There are four serious voice agents in the trades in 2026 — Avoca, Jobber AI Receptionist, Housecall Pro AI Agents, and ServiceTitan Voice — and they are not interchangeable. They are competing for the same surface (the inbound call before a CSR picks up, the after-hours overflow, the in-hours abandon, the warm-transfer to a human when the call is too complex), but they sit on different platforms, integrate at different depths, charge at different bands, and produce different booking-percentage outcomes on the same call volume. The owner choosing among them in 2026 is not picking a chatbot — they are picking which voice answers the phone the next time a homeowner calls at 6:47 a.m. with a no-heat. The wrong pick costs $80K-$150K of margin annually because of leaked bookings; the right pick produces the HL Bowman case math (100% answer rate, 70% YoY revenue growth, cost-per-conversion down 39%). This lesson is the rubric the owner runs in the 30-day pilot — the questions to ask each vendor, the numbers to demand on the demo call, what to measure during the pilot, and the cost-per-booked-call math that decides the commit.

The Four Contenders and Where Each Actually Fits

Avoca is the category leader by capital, by case-study depth, and by feature surface in 2026. April 2026 Series B of $125M led by Meritech and General Catalyst at roughly $1B valuation, on top of the Kleiner Perkins Series A. The HL Bowman case is the published reference: 100% answer rate, 70% YoY revenue growth, cost per conversion $350 → $215 (39% reduction). Avoca's depth advantage sits in warm-transfer logic (when to keep the AI on the call vs. escalate to a human CSR), intent classification (emergency vs. routine, service vs. sales, commercial vs. residential), after-hours booking quality, and post-call summary fields. Pricing in 2026: $1K-$3K/month for a 6-12 truck shop depending on call volume. Best fit: shops with measured missed-call rate above 15% on any FSM platform; multi-FSM portfolios; shops where the depth premium clears the integration cost.

Jobber AI Receptionist is built inside Jobber and integrates natively with Jobber's CRM, scheduling, and quoting workflows. Built for the smaller-shop, multi-trade audience Jobber serves — lawn, landscape, cleaning, pest, handyman, smaller HVAC and plumbing shops. The voice agent answers, classifies, books direct into Jobber's calendar, and warm-transfers to the shop's phone when needed. Pricing in 2026: bundled into Jobber's higher-tier plans, effectively $50-$200/month incremental on top of Jobber's base seat cost. Best fit: shops already on Jobber, especially Tier 1 and Tier 2 (under 15 trucks), where the integration advantage and bundled pricing dominate the depth premium calculation.

Housecall Pro AI Agents is Housecall Pro's voice and chat agent stack. Like Jobber, it lives inside the HCP platform and integrates natively with HCP's calendar, customer record, dispatch, and invoicing. HCP's 2026 push on the AI Agents surface has narrowed the gap to Avoca on basic answer-and-book workflows; it remains behind on the deeper after-hours edge cases and multilingual handling. Pricing: bundled into HCP's higher-tier plans, $50-$300/month incremental. Best fit: HCP shops; especially 4-20 truck shops with moderate measured missed-call rates (under 15%) where the bundled pricing and zero-integration cost dominate the depth question.

ServiceTitan Voice is ServiceTitan's voice agent, the in-platform challenger to Avoca on the ServiceTitan installed base. Integrates natively with Titan Intelligence, the dispatch board, the customer record, and the call summary stack. ServiceTitan's 2026 release schedule has accelerated the feature catch-up to Avoca but the depth gap on warm-transfer logic and after-hours booking quality remains visible in side-by-side pilots. Pricing: $300-$600/month per shop add-on. Best fit: ServiceTitan shops with measured missed-call rates under 12%, especially shops that prioritize stack consolidation over the depth premium. Above 15% missed-call rates, Avoca still wins on the ServiceTitan installed base by depth.

The Six Criteria That Decide the Bake-Off

Six criteria decide the bake-off in 2026. Score each from 1 to 5 in the 30-day pilot; the highest aggregate wins; published vendor numbers do not substitute for measured pilot data.

Criterion 1: Booking % on AI-handled calls. The headline metric. What percentage of inbound calls answered by the AI agent end in a booked job (either booked direct or warm-transferred and booked by a human CSR)? Industry baseline for unanswered + abandoned calls: 0% bookings. Industry baseline for human-CSR answered calls: 65-85%. AI agent target: 70-85% on routine calls (the AI books direct); 90-95% pass-through on complex calls (the AI warm-transfers to a human who closes). Avoca's HL Bowman case reports 100% answer rate; booking % on the answered band is in the 75-85% region. ServiceTitan Voice, Jobber AI Receptionist, and HCP AI Agents land 5-15% lower on booking % in 2026 published cases. Measure during pilot, do not trust the demo.

