398 | Bought list | Not known | None | None: hold | Not applicable |
| 0377 | Web form | Day 1, 09:51 | Form v2 | Call, text | Day 3, 16:40 |
Treat any clear refusal as an opt-out, in any wording ("stop", "don't text me", "remove me"), and make sure it reaches every tool, not only the one that received it. An AI assistant that has been told to stop in the chat window but still sits on the follow-up schedule is the classic way to turn a polite customer into a complaint.
The hail triage script: eight questions in a fixed order
A fixed order lets a person, a chatbot or an inbound AI voice collect the same facts, so the scoring afterwards compares like with like. Question one is the safety gate, and a yes stops everything else.
1. Is water coming into the building right now?
If yes: stop. Keep clear of any dripping or bulging ceiling and anything electrical, do not go on the roof, and hand over to a person.
2. Are you the owner, or do you rent? (If renting: who owns it?)
3. When did the hail fall, and how big was it? (Pea, marble, golf ball, tennis ball: use their word.)
4. What have you noticed since? (Dented gutters or vents, missing or cracked shingles, stains inside, nothing yet.)
5. Roughly how old is the roof, if you know?
6. Do you have photos from the ground? (Send an upload link. Never ask anyone to climb up.)
7. Which days and times suit an inspection?
8. May we text you the details? (Record yes or no, with the time.)
If asked about insurance, cost or what is covered: "I can't advise on insurance or prices. Our inspector will give you a written report with photos that you can share with anyone you choose."Each question earns its place. One sets the lane and the safety path. Two tells you whether the person can authorise work. Three and four give the signals for scoring, and the size in the caller's own words matters more than a number, because it is a rough sort and never a diagnosis; only an inspection decides what the roof needs. Five and six prepare the visit. Seven feeds the slot offer, and eight is the consent record. Keep the script to these eight, because every extra question loses callers who are already anxious.
Priority rules: score the lead, then sort into four lanes
Scoring turns a pile of enquiries into an order of work. These weights are my starting point (illustrative); adjust them after the first storm, when you can see which leads turned into jobs. Your CRM may already score leads, and AccuLynx's home page, for instance, advertises AI-powered lead intelligence, but no software scores on the hail answers unless you collect them first.
| Signal | Points |
|---|---|
| Water coming in now, or a new ceiling stain | Lane 1 straight away |
| Damage reported: dented gutters or vents, missing or cracked shingles | +3 |
| Hail described as golf ball size or larger | +3 |
| Hail described as marble size | +1 |
| Owner-occupied | +2 |
| Roof 15 years old or more, or age unknown | +1 |
| Ground-level photos supplied | +1 |
| Asked for a particular day | +1 |
| Renting, or not the person who decides | -2 |
Lane 1 is any active leak or structural worry. Lane 2 is a score of 7 or more. Lane 3 is 3 to 6. Lane 4 is below 3. Try three leads. The first reports a bedroom ceiling stain that appeared overnight: Lane 1, whatever else the score says. The second is an owner (+2) who saw golf ball hail (+3), has dented gutters (+3) and an 18-year-old roof (+1), which makes 9 and Lane 2. The third rents (-2), saw pea-sized hail (0) and has noticed nothing (0), which makes -2 and Lane 4. The reply to that third person asks them to pass the details to the landlord, because the landlord has to book the inspection.
Prompt: score an enquiry and draft a safe first reply
Put the rules above in a chat project, together with the triage script and the consent ledger columns, and use this standing instruction. For a website version of the same sort, see the comparison of roofing chatbot platforms.
You triage roofing enquiries after a hail storm for [COMPANY NAME]. Use only the scoring rules, triage script and handover triggers in this project.
For each enquiry I paste, return:
LANE: 1, 2, 3 or 4, with the points added up line by line.
MISSING: which of the eight triage questions are still unanswered.
HANDOVER: yes or no, naming the trigger.
CONSENT: copy the consent status I give you. If it is not YES for texts, write a call script for a person instead of a text.
REPLY: a text of 60 words or fewer. Say you are an AI assistant. Include an opt-out. Offer no more than two inspection slots, and only from the slots I paste. Never mention insurance cover, claims, price or timing you were not given.
