At 9:47 p.m. on a Tuesday in January, a homeowner with a dead furnace and a 52-degree house picks up the phone. The shop's answering service in Tampa charges $1.85 a minute, books one in three after a 4-minute hand-off, and reads the address back wrong half the time. The homeowner hangs up. By 7:20 a.m. Wednesday they are on a competitor's truck because the competitor's Avoca-style AI receptionist answered on the second ring, captured the booking, and texted the on-call tech only because the symptom met the emergency rule. The shop with the answering service captured 0-15% of after-hours calls at $1,140 of monthly invoice cost. The shop with AI captures 80%+ for $1K-$3K/month — and the leak shrinks from 22% missed-call rate to under 5%. This lesson is the build instruction for the after-hours AI workflow that replaces the answering service: the Avoca-style, Jobber AI Receptionist, or Housecall Pro AI Agents stack that books direct to ServiceTitan / Sera / HCP after hours, warm-transfers in-hours, and texts the on-call tech only for true emergencies. The named workflow targets cutting missed-call leakage from 22% to under 5%, lifting after-hours capture from 0-15% to 80%+, and protecting the on-call tech's sleep on every call that does not meet the emergency rule.

Why the Answering Service Is a 22% Leak That Costs More Than the Invoice

Most residential trades shops still run a third-party answering service for after-hours calls. The monthly invoice — typically $800-$1,500 per month at $1.50-$2.00 per minute, depending on call volume — is the visible cost. The invisible cost is the booking-rate disaster. Industry data on traditional answering services consistently shows 25-35% booking rates: the operator answers, takes a message, occasionally schedules a callback, and hands the lead to the shop's voicemail or CSR queue for the next morning. By the time the morning CSR returns the call, 6-12 hours have passed, the homeowner has called the next shop, and the lead is gone. Aggregate after-hours capture across the industry baseline lands at 0-15% — the answering service captures so few that the shop's effective after-hours leak is essentially total.

Layer that 22% missed-call number from L1 against the after-hours math. At a 7-truck residential HVAC shop with $387 average diagnostic-plus-repair ticket, after-hours calls are typically 18-25% of total inbound by volume. Multiply: roughly 12-18 after-hours calls per week × 87% leak rate × $387 average ticket = $4,000-$6,000 per week of pure margin walking out, $208K-$312K annualized. The answering service invoice covers a fraction of one week's leak. The shop is paying twice — once for the service, again for the leak the service does not close.

The leak compounds beyond the single lost call. A homeowner calling at 9:47 p.m. with a dead furnace and a 52-degree house who reaches voicemail is not a one-time-lost lead. They are a permanent customer transfer to the competitor that picked up. They will not call back. They will recommend the competitor to neighbors (NPS effect). They will call the competitor for their next service in 14 months (CLV effect). One after-hours leak at $387 ticket compounds to $1,500-$3,500 of forgone lifetime revenue with referrals. The 7-truck shop's annualized $208K-$312K visible leak roughly doubles when CLV and referral effects are modeled out — call it $400K-$600K of true margin impact. The answering service is the most expensive cheap solution in the trades.

The Three-Product Bake-Off for After-Hours AI

Three AI products dominate the after-hours surface in 2026 for residential trades. Each one has a different bet, a different integration story, and a different fit by shop size and FSM platform. The decision tree is not "which is best" — it is "which fits the shop."

Avoca — The Bolt-On Standard for After-Hours

Avoca is the bolt-on AI receptionist that pioneered the residential trades after-hours capture market. The April 2026 Series B at $125M+ and $1B valuation (led by Meritech and General Catalyst with Series A money from Kleiner Perkins) priced Avoca's category-defining position. After-hours is where Avoca's value is largest because the calls Avoca handles after-hours are almost entirely calls that would have leaked through an answering service. The HL Bowman case study Avoca published shows after-hours capture climbing to 80%+ within 60 days of deployment, with cost-per-conversion collapsing from $350 to $215 (39% drop) and 70% YoY revenue growth in the case window. Avoca's pricing lands at $1K-$3K per month for a 6-12 truck residential shop, which means the payback on after-hours alone runs 2-4 weeks at typical leak math.