Criterion 2: Missed-call recovery rate. Of the calls the shop's previous setup missed (after-hours, in-hours abandon, voicemail), what percentage does the AI agent now answer and book or warm-transfer? Pre-AI: 0-15% (the rare after-hours answering service or callback workflow). Post-AI: target 80%+. Avoca routinely lands 85-95% on shops with previously-leaking after-hours. ServiceTitan Voice and HCP AI Agents land 70-85% in 2026 cases. Jobber AI Receptionist's recovery rate depends on Jobber's call-routing setup; well-configured shops land 80-90%.

Criterion 3: Transfer-to-CSR accuracy. When the AI decides the call exceeds its scope and warm-transfers to a human CSR, is the human CSR getting the right call? Accuracy here means: AI correctly identifies routine vs. complex; transfers complex; books routine. Misclassification cost: the CSR spends 4-8 minutes on a call the AI should have closed (CSR floor capacity wasted), or the AI books a call it should have transferred and the customer arrives Monday irritated about a misunderstanding. Target: 90%+ transfer-decision accuracy. Avoca's depth advantage sits heavily here; the warm-transfer logic has been refined across 24-36 months of focused engineering. ServiceTitan Voice and HCP AI Agents are closing the gap; Jobber AI Receptionist's accuracy depends on call-flow configuration.

Criterion 4: Integration depth. Does the booked call land in the right slot in the FSM platform with the right customer record, equipment tag, and dispatch instructions? Or does it require a CSR to re-enter half the data after the fact? Integration depth measures the data flow quality from voice agent to FSM platform. In-platform agents (Jobber AI Receptionist, HCP AI Agents, ServiceTitan Voice) win this by definition — the voice agent is the FSM platform. Avoca, the bolt-on, wins or loses by the maturity of its FSM connector; Avoca's ServiceTitan and HCP connectors are mature in 2026 (sub-30-second sync, no data loss on platform updates in published shop reports); Jobber and Sera connectors are less mature.

Criterion 5: Cost per booked call. The denominator the owner pays attention to on Friday. Compute: monthly tool cost ÷ AI-booked + AI-transferred-then-booked calls = cost per booked call. Industry pre-AI cost per booked call via human-only CSR + answering service: $4-$12 depending on shop volume and answering-service rates. Post-AI: target $2-$6 on Avoca, $1-$4 on the in-platform agents (because the in-platform pricing is bundled lower). The math: a 7-truck shop with 1,800-2,400 booked calls/month at $1K-$3K Avoca cost = $0.50-$1.50 per booked call attributable to Avoca; same shop with ServiceTitan Voice at $400/month = $0.18-$0.25 per booked call. Cost per booked call alone does not pick the winner because booking % drives revenue; high booking % at $1.50 cost beats lower booking % at $0.20 cost on margin contribution.

Criterion 6: After-hours booking quality. The hidden differentiator. What does the call sound like at 11:30 p.m. on a Sunday when the homeowner has a flooded basement? Does the AI handle the emergency triage correctly (collect address, equipment, severity, dispatch to the on-call tech with a clear ticket; or warm-transfer to the after-hours phone if severity exceeds AI scope)? Does it book a routine non-emergency for the next morning without committing the on-call tech? Does it accidentally book emergencies as routine slots (worst-case failure)? Avoca's edge-case handling is documented across more shop reports than any competitor; ServiceTitan Voice's 2026 emergency-triage release closed much of the gap; HCP and Jobber lag on the deepest edge cases.

The Questions to Ask the Vendor on the Demo Call

Vendor demo calls follow a pattern: the vendor demos the happy-path booking flow, shows the call summary, references the published case study, and quotes a number. The owner's job is to break the demo with specific questions that exercise the depth dimensions. Six questions surface the differences fast.

One. "Show me the after-hours emergency call where the homeowner says 'my basement is flooding' at 11:30 p.m. on a Sunday. How does the AI distinguish emergency from routine, and what's the warm-transfer logic?" If the vendor can play a real anonymized recording of this flow, the depth is real. If the vendor flips to a slide showing the call-flow diagram, the depth is theoretical.

Two. "What's your transfer-to-CSR accuracy in measured pilots, not in the case study? I want the median and the bottom-quartile pilot." The case study is the top pilot. The median tells you what to expect; the bottom quartile tells you the failure mode. Vendors who only quote case-study numbers are filtering hard; ask for the pilot distribution.