ENQUIRY: [paste, with phone number and address removed]
CONSENT STATUS: [YES for text and call / TEXT ONLY / NO]
FREE SLOTS: [paste two]Run it on this enquiry (illustrative): "Hail hit us last night, golf ball size. Gutters dented and there's a stain on the bedroom ceiling that wasn't there yesterday. Roof is about 18. House is mine. Can someone come Thursday?" A typical first draft of the reply reads:
Hi [first name], sorry to hear about the hail! We can absolutely book you in on Thursday, and we will help with the insurance claim too. A new ceiling stain could mean an active leak, so we will prioritise you.Three faults. It promises Thursday, but the assistant was never given Thursday's slots. It offers help with the insurance claim, which is claim advice and a promise in one phrase. And it never says it is an AI or offers an opt-out. The version a person approves:
Hi [first name], this is the assistant for [COMPANY NAME] (I am an AI). A new ceiling stain can mean water is still coming in, so a member of our team will ring you within 15 minutes. Until then, please keep clear of the stained area and don't go on the roof. Reply STOP to opt out of texts.The lane is 1 because of the new stain, so the assistant's real job here was recognising it and getting out of the way. Notice what the approved text omits: no slots, no price, no mention of insurance. Slots come later, from a person who has heard the details.
Inspection slots: eighty enquiries, six slots a day
Capacity, not interest, is the limit after hail. Picture a small firm with one inspector who can do six inspections a day. Three days after the storm it has 80 enquiries, and triage has sorted them: 6 in Lane 1, 24 in Lane 2, 30 in Lane 3 and 20 in Lane 4 (illustrative). The arithmetic is plain: 80 inspections at six a day is 14 working days of one inspector's time once partial days are rounded up.
| Lane | Leads | Slots at 6 a day | What customers are told |
|---|---|---|---|
| 1 | 6 | Day 1 | A person rings within 15 minutes; temporary protection the same day where it is safe |
| 2 | 24 | Days 2 to 5 | Two slots offered from days 2 to 5 |
| 3 | 30 | Days 6 to 10 | Two slots offered from days 6 to 10 |
| 4 | 20 | Days 11 to 14 | On the waiting list; messaged when a slot opens |
Telling Lane 3 customers "days 6 to 10" on day one feels harsh, yet it beats offering Thursday and cancelling it. The assistant should quote windows only from your calendar, and say plainly when the earliest slot is a week away. Three inspectors instead of one turn 14 days into about five, so it can be worth borrowing a crew member for the first week. Either way, rewrite the lane windows daily from the real calendar rather than leaving them in a template.
Two habits protect the slots. Ask every booked customer to confirm by text, and release any slot not confirmed 24 hours ahead to the next person in the same lane. And hold a slot for two hours at most while you wait for a reply, or the calendar fills with maybes. Once the inspection happens, the written report and the quote follow; the tutorial on adding AI-written scopes and warranty wording covers the proposal that comes next.
Follow-up timing: five touches in fourteen days, then stop
Two things can stall a hail enquiry: the customer books someone else, or they wait on their insurer and drift. A short, capped sequence handles both without turning into pestering. This is my suggested starting plan (illustrative); your own logs will show where to move it.
| Touch | When | Channel and who | Content |
|---|---|---|---|
| 1 | Within minutes, in business hours | Text from the assistant if consented; thank-you page otherwise | AI disclosure, one question, opt-out |
| 2 | Within 15 minutes for Lanes 1 and 2; same day otherwise | A person rings | Triage script, slots |
| 3 | Two hours after a missed call | Text, if consented | "Sorry we missed you" and a booking link |
| 4 | Day 3 | Text, if consented | Next two free slots in their lane |
| 5 | Day 14 | Text if consented; otherwise a person rings once | A polite close that says how to book later |
Stop the sequence on a booking, a reply asking you to stop, or any opt-out wording. Choose sending hours inside the quiet-hours rules where you work, and keep them dull: daytime, weekdays first. Here are the touch 4 and touch 5 messages (illustrative), and note how neither pretends urgency the roof doesn't have:
DAY 3
Hi [first name], [COMPANY NAME]'s assistant (an AI) again. Our next free inspection slots are [SLOT 1] and [SLOT 2]. Reply 1 or 2 to book, or STOP to opt out.
