How an AI Employee actually works.

Overview

An AI Employee is not a chatbot. It's an operational layer that sits between every inbound channel a trade business has — website, SMS, missed-call text-back, Google Business Profile, paid landing page — and that business's dispatch system.

Its job is the job a CSR (customer service representative) would do, with three differences that matter: it answers in under 45 seconds every time, it works 24/7, and it speaks the language of one specific trade.

The pipeline has four stages, and each one is where most generic chatbots fail. We'll walk through each in turn.

1. Capture — meeting the customer where they are

The first stage is the unglamorous one: making sure the AI Employee is on every channel a customer might use, and that every channel routes into the same pipeline.

The most expensive failure mode in the trades is the inbound that doesn't get caught. Industry data is consistent across verticals: between 40% and 60% of inbound calls go unanswered — most of them after-hours or during overflow events (storms for roofing, heatwaves for HVAC, season-opening rushes for pool service). Every unanswered call is a lead the next contractor on Google captures.

The capture layer normalises the inbound channels:

• Web chat widget — one-line embed on the niche site, with a customised greeting and trade-specific placeholder copy.
• SMS / missed-call text-back — a Twilio number that auto-replies to any missed call within 60 seconds. This is the highest-ROI channel for most shops because it converts the existing voicemail funnel into a captured lead.
• Google Business Profile messaging — for the segment of customers who message directly from the GBP listing.
• Facebook / Instagram lead forms — when paid social is part of the lead mix.
• Email-to-conversation — for the customer who lands on the website but only feels like emailing.

All of these route into the same conversation engine. From the AI Employee's perspective, channel doesn't matter; what matters is that the conversation has started.

2. Qualify — the part where verticality matters

This is the stage that separates a real AI Employee from a generic chatbot.

A roofing inbound and an HVAC inbound look superficially similar — both are emergencies, both involve a homeowner who's stressed, both want someone on-site fast. But the qualification logic is completely different.

The qualifier doesn't just collect data — it routes. A solar lead with a 25-year-old roof gets a different conversation flow than one with a new roof. A no-cool HVAC call from a customer with an infant in the house gets emergency priority. A plumbing leak that's "still actively dripping" gets dispatched today; one that's "stopped after we shut the valve" gets booked tomorrow at a more efficient slot.

This is also where the AI Employee handles the part of the job most generic tools don't touch: objection handling and reassurance. The customer who asks "do you do warranty work for [specific brand]" gets the right answer. The customer who hesitates because they think they need three quotes gets the gentle, accurate response your best CSR would give. The customer who's panicking about water on the kitchen floor gets calmed down enough to give you the address.

A generic chatbot collects information. A trade-trained AI Employee qualifies, reassures, and routes — the way a senior CSR with ten years on the desk would.

3. Dispatch — writing into the calendar, not into a queue

This is where most "AI for the trades" products quietly fall apart. They collect the lead and drop it into a queue for a human to act on. By the time the human picks it up — sometimes hours later — the customer has booked with the next contractor.

An AI Employee dispatches directly. It checks live calendar availability against tech skill matrix, parts availability, and service-area routing, then books the slot, sends the customer a confirmation SMS, and writes a structured record into the CRM — usually GoHighLevel or whatever the shop uses.

The dispatch layer answers four questions before booking:

• Which tech? Skill matrix (commercial vs residential, brand certifications, lead vs apprentice) matched against the qualified job type.
• Which slot? Live calendar availability, with severity-aware prioritisation. An emergency leak displaces a maintenance visit; a panel upgrade gets the lead electrician's morning slot.
• What parts? Parts pre-pull list pushed to the truck-loading workflow when the job type is known in advance (water heater swap, panel upgrade, AC compressor replacement).
• What expectations? The customer gets a precise confirmation: tech name, arrival window, what they'll be doing, what it'll cost (or how the estimate will be built on-site).