Three. "Walk me through your ServiceTitan / Jobber / HCP / Sera integration — how does the data flow, what's the sync latency, and what breaks when the platform releases a quarterly update?" The honest answer covers sync latency (target sub-30 seconds), data fields synced (target all booking-relevant fields), and breakage cadence on platform updates (target zero broken updates in last 12 months on the major platform). Vendors who hand-wave on this question are warning you about integration debt.

Four. "What's your cost per booked call in a pilot at a 7-truck shop with 22% measured missed-call rate? Show me the math." Forces the vendor to compute the metric the owner uses, not the metric the vendor prefers. Avoca's response should land at $0.50-$1.50; in-platform competitors should land at $0.18-$0.60. If the vendor can't or won't compute it on the call, walk.

Five. "How do you handle multilingual calls? Spanish-only, Spanish + English code-switching, Vietnamese, Mandarin, Portuguese, Tagalog?" In most metros the trade customer base includes 8-20% non-English-primary callers; the AI agent's handling of these calls determines whether the shop captures or leaks this segment. Avoca leads on multilingual depth in 2026; in-platform agents are catching up but still trail.

Six. "What does your 30-day pilot look like? Can I get month-to-month for the first 90 days against an annual commit at month 4?" Test the rubric from Lesson 1. Vendors confident in pilot conversion accept the term; vendors refusing signal walk.

What to Test in the 30-Day Pilot

The 30-day pilot for a voice agent has six measurement workstreams. Set them up before day 1; measure daily; review with the team weekly; commit-or-walk decision at day 30.

Workstream 1: Booking-percentage baseline. Two weeks pre-pilot, measure the current booking % on inbound calls (% of inbound that ends in a booked job). During pilot, measure the same metric on AI-handled calls separately from human-CSR-handled calls. Compare. Target lift: 10-25 percentage points on AI-handled calls vs. previous unanswered/abandoned bucket.

Workstream 2: Daily integration health log. Each morning, the ops manager or owner checks: did all overnight bookings flow into the FSM platform correctly? Are there any data fields missing or mis-mapped? Were there any failed transfers? 30 days of daily logs reveals integration patterns vendor demos hide.

Workstream 3: CSR adoption rate. The CSR floor has to use the AI agent's warm-transferred calls — handle the transfer, see the AI's pre-call summary, take the call to close. Measure % of CSRs using the workflow daily; identify CSRs avoiding it; coach the holdouts. A pilot that produces 80%+ CSR adoption produces real metrics; a pilot at 40% CSR adoption produces noise.

Workstream 4: Call-quality audit. Sample 10 AI-handled calls per week (5 booked direct, 3 warm-transferred, 2 hung up by the customer). Listen end-to-end. Score against the rubric (intent classification accuracy, booking-data accuracy, customer tone handling, AI confidence calibration). Identify failure patterns; report to vendor support; track resolution time.

Workstream 5: Cost-per-booked-call computation. At day 30, compute monthly tool cost ÷ AI-attributable booked calls (AI-direct + AI-transferred-and-booked). Compare against pre-pilot baseline (typically $4-$12 with answering service + human CSR overflow). Target post-pilot: $1-$6 depending on tool.

Workstream 6: Owner / ops hours during pilot. Track hours spent on integration troubleshooting, vendor support tickets, CSR training, and call audits. Target under 8 hours/week at week 1, dropping to 1-2 hours/week by week 4. Vendors whose tool requires more than 8 hours/week of ongoing owner time after week 4 are not production-ready for the shop.

Cost per Booked Call — The Math That Decides

Cost per booked call is not the decision metric; cost per booked call combined with booking % is. The math: each booked call produces a revenue range based on shop close rate and average ticket. At a 7-truck residential HVAC shop with $387 average diag-plus-repair value, 35% close-to-full-job rate on $5K replacement, and $14K average replacement ticket, each booked call has a blended revenue value of $700-$1,400. Margin: 38% gross on the blended mix = $266-$532 margin per booked call.

The cost-per-booked-call comparison: Avoca at $1.50 produces $264.50-$530.50 net margin per booked call after tool cost. ServiceTitan Voice at $0.25 produces $265.75-$531.75 net margin per booked call after tool cost. The per-booked-call delta is $1.25 — negligible. The decision is booking %, not cost per call. If Avoca produces 80% booking on AI-answered calls and ServiceTitan Voice produces 70%, Avoca produces 14% more booked calls on the same volume. At 1,800 monthly inbound calls × 70% answered by AI (rest by human CSR) = 1,260 AI-touched calls × 10% booking delta = 126 additional booked calls per month × $300 average margin = $37,800/month margin difference. The $1,500/month price gap between Avoca and ServiceTitan Voice is irrelevant against $37,800 in incremental margin.

