AI Agent for Chimney Sweeps: Fill the Fall Calendar Without Missing Calls
Chimney sweep demand collapses into about 20 weeks a year. From late August through the first hard freeze the phone will not stop ringing, and the rest of the year the schedule looks half empty. An AI agent on phone, WhatsApp, and web chat is what lets a 2-to-5-truck operation pick up every one of those fall calls, trigger the annual recall that fills January, and stop losing dormant customers to the sweep across town.
Why the chimney sweep calendar is broken
Vendor data and agency estimates line up on the same shape: bookings run hot from September through January and go quiet from April to August. One field-service vendor describes the fall rush as starting in late August and clearing out the calendar by the first cold snap. Smart Service recommends locking in seasonal hires by August at the latest because a new tech needs six to eight weeks of ride-alongs before they can run a Level 1 inspection alone.
That seasonal shape creates four problems the owner feels every year:
- Missed fall calls. One owner on the truck plus one person at the desk cannot answer 40 to 80 inbound calls per day in October. Calls that go to voicemail at 8:15 a.m. become bookings for the competitor by 9:00 a.m.
- No annual recall. NFPA 211 recommends an annual inspection for any venting system that burns solid fuel. Most sweeps never contact last year's customer in month 11. The customer books whoever shows up first in a search result.
- Dormant customer database. A sweep with 5 years of jobs is sitting on 2,000 to 4,000 households who have not been touched. Industry reactivation benchmarks run from 5 to 15 percent on a single campaign.
- Off-season cliff. April through August the phone rings maybe 6 times a day. Dryer vent, animal removal, waterproofing, cap and damper repair all exist as revenue, but the business is not marketing them and the office does not pitch them.
An AI agent does not fix the seasonality itself. It fixes the four leaks around it: the missed calls at peak, the recall nobody sends, the dormant list nobody works, and the off-season services nobody pitches.
What "AI agent" means for a 2-to-5-truck sweep
In this article an AI agent is the front desk, nothing more. It is a voice agent on the main business line, a WhatsApp Business API number on the website and Google Business Profile, a web chat on the site, and a shared layer behind them that books into the field-service calendar, writes into the CRM, and hands off to a human when the question is out of scope. It is not a phone tree, a generic voicemail-to-text tool, or a social-media auto-reply.
Built right, one agent handles all seven front-desk jobs below. Built wrong, you end up with three disconnected tools, three different voices, and a dispatcher who still has to re-ask every customer for their address.
The 7 jobs the AI agent actually does
- First-ring pickup at peak. Target: answer under 3 seconds, every call, 24/7, in English and Spanish. No ring groups, no voicemail during business hours. The agent captures name, address, fireplace type (wood, gas, pellet, insert), last sweep date, reason for the call, and any smoke or draft complaint.
- Qualification and routing. Routine Level 1 sweep plus visual inspection goes straight onto the calendar at the closest open slot. A real-estate sale, post-fire, post-quake, or appliance swap gets flagged as Level 2 (camera scan required) and routed to the senior tech. Suspected creosote stage 3 or a visible crack in the clay flue tile gets a "we need to see this before we quote" booking at a 30-minute diagnostic rate.
- Calendar-aware booking with deposit. The agent reads the field-service calendar's real availability, respects drive-zone rules (a 90-minute drive to a $200 sweep is not acceptable in October), and takes a $39 to $75 non-refundable deposit at booking. A Smart Service case study calls deposits the single most underused booking discipline in the category; a Contractor Plus case credits pre-pay for a sweep booking 6 months of fall work by mid-summer.
- Confirmation and no-show prevention. Three automated touches: 72 hours before (confirm the appointment and ask the homeowner to clear 4 feet around the hearth), 24 hours before (restate address and tech name), and 1 hour before (tech en route with ETA and a map link). Over WhatsApp the 24-hour touch typically clears north of 90 percent open rate; the 1-hour touch cuts the "nobody home" ride by more than half.
- Annual recall at month 11. Every homeowner who had a service last year gets a WhatsApp or SMS at month 11 with a single tap-to-book link and a $25 early-bird credit if they book before July 31. This is the single highest-ROI automation in the stack. A sweep with 1,800 active customers who books 22 percent of them at $285 average ticket is pulling $112k in work scheduled before Labor Day, which is when the fall rush starts.
- Dormant reactivation at 18 to 36 months. A sweep that has been running 5+ years has 2,000 to 4,000 households who skipped at least one year. A quarterly reactivation campaign pulled at the 18-, 24-, and 36-month marks typically recovers 5 to 10 percent per run with a modest offer ($20 off, or $99 cap-and-damper check-up). This is the April-through-August bread and butter.
