The Challenge
A residential HVAC business with no commercial pipeline
AirCentral had spent years building a reliable residential HVAC business - Google Ads, trucks, technicians, great reviews. The economics worked, but ceiling was real: residential jobs average $4,000 and you fight the same local competitors on price every week.
Commercial HVAC contracts are an entirely different game. Deals run $30K - $150K. Contracts renew annually. Reputation travels further. But the buyers live inside facility management teams at office buildings, warehouses, retail chains, and multi-unit residential - places no homeowner funnel will ever reach.
AirCentral had tried the normal playbooks: a LinkedIn outreach VA (zero pipeline, 2 months), a "cold email agency" (got 10,000 emails sent, 0.3% reply rate, one meeting - it didn't show up), and sponsoring two trade associations (cost $14K, produced one inbound lead). None of it worked because commercial HVAC buyers don't respond to generic pitches about "25 years of experience and satisfied customers." They respond to signals about their specific building, their specific problem, and their specific moment of need.
"We spent two years trying to crack commercial and made zero dollars. Deep-Y installed the system and we signed $540K in 90 days - with no new headcount. This wasn't a tool; it was a system someone built, deployed, and ran for us."
I
Itay M.
Operations Lead, AirCentral
The Approach
How Deep-Y built the commercial outreach system
Deep-Y is a done-for-you agency. We built every layer - signal sources, ICP, infrastructure, personalization, sequencing, qualification, and handoff - and kept it running. AirCentral's team did not write a single email.
01
Commercial ICP from scratch
AirCentral had no commercial ICP. We built one: building size (50K - 500K sqft), HVAC system age (7+ years - peak replacement window), facility manager tenure, recent building ownership changes, and energy cost pressure. 2,800 target accounts across the metro area.
02
Signal sources nobody watches
We wired signals from building permit filings (HVAC replacement permits), tenant departure announcements (vacancy creates maintenance pressure), property management company changes, energy audit news, and local temperature event forecasts. These aren't in Apollo or ZoomInfo. We built the pipeline.
03
Deliverability infrastructure (built to last)
We set up 6 sending domains, warmed them for 21 days, and rotated inboxes with anti-spam logic. AirCentral's legacy domain was already in poor sender reputation - we never touched it. Result: 89% open rate across 14,000 sends with zero deliverability complaints.
04
AI-written, signal-grounded personalization
Every email referenced something specific: "Saw the permit filing for 420 W Main - if you're replacing the chiller, here's what we do for buildings in the 80K sqft range…" AI wrote each one using the signal data as the hook. No templates. No merge tags. Each message was structurally custom.
05
Meeting-booking auto-loop
Replies hit an AI triage layer that routed interested prospects directly into AirCentral's calendar with zero human touch. The Operations Lead just showed up to qualified calls. 340 booked in 90 days, 78% show-up rate.
06
Weekly review + ongoing optimization
Deep-Y reviews every campaign weekly, rewrites underperforming angles, adds new signals as we find them, and tunes the ICP as contract data comes back. The system compounds - month 2 was 40% better than month 1.
Under the Hood
How the system was built - step by step
The AirCentral AI outreach system had six components, each solving a specific failure mode in their previous manual outreach.
01
Domain architecture (week 1)
Built 4 sending domains with proper SPF, DKIM, and DMARC authentication. AirCentral's primary domain was excluded from cold sending to protect deliverability of existing client communications. Each sending domain was aged 30+ days before first use.
02
ICP signal model (week 1)
Mapped the commercial HVAC buyer signals: facility manager job titles, building square footage data, recent construction permits, and LinkedIn job-change events indicating new facility managers. Built a scoring model that ranked 3,400 target accounts by buying probability.
03
AI personalization engine (week 2)
Trained the AI on 40 real conversations from AirCentral's past deals to understand the language their buyers responded to. Each email referenced a specific signal about the prospect's building or recent activity - not generic personalization tokens.
04
Warm-up and ramp protocol (weeks 2-3)
Started at 8 emails/day per domain, increasing by 5 per day over 21 days to reach full volume. Monitored sender reputation scores daily. Achieved 89% open rate before any sequence optimization was needed - a deliverability baseline.
05
Reply handling and routing (week 3)
Built an AI reply classifier that sorted responses into: interested (routed to AirCentral sales), not now (scheduled follow-up at 60 days), referral (extracted new contact), and objection (fed into sequence optimization). 340 interested replies were routed without a human touching the inbox.
06
Optimization loop (ongoing)
Every Friday: AI analyzed the week's reply patterns, updated subject line variants, and reweighted the signal model based on which prospect types converted. By month 3, conversion rate from email to meeting had improved 40% over the baseline.
The Result
The 90-day scoreboard
- Signed commercial contracts12 contracts, $540K total pipeline
- Average deal size$45K (vs $4K residential)
- Email open rate89% (vs 20% industry average)
- Reply rate24% across 14,000 sends
- Commercial meetings booked340 in 90 days
- Meeting show-up rate78%
- Cost per qualified lead$0.80
- First contract closedDay 18 from launch
- ROI on Deep-Y engagement4.2x in the first 90 days
- New headcount requiredZero
AirCentral is now a residential-and-commercial HVAC business, and the commercial side is growing faster than residential ever did. The Deep-Y system continues to run - every week, more contracts, same zero manual effort.
Client
HVAC · Commercial Field Services · Outreach