The owner who picks on cost per booked call alone underweights booking %. The owner who picks on booking % alone ignores integration overhead. The owner who runs the full 30-day pilot with all six criteria scored produces the defensible decision. The pilot rubric replaces the vendor pitch deck as the decision artifact.

The Decision Rules by Shop Size and FSM Platform

The default recommendations by shop profile, distilled from the 2026 pilot data and the criteria above.

1-5 trucks on Jobber: Jobber AI Receptionist. Bundled pricing, native integration, sufficient depth on Tier 1 call volumes. Don't add Avoca unless missed-call rate is measured above 18% and Jobber AI Receptionist has produced under 75% booking % over a 60-day baseline.

1-5 trucks on Housecall Pro: HCP AI Agents. Same logic. The bundled pricing and zero-integration cost dominate the depth question at this size.

1-5 trucks on ServiceTitan: ServiceTitan Voice if measured missed-call rate under 12%. Avoca if above 15%, even at the depth-premium price.

6-15 trucks, any FSM: Hybrid. Run the platform-native agent as default. Bolt-on Avoca specifically for the after-hours and overflow surface where depth premium clears integration cost. This stack covers in-hours with low cost and after-hours with Avoca depth.

16-40 trucks, any FSM: Avoca as primary. The depth premium dominates at this volume; platform-native agents lose 5-10 points of booking % vs. Avoca, which translates to $200K-$500K of annual margin at this size. The integration cost is amortizable across the dedicated ops/IT capacity.

40+ trucks or multi-shop: Platform-mandated. Most PE platforms (Wrench, Authority, Apex, Sila, Path Light, Redwood) standardized on Avoca as the cross-portfolio voice agent in 2025-2026 because of the depth-premium math at portfolio scale. Franchisees of franchised systems may face HQ-mandated alternatives; the override-memo template covered in Level 5 documents the case.

Key Takeaways

• Four voice agents compete in 2026: Avoca (category leader, $125M Series B at ~$1B), Jobber AI Receptionist (bundled in Jobber), Housecall Pro AI Agents (bundled in HCP), ServiceTitan Voice (in-platform challenger). Not interchangeable.
• Six bake-off criteria scored 1-5 each in 30-day pilot: booking % on AI-handled calls, missed-call recovery rate, transfer-to-CSR accuracy, integration depth, cost per booked call, after-hours booking quality.
• Booking % target: 70-85% on routine AI-direct bookings; 90-95% pass-through on warm-transferred complex calls. Avoca leads at 75-85%; in-platform competitors land 5-15% lower in 2026 published cases.
• Missed-call recovery target: 80%+ on previously-leaking after-hours and overflow. Avoca at 85-95%; ServiceTitan Voice and HCP AI Agents at 70-85%; Jobber AI Receptionist at 80-90% with well-configured routing.
• Cost per booked call: Avoca $0.50-$1.50; in-platform agents $0.18-$0.60. Per-call delta is negligible — decide on booking %, not cost. 10% booking-% delta on 1,800 monthly calls = $37,800/month margin difference at typical shop math.
• Six vendor questions on the demo call: after-hours emergency flow with audio; pilot distribution numbers not case-study numbers; integration sync latency and breakage cadence; cost per booked call math live; multilingual handling depth; 30-day pilot terms.
• Six pilot workstreams measured daily for 30 days: booking-% baseline, daily integration health log, CSR adoption rate, weekly call-quality audit, cost-per-booked-call computation, owner/ops hours.
• Default recommendations by shop profile: 1-5 trucks on Jobber → Jobber AI Receptionist; on HCP → HCP AI Agents; on ServiceTitan → ServiceTitan Voice if missed-call <12% else Avoca. 6-15 trucks → hybrid platform-native + Avoca on overflow. 16-40 trucks → Avoca primary. 40+ trucks or multi-shop → platform-mandated, typically Avoca at PE-backed portfolios.
• The depth premium math: at 16+ trucks, Avoca's 5-10 point booking-% lead vs. platform-native translates to $200K-$500K annual margin. Premium pricing is irrelevant against this delta.
• The pilot rubric is the decision artifact, not the vendor pitch deck. Vendors who refuse month-to-month-for-90-days against month-4 annual commit are signaling pilot conversion risk. Walk.