DAY 14
Hi [first name], we haven't heard back, so we will stop messaging now. If you would like an inspection later, reply to this message or call [NUMBER] and a person will help.The first reply is the part you control, and the tutorial on replying to every enquiry in under five minutes sets up that side. Missed calls during a storm are their own problem, and missed call text back covers the setup, including the consent question it raises.
What the assistant must never say about claims
After hail, the question behind half the enquiries is "will my insurer pay?" It is also the question an AI is most tempted to answer, because it sounds helpful. Rules about what a contractor may say or offer around insurance claims differ by region and can be strict. In some places they cover rebates of an excess or deductible, and acting for the customer in the claim. The simplest safe policy is one sentence the assistant repeats: it can't advise on claims or prices, the inspector will provide a written report with photos, and the customer may share it with anyone.
Put a short ban list in the prompt and in your review checklist: covered, approved, free roof, "we'll handle your claim", guaranteed approval, and any promise to absorb the excess or deductible. The slip to watch for is a warm reassurance rather than a deliberate claim. A customer writes "I'm so worried they won't pay" and a draft answers "Don't worry, claims like this are normally accepted". Nobody on your side knows that, and the customer will remember the sentence.
Handled properly, the exchange runs like this (illustrative):
Customer: Will my insurance cover this?
Assistant: I can't advise on insurance or prices. What I can do is book an inspection. Our inspector will send you a written report with photos that you can share with anyone you choose. Would [SLOT 1] or [SLOT 2] suit you?If the customer presses ("but do you think they'll pay?"), the assistant repeats the same sentence once and then hands over to a person who knows your local rules. Repeating is better than improvising.
Buying storm leads: three consent questions to ask before you pay
Lead sellers' pages describe volume and speed. They rarely say what each person agreed to, which is the part that decides whether your AI texts and calls are allowed. Before you pay, ask the seller:
- Whose consent is it? Does the wording the person saw name your company, or only "partners" or "service providers"?
- When and how? Can the seller show the date, time and page where each person agreed, for every lead you buy?
- To what? Does the wording cover automated or AI-assisted texts and calls, or only a call from a person?
Then do the sum on cost. If a list charges $40 a lead and one lead in ten books an inspection, each inspection costs $400 before anyone has driven anywhere (illustrative figures). Compare that with the leads from your own form, where the consent record is yours. The buying guide for AI lead generation in roofing compares pay-per-appointment and retainer models.
Five numbers to check after the first storm
After the first storm, spend half an hour with the ledger and the call log and work out five figures. Suppose 80 enquiries produced the following (illustrative):
- Time to first contact: the median across all 80. For Lanes 1 and 2 it should be minutes, not hours.
- Contact rate: 68 of 80 reached, which is 85 per cent.
- Booking rate: 52 of 80 booked, which is 65 per cent. Compare it by lane; a low Lane 2 figure means the slot offer or the timing needs work.
- Opt-outs and complaints: 3 opt-outs, 3.75 per cent, and none about the assistant. More than a handful points at the follow-up cadence or the wording.
- Handovers: how many conversations went to a person, and why. A rising count of "insurance" handovers suggests your first message should say earlier that you can't advise on it.
Change one thing at a time and write down the date. The weights in the scoring table are the likeliest to need adjusting, since the first storm will show which signals really predicted a booking. Then update the prompt and the script together, so the rules a person follows and the rules the assistant follows stay identical.
Storm lead questions roofers ask about AI and consent
Can an AI voice agent phone storm leads I bought?
Treat that as a no until an adviser says otherwise. Consent usually has to come from the person, name your company and cover the channel, and at least one regulator has ruled that AI-generated voices count as artificial voices. Ask the seller for the consent record for each lead, and check do-not-call rules before anyone, human or AI, dials.
How fast should I reply to a storm lead?