The Avoca after-hours flow: inbound rings the shop's phone tree, after-hours routing engages Avoca, the AI handles the call end-to-end (booking direct to ServiceTitan / Sera / HCP via API), texts the homeowner the confirmation, and texts the on-call tech only when the system prompt's emergency rules trigger. The CSR floor reads the AI-handled call notes at 7:30 a.m. and absorbs the booked-overnight load into the morning routing. Avoca's depth — hundreds of hours of trades-call training, tuned escalation rules, calibrated rebuttal handling — is the reason it commands a price premium versus FSM-native alternatives. For shops with material after-hours volume and significant leak, the depth is worth the premium.

Jobber AI Receptionist — For 1-8 Truck Shops

Jobber's AI Receptionist (Copilot bundle, $99/month add-on on the Plus plan) is the in-platform after-hours engine for Jobber shops. It launched broadly in 2026 across the $1M-$3M residential service segment and is built for shops that are too small to justify Avoca's price band. The integration is native: bookings drop into the Jobber schedule, customer records update automatically, the on-my-way text fires from the same Jobber automation engine the shop already uses. Trade-offs versus Avoca are real — fewer hours of trades-call training, a more conservative escalation policy (Jobber AI Receptionist defaults to voicemail handoff on edge cases where Avoca would attempt to book), and slightly less aggressive rebuttal handling on price-shopper objections after-hours. For a 4-truck plumbing shop running Jobber, the AI Receptionist is the rational first AI bet — low setup friction, low monthly cost, documented after-hours capture lift of 8-14 points in published 2026 case studies at $1M-$3M shops.

Housecall Pro AI Agents — For HCP Shops

Housecall Pro's AI Agents module launched broadly in 2026 across the $1M-$15M HVAC, plumbing, and electrical segment. Same in-software logic as Jobber: after-hours bookings land in HCP, customer history pulls in real time, the dispatcher sees the booked-overnight load on the same board they have always used. The 2026 HCP case studies show after-hours capture rising from 12% to 70%+ within 60 days, alongside the 8-14 point booking-% lift documented for HCP AI Agents in routine inbound. Pricing rolls into HCP's plan tiers — aggregate cost typically falls between Jobber and Avoca. The strength for HCP shops is integration tightness and a single vendor relationship; the trade-off versus Avoca is the same as Jobber's — shallower training depth on the highest-volume objection patterns.

Decision rule: 1-8 trucks on Jobber, use Jobber AI Receptionist. 4-15 trucks on HCP, use HCP AI Agents. 10+ trucks on ServiceTitan with significant after-hours leakage and appetite for a 30-day pilot, run Avoca. 25+ trucks on ServiceTitan as the core platform, run ServiceTitan Voice (Titan Intelligence) with Avoca as benchmark comparison if after-hours leakage exceeds 8% post-deployment. The bake-off framework lives in L4 Ch2; the after-hours implementation lives here.

The Emergency Rule That Protects the On-Call Tech

The single most important configuration in the after-hours AI workflow is the emergency rule — the system-prompt logic that determines when the AI texts the on-call tech versus when it books the call for the next morning. Set the rule too loose and the on-call tech gets woken up for non-emergencies, burns out within 60 days, and the floor revolts. Set it too tight and true emergencies sit until 7:30 a.m. and the homeowner ends up on a competitor's truck before the morning CSR reads the booking. The rule has to be precise, documented, and trained.