- Off-season upsell. The same agent asks dryer vent, animal removal, waterproofing, and cap-replacement questions year-round. Dryer vent cleaning has a 12 to 18-month recall cycle that overlaps the quiet summer months; a $189 service on the back of a $285 sweep is pure margin. The agent does not negotiate price. It books a diagnostic or a flat-rate add-on and nothing else.
Case study: 3-truck Columbus, Ohio operation, 90 days
A CSIA-certified sweep operating three trucks out of Columbus, Ohio, running mostly within Franklin and Delaware counties. Pre-period: 2,180 lifetime customers, one owner-operator plus one dispatcher, 2 full-time techs, 1 seasonal tech from September through December. Average ticket $285 for sweep plus Level 1, $420 for Level 2 with camera, $189 for dryer vent. Shoulder-season (April to August) averaging 42 completed jobs per week. Fall (September to January) averaging 108 completed jobs per week. Lost-call rate measured at 31 percent in October.
They launched the AI agent on 2026-07-28, 5 weeks before the fall ramp. Measured 90-day window: 2026-07-28 to 2026-10-26. Comparison window: same 90 days of 2025.
Swipe to see the full table →
| Metric | Before | After (90 days) | Change |
|---|---|---|---|
| Business-hours call pickup | 72% | 99% | +27 pts |
| After-hours pickup | 0% | 96% | +96 pts |
| Median first reply | 3h 12m | 18s | -99% |
| Call-to-booking rate | 38% | 57% | +50% |
| No-show rate | 14% | 6% | -57% |
| Deposit-at-booking take rate | 0% | 88% | +88 pts |
| Annual recall bookings (month 11) | 7% | 24% | +243% |
| Dormant 18-36mo reactivations | N/A | 142 jobs | new |
| Fall calendar filled by Aug 15 | 52% | 91% | +39 pts |
| Dispatcher hours/week | 44 | 19 | -57% |
Dollar math over the 90 days:
- 142 dormant reactivations at a blended $264 ticket = $37,488 in work that was not on the calendar in 2025
- Annual recall bookings went from 12 (7% of 171 eligible last year) to 48 (24% of 200 eligible this year); 36 incremental sweeps at $285 = $10,260
- Call-to-booking rate lift from 38% to 57% on roughly 1,840 inbound calls = +350 extra bookings; at a conservative $200 contribution margin = ~$70,000
- No-show rate from 14% to 6% on ~1,300 completed appointments = ~104 fewer wasted ride-outs; at a $95 drive-and-time cost = ~$9,900 avoided
Total measurable lift over the 90 days: roughly $127,000 in bookings that were not in the 2025 window, against a setup fee of $4,900 and $1,100 per month in software and WhatsApp conversation costs. The owner moved the dispatcher off the phone and onto estimates-and-invoicing, which is where the business was bleeding another half-week of unbilled service agreements.
Reference stack: what actually ties this together
You do not need all of it. You need one tool per row, chosen to integrate with the field-service software you already run.
Swipe to see the full table →
| Layer | Options | Notes |
|---|---|---|
| Field-service CRM | ServiceTitan, Housecall Pro ($59+/mo), Jobber ($39+/mo), Smart Service, FieldEdge | Housecall Pro and Jobber fit 1-to-5 trucks; ServiceTitan starts to make sense around 6+ trucks. Confirm calendar-write API access before you sign. |
| Voice agent | Vapi + Claude, Retell AI, Thoughtly, Dialzara, MyAIFrontDesk, Sameday | Pick one that supports a warm handoff to a human extension for Level 2 flags. |
| WhatsApp layer | Meta Cloud API direct, 360dialog, Twilio BSP | Required for template messages on recall and confirmations. Budget ~$0.014 to $0.08 per utility conversation in the US. |
| SMS fallback | Twilio with 10DLC registered campaign, Telnyx | 10DLC registration takes 2 to 4 weeks. Start it on day one of the project. |
| Web chat | Same agent surfaced as a web widget, or Chatbase / Intercom with the same backend | Match the voice to the phone agent. Do not run two separate scripts. |
| Payments at booking | Stripe, Square, QuickBooks Payments | Non-refundable deposit of $39 to $75 is the number that moves no-shows the most. |
| Inspection reports | Jobber custom forms, SweepSmart, or a tablet-first PDF template | Attach to the customer record so the next year's recall knows what was found. |
| Review funnel | Podium, Birdeye, NiceJob, or a simple 48-hour post-service WhatsApp template | 5 to 15 new reviews per quarter is the floor to move local-pack ranking. |
Explicit refuse-list: what the agent should never do
A good AI agent for a chimney sweep is boring on purpose. It books, confirms, and hands off. It does not improvise on any of the following:
- No firm quotes on anything beyond flat-rate services (standard sweep plus Level 1, dryer vent, cap replacement with a pre-measured cap). Masonry repair, relining, insert install, and fire-damage work all require an onsite estimate.