Aim for a first reply within minutes in business hours: a short text, or a call for leaks and likely damage. After hours, an instant acknowledgement saying when a person will ring is enough. I can't point to a verified figure for an exact optimum, so use your own call logs after the first storm to see where contact rates fall away.
What should the AI say when a customer asks if insurance will pay?
It should say it can't advise on insurance or cost, offer the inspection, and explain that the inspector will send a written report with photos the customer can share with anyone. It must not say what is covered, promise approval or offer to deal with the insurer. If the customer wants more, it hands over to a person.
How many follow-up messages are too many?
Five touches in fourteen days is a sensible ceiling for people who haven't booked: an instant acknowledgement, a call from a person, a text after a missed call, a day-3 message with two free slots and a day-14 closing note. Stop at once on a booking or any opt-out wording, not only the word STOP. Check the quiet-hours rules where you work before choosing send times.
Further reads
- Roofing Google Ads: Storm-Season Campaigns and Negative Keywords With AI — Where many storm leads come from, and how to filter the wrong ones.
- AI Marketing for Roofing Companies: A Worked Storm-Season Content Plan — Posts and pages to publish while the hail is still news.
- Door-to-Door Roofing Sales: Scripts and Follow-Up Texts Written With AI — Follow-up texts for leads collected on doorsteps.
- Roofr's AI Receptionist: Is the Add-On Worth It for a Small Roofer? — What an AI receptionist add-on costs against missed storm calls.
- What Can AI Agents Do for Roofers Between Lead and Signed Contract? — What agents can handle between first enquiry and signed contract.
- Roofing Social Media: Turn Finished Roofs Into a Month of Posts With AI — Turn finished storm repairs into posts, with photo consent.
- How to Respond to Negative Roofing Reviews With AI: Claims and Delays — Replies for complaints about waiting times after a storm.
- AI Automation for Roofers: Six Workflows Worth Setting Up First — Call answering, estimate follow-up, measurements, instant estimates, inspection reports and job close: what each costs, how to start and what to watch.
- AccuLynx vs JobNimbus for a Small Roofer: What to Ask About AI — AccuLynx and JobNimbus are sold by quote. A five-person crew's question list, with Roofr and Jobber prices to hold their quotes against.
- Foundation Repair Leads: Screen Crack Photos With AI Before the Visit — Six intake questions, a four-photo request, an AI summary that describes but never diagnoses, an engineer-referral rule, reply scripts and the lead-cost sums.
- Roofers AI: What a Starter Setup Costs for a Small Roofing Company — Three priced bundles for roofers AI, from about $20 a month to a roofing suite, plus what each covers, the hidden costs and break-even sums.
- Roofr vs EagleView vs Hover: AI Roof Reports, Cost per Job, Turnaround — Published per-report prices for Roofr, EagleView and Hover, annual cost at three volumes, turnaround, file formats and a three-roof accuracy test.
- Which AI Tools Do Roofers Need? Measurements, Calls, Estimates — Roofr, EagleView and Hover reports priced for 25 roofs a month, plus AI receptionist, instant estimator and estimate-writing costs and a decision by budget.
- AI in Roofing: Weather-Delay Messages and Rescheduling Checklist — A roofing weather-delay checklist: five customer notices, a crew rescheduling routine and a named forecast, with the foreman making every go or no-go call.
- ChatGPT for Roofing: 10 Prompts for Estimates, Inspections and Warranty — Ten ChatGPT prompts for a roofing company's inspections, estimates and warranty replies, each with a filled example and a check before it is sent.
- Gutter Cleaning Marketing: Eight Spring and Autumn Campaigns With AI — A Leaf Week calendar of eight campaigns, with AI prompts, sample copy and a tracking sheet so you know which gutter cleaning marketing booked jobs.
- Roof Damage Reports for Insurance Claims: Let AI Document, Don't Adjust — Photo order, caption wording and a measurements log that let AI organise roof damage evidence without drifting into coverage opinions.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Roofr help-centre article on creating a proposal (texting and opt-in); AccuLynx home page; a regulator's February 2024 ruling on AI-generated voices; legal commentary on later court decisions about consent.