What Counts as a True Emergency

The trade-by-trade emergency definition for residential service in 2026 standardizes around five categories. HVAC: no-heat below 50 degrees outside ambient OR with elderly/medical-vulnerable household members; no-cool above 88 degrees outside ambient OR with elderly/medical-vulnerable household members; gas smell; carbon monoxide alarm. Plumbing: active leak that cannot be shut off at fixture or main; sewage backup; water heater leak threatening property; frozen pipes with active flow loss. Electrical: power loss to medical equipment; burning smell or sparking; tripping main breaker that will not reset; downed line affecting service entrance. Drain & sewer: full backup with no usable fixtures; basement flooding; sewer line failure. Roofing: active leak during precipitation event; structural compromise with weather exposure.

What does not count as a true emergency: minor drip, temperature complaint within tolerance band, single-fixture outage with workaround available, scheduled-maintenance question, financing or pricing inquiry, anything that surfaces at 11 p.m. with a "I can wait until morning" qualifier from the homeowner. The 80% of after-hours calls fall into the not-emergency category; the system books them for first-out morning and the on-call tech sleeps through the night.

The System-Prompt Fragment That Implements the Rule

The system prompt's emergency-trigger logic asks the homeowner a fenced set of questions in sequence to classify the call. For HVAC no-heat: "Is it below 50 outside right now?" "Is there anyone elderly, very young, or with a medical condition in the home tonight?" "Is the heat completely off, or is it running but not warming?" For plumbing leak: "Can you shut the water off at the fixture or at the main?" "Is the leak actively damaging property right now — water on floors, ceiling, drywall?" The questions are not vibes-based — they map to specific yes/no answers that trigger or do not trigger the emergency escalation. The system prompt forbids the AI from inferring emergency status; only the explicit yes-answers fire the tech text.

When the rule fires, the AI texts the on-call tech with a structured page: customer name, address, symptom, urgency tier, callback number, and a brief context paragraph from the call. The tech reads in 12 seconds, decides whether to roll, and texts the homeowner back through the AI receptionist's threading. When the rule does not fire, the AI books a first-out slot for the next morning, texts the homeowner the confirmation, and queues the booking for the morning CSR's 7:30 a.m. review. The on-call tech's phone stays silent for the night.

The Three-Tier Routing — After-Hours, Warm Transfer, and Book-the-Morning

The named workflow for the after-hours AI receptionist has three tiers. Each tier corresponds to a clear set of triggers and a clear set of actions. The tiers are mutually exclusive — every after-hours call lands in exactly one.

Tier One: Book the Morning

The default after-hours tier. The AI runs the inbound call through the standard rebuttal library from Lesson 1 (price-shopper, just-looking, already-have-a-tech variants adapted to after-hours context), books a first-out morning slot, texts the homeowner the confirmation in 90 seconds, and queues the booking for the morning CSR's 7:30 a.m. review. Roughly 70-80% of after-hours calls land in Tier One. The on-call tech is not paged. The homeowner has a confirmed slot before the call ends and goes to bed knowing the truck is coming at 8 a.m. Booking rate on Tier One calls: 80-87%.

Tier Two: Warm Transfer for the Edge Cases

Roughly 10-15% of after-hours calls fall outside the AI's confident handling — service-area edge cases the AI's lookup table flags as uncertain, complex multi-system scenarios the AI cannot triage in confidence, pricing complexity the AI's system prompt routes to human judgment, or high-emotion calls (angry recall, dispute, escalation request) where AI mishandling would worsen the customer relationship. For these, the AI warm-transfers to the on-call CSR or the on-call manager (depending on the shop's escalation policy) with a structured handoff: the call audio plus a 4-field summary (customer, symptom, escalation trigger, AI's recommended path). The human resolves in 3-6 minutes and the AI re-engages to handle booking confirmation and on-my-way scheduling.

Tier Three: True Emergency — Page the Tech

Roughly 5-10% of after-hours calls meet the emergency rule from the prior section. The AI texts the on-call tech with the structured page, captures the homeowner's confirmation that the tech is rolling, and threads the AI receptionist into the resulting tech-to-homeowner communication for the rest of the night. The tech texts arrival ETA through the AI receptionist; the AI updates the homeowner. The morning CSR reads the entire overnight thread at 7:30 a.m. as one consolidated record. Tier Three is the smallest volume and the highest revenue per call — emergencies are typically the highest-ticket residential service work in the trades, and the customer relationship lifetime value is the largest because the shop showed up when the homeowner needed them most.