- No safety advice. If a caller describes active smoke, carbon monoxide symptoms, or a chimney fire in progress, the agent says "hang up and call 911" and texts a 911 reminder. No diagnosis over the phone.
- No code interpretation. NFPA 211 and local code questions get a "our tech will confirm onsite" response. The agent does not state a clearance, a flue size, or a liner requirement.
- No insurance-claim negotiation. Post-fire or post-storm work routes to a human estimator who handles the adjuster.
- No HOA or landlord authority questions. If the caller is not the homeowner or legal tenant, the agent requires written authorization before scheduling.
- No pricing negotiation. Published flat rates are the floor. Discounts route to the owner.
Write these rules into the system prompt and keep them there. Every time the agent gets caught making up a quote or a code answer, you lose a customer in a way that is harder to measure than a missed call.
2026 US pricing bands
What a sweep should expect to pay for the full stack above, not counting the field-service CRM you already have:
- Entry point: $199 to $499 per month plus $499 to $1,500 one-time setup. Voice agent with basic WhatsApp and SMS, no deep calendar writeback, no annual-recall automation. Fine for a 1-truck operator who wants to stop missing after-hours calls.
- Mid-tier: $599 to $1,200 per month plus $1,500 to $4,900 setup. Voice + WhatsApp + web chat under one agent, calendar writeback into Jobber or Housecall Pro, deposit at booking, annual recall automation, dormant reactivation sequences. Fits 2-to-5 trucks.
- Custom build: $1,200+ per month plus $4,900 to $15,000 setup. Multi-location, call-recording QA, custom qualification tree, ServiceTitan or FieldEdge integration, bilingual voice, HIPAA-adjacent vendor compliance for insurance work. 6+ trucks.
For reference, a human answering service runs $1.25 to $2.80 per minute (roughly $1,100 to $2,400 per month at typical fall volumes and nothing in the shoulder season), and a full-time dispatcher runs $38k to $55k per year plus benefits. A mid-tier AI agent pays for itself on missed-call recovery alone in the first full fall.
2-week rollout, WhatsApp first, voice last
The mistake owners make is turning on the voice agent first. The voice stack has the most brittle failure modes (a bad transcription on an address kills the booking) and the highest stakes (a mishandled fire call is a public complaint). Start with the channel where mistakes are quietly fixable.
Week 1: data, WhatsApp, and the recall list.
- Day 1-2: export the last 3 years of customer records from the field-service CRM. Normalize phone numbers, flag opt-outs, tag by last-service date and fireplace type.
- Day 3: provision the WhatsApp Business API number and submit the 3 templates you need (annual recall, 24h confirmation, dormant reactivation). Approval runs 24 to 72 hours.
- Day 4: start 10DLC registration for SMS fallback. This runs in the background for the next 2 to 4 weeks.
- Day 5-7: wire the WhatsApp agent to the calendar. Test 20 real booking flows. Catch every edge case (apartment, HOA, unlisted street) before going live.
Week 2: web chat, voice, and the live run.
- Day 8-9: web chat widget live on the site and Google Business Profile. Same backend as WhatsApp, so no duplicated scripts.
- Day 10-11: voice agent live on a forwarded number (not the main business line yet). Shadow the dispatcher for 2 days. Review every call transcript end of day.
- Day 12: switch the main business line to the voice agent for after-hours only. Keep the dispatcher on daytime for one more week.
- Day 13-14: fire the annual recall campaign to everyone with a last-service date between 10 and 13 months ago. Fire the dormant reactivation to everyone between 18 and 36 months.
Day 30 review criteria: pickup rate above 95 percent on business hours and above 85 percent after-hours, call-to-booking above 45 percent, deposit take-rate above 75 percent, and no customer complaint about the agent misidentifying itself. Any one of those fails, the owner pulls the voice agent back to after-hours only and the vendor fixes before re-enabling.
For the closely related front-desk build on the voice-first side, see our breakdown of the AI receptionist pattern for solar energy, and for the CRM-side cousin the AI-enabled roofing CRM. The shape of the problem is the same; the qualification tree and the refuse-list are what change by trade.
Want the AI agent running in your sweep business before the fall rush?
We build the CRM, the voice agent, and the WhatsApp layer around your existing field-service software. 2-week rollout. Deposit at booking. Annual recall automation on day one.
Message us on WhatsApp Book a 30-min call