4. Hand-off — the part nobody asks about, but everyone notices

The final stage is what your team experiences. When dispatch picks up the job in the morning — or the on-call tech opens it at 2am — they don't see "AI booked: leak, 8:30 AM." They see the full transcript, the qualification answers, the recommended action, and any flags the AI Employee raised ("customer mentioned the home is for sale — likely wants quick patch, not full replacement").

This matters because the worst experience for a customer is being asked the same questions twice. They told the AI Employee about the leak; if the tech shows up and says "tell me what's going on," the magic breaks.

The hand-off output is structured, not conversational. Something like:

job_id:           ROOF-2026-04-29-0042
intent:           emergency_leak
severity:         high (active drip, slowing)
property:         412 Maple Ave, single-family, 2-story
storm_event:      hail/wind, night of 04-28
insurance:        State Farm, claim not yet filed
qualified_by:     ai_employee_v2.4
booked_for:       2026-04-29 08:30
assigned_tech:    jared_m (lead, residential)
truck_load:       tarp_roll_20x20, blue_tarp_x4, ladder_ext, ...
flags:            home_for_sale_implied · file_claim_today · wife_decision_maker

That's what dispatch sees. That's what the tech sees on their tablet when they open the job. No friction, no re-asking, no lost context.

The stack

For the technically-curious: each AI Employee is built on the same four-layer stack, with the verticalisation living in the middle layers.

Why vertical-specific (and not "one chatbot fits all")

The fastest way to build a generic AI assistant is to make it talk to anyone about anything. The fastest way to ruin a trade business's customer experience is to put that generic assistant on its phone line.

The reason we operate as a network of vertical-specific products — rather than a single configurable platform — comes down to four things:

• Vocabulary. A homeowner saying "the AC is short-cycling" means something specific. A generic AI guesses. A trade-trained AI knows it's likely a refrigerant or thermostat issue and qualifies accordingly.
• Trust signals. Customers in distress test the assistant with trade-specific questions to see if it knows what it's talking about. "Do you do TPO or only EPDM?" is a question a roofing customer will ask. A vertical-specific AI Employee answers correctly and keeps the conversation alive.
• Dispatch logic. The decision tree for booking a no-cool call vs a panel upgrade vs a green-pool emergency is fundamentally different. Configuring this into a single "platform" means most of it never gets configured.
• SEO and lead-source authority. A homeowner Googling "roofer near me at 2am" lands on roofersaiemployee.com, not a generic AI tools directory. The verticalised architecture is also how we win organic traffic.

Integrations

The AI Employee writes into your existing stack, not its own. Out of the box:

• GoHighLevel — primary CRM and calendar integration. Two-way sync.
• Twilio — SMS layer for missed-call text-back and confirmation flows.
• Resend — transactional email for confirmations, summaries, and weekly reports.
• Google Business Profile — direct messaging integration for GBP-sourced inbounds.
• Meta lead forms — webhook-driven for paid Facebook / Instagram leads.
• Cloudflare Workers — the runtime layer; sub-50ms response globally.

Custom integrations (ServiceTitan, Housecall Pro, Jobber, FieldEdge, etc.) are deployed per-customer during onboarding.

Deployment timeline

From signed contract to AI Employee live on the customer's site:

• Day 1–2: Discovery call. We capture trade specifics, dispatch quirks, common objections, brand voice, service area, tech roster, calendar logic.
• Day 3–5: AI Employee tuned and trained against the customer's actual historical inbound data (anonymised). Reviewed by the customer.
• Day 6–7: Soft-launch on one channel (usually missed-call text-back) with human shadow review.
• Day 8–14: Full deployment across all channels. Daily monitoring, weekly tuning.
• Day 15+: Handed to the customer's normal operations cadence. Ongoing model improvements rolled in monthly.

Most customers see measurable lift in week two. The customers who don't, almost always have an upstream problem (lead source quality, dispatch capacity, tech shortage) that no AI Employee can solve.

Want to see this on a real call from your shop? Send us 60 seconds of historical inbound audio (or a transcript), and we'll show you the AI Employee handling it in your demo. Book the demo →