The math on the three tiers at a typical 7-truck residential shop: 14 after-hours calls per week, 70% Tier One (10 calls booked for morning at ~85% booking-%, 8-9 morning shows), 15% Tier Two (2 warm transfers, 90% resolved), 10% Tier Three (1-2 emergencies, tech rolls). Pre-AI baseline: 14 after-hours calls × 87% leak = 12 lost calls weekly. Post-AI: 14 calls × 80% capture = 11 captured. Net delta: roughly 10 recovered calls per week × $387 average ticket = $3,870/week, $200K+ annualized. The on-call tech gets paged 1-2 times per week on true emergencies; the rest of the volume sleeps through the night for the tech but books for the shop.

The Tech Rotation and the On-Call Comp Plan

The after-hours AI workflow has an implication most shops miss in the rollout: the on-call rotation gets easier to staff and more valuable to be on. Pre-AI, on-call meant 8-10 phone pages per night, half of them not real emergencies, sleep destroyed, weekend rotation hated by every tech on the bench. The standard comp arrangement was on-call stipend of $50-$150/night plus full ticket commission on any work done — and the comp was structurally underpaid because the burnout cost was not in the model.

Post-AI, on-call means 1-2 pages per week on actual emergencies, sleep mostly preserved, and the calls that do come are high-ticket, high-CLV emergency work. The rotation becomes a coveted shift among the bench's top performers because the math improves: same stipend, same commission, but 3-4x the per-page revenue and 80% less sleep disruption. Owners running the after-hours AI rollout consistently report that the tech-rotation conversation is the second-biggest cultural change after the CSR-row evolution from Lesson 1.

The comp redesign that supports the rollout: keep the on-call stipend in place ($50-$150/night signals the shop values the rotation), keep the full ticket commission on any emergency work, and add a "no false page" bonus tied to the AI receptionist's emergency-rule precision. If the on-call tech gets paged and rolls, then the ticket pays at full commission. If the tech gets paged and the symptom does not meet the emergency definition on arrival, the AI rule needs tuning and the false page generates a $25-$50 inconvenience credit to the tech. The bonus structurally aligns AI-rule tuning with tech sleep quality; floor leads tighten the system prompt week over week because the false-page bonus shows in the shop's labor cost.

The Metrics and the 90-Day Rollout for After-Hours AI

The after-hours workflow moves four headline numbers and three diagnostic numbers. The owner reads them on the dashboard before the truck rolls.

After-hours capture rate. The single most important number. Industry baseline 0-15%; target 80%+. The Avoca HL Bowman case landed at 80%+ within 60 days; the HCP AI Agents 2026 case studies match the band. Pre-rollout this number is roughly invisible because the answering-service-handled "messages" rarely convert to bookings; the post-rollout dashboard makes it visible daily.

Missed-call rate. Industry baseline 22%; target under 5%. Includes after-hours, in-hours overflow, and hold-abandonment. After-hours AI handles the largest share; in-hours overflow handled by the same AI on overflow routing.

Cost per booked call. Marketing spend divided by booked jobs. Pre-AI the cost per booked call inflates by exactly the leak; post-AI it drops mechanically. The HL Bowman $350 → $215 drop (39% reduction) lives largely in the after-hours capture.

On-call tech sleep quality (false-page rate). Number of pages per night the on-call tech gets where the symptom did not meet the emergency rule on arrival. Healthy target under 1 per week; if the number runs higher, the system prompt's emergency-trigger logic needs tightening. The metric is unique to the after-hours workflow and the only one the on-call tech cares about personally.

Three diagnostic numbers: tier distribution (Tier One should land at 70-80%, Tier Two at 10-15%, Tier Three at 5-10%; significant drift means system prompt is mis-classifying); morning CSR overnight-review time (target 8-15 minutes at 7:30 a.m. — if it runs longer, the call summaries are not in the six-field format from Lesson 2); show rate on Tier-One bookings (target 90%+ — Tier One morning slots show higher than daytime average because the booking is fresh and the homeowner urgently needed the slot).

The 90-day rollout sequence: Week 0, configure system prompts for the three tiers, set the emergency-rule trigger questions, define the warm-transfer escalation policy, and run the on-call rotation conversation with the tech bench. Week 1, deploy to after-hours window (typically 6 p.m. - 7 a.m. weekdays, all-day weekends). Run shadow mode for the first 3 nights — AI handles calls but routes everything to the on-call tech to verify the rule. Week 2, switch to live mode; on-call tech only paged on emergency-rule triggers; morning CSR runs the 7:30 a.m. overnight review and flags any mis-classification. Weeks 3-8, weekly tuning of the emergency rule based on the false-page audit and the missed-emergency audit (any Tier-One booking that should have been Tier Three). After-hours capture climbs from 0-15% to 50-65% by week 4 and 75-85% by week 8. Weeks 9-12, edge cases land (snowstorm overload, holiday volume spikes, multi-property landlord calls); polish phase. By day 90 the workflow stabilizes at 80%+ after-hours capture, missed-call under 5%, on-call false page under 1/week.

Key Takeaways

• The answering service is a 22% leak that costs more than the invoice. At a 7-truck residential shop, 87% after-hours leak × 14 calls/week × $387 ticket = $208K-$312K annualized visible leak; $400K-$600K when CLV and referral effects are included.
• Three after-hours AI products dominate in 2026: Avoca (bolt-on standard, $1K-$3K/mo, deepest training), Jobber AI Receptionist ($99/mo on Plus plan, for 1-8 truck Jobber shops), Housecall Pro AI Agents (in-HCP, $1M-$15M segment). ServiceTitan Voice (Titan Intelligence) is the 25+ truck ServiceTitan-native option.
• The three-tier routing is the workflow: Tier One book-the-morning (70-80% of calls, AI handles end-to-end), Tier Two warm-transfer for edge cases (10-15%), Tier Three true-emergency tech page (5-10%). Mutually exclusive; every call lands in exactly one.
• The emergency rule has trade-by-trade definitions: HVAC (no-heat under 50 ambient, no-cool over 88, gas, CO), plumbing (active uncontrolled leak, sewage, water heater leak, frozen pipes), electrical (medical-equipment power, sparking, burning smell, main breaker), drain (full backup, basement flooding), roofing (active leak in weather, structural). The system-prompt logic asks fenced yes/no questions; AI cannot infer emergency status.
• The on-call tech gets paged 1-2 times per week instead of 8-10 per night. Sleep mostly preserved; calls are high-ticket emergency work. The rotation becomes coveted instead of hated. False-page bonus structurally aligns AI-rule tuning with tech sleep.
• Four headline metrics: after-hours capture (0-15% → 80%+), missed-call (22% → under 5%), cost per booked call ($350 → $215 / 39% drop, HL Bowman benchmark), on-call false-page rate (under 1/week target).
• Three diagnostic metrics: tier distribution (70-80/10-15/5-10), morning CSR overnight-review time (8-15 minutes at 7:30 a.m.), Tier-One show rate (90%+ because the slots are fresh and urgent).
• The 90-day rollout: shadow mode week 1, live mode week 2, weekly emergency-rule tuning weeks 3-8, polish weeks 9-12. After-hours capture lands at 75-85% by week 8; missed-call rate lands under 5% by day 90.
• Payback is 2-4 weeks at typical leak math. $200K+ annualized capture at a 7-truck shop versus $12K-$36K annualized AI receptionist cost. The answering service invoice ($800-$1,500/mo) goes to zero alongside the leak closure.