A workflow map names the 38 steps. The decision rules tell the manager which steps are AI, which are human, and which are mixed. But neither tells the team what crosses the seam between AI and human at each handoff — what fields move, in what shape, at what timing, with what escalation triggers if something is missing or wrong. That handoff design is what separates a workflow that runs from a workflow that breaks at the seams. AI hands the tech a pre-route brief at Step 11; the tech hands the manager a structured note at Step 30; the manager hands the owner the Friday recap. Each handoff has a contract — the fields that must be present, the timing that protects the next step, and the escalation that fires if the contract is broken. This lesson is the handoff design discipline. The manager who finishes this lesson can build, audit, and defend the seams between AI and human at every step of the call, with named fields, named timing, and named escalation triggers — the operating infrastructure that makes a 6-truck shop's AI workflow look identical in quality to a 60-truck shop's at the per-call level.

Why the Handoff Is Where Workflows Break

Shops that fail at AI in 2026 do not fail at the AI step itself. The AI generates output competently — Avoca answers, ServiceTitan summarizes, Dispatch Pro re-routes, Rilla scores. The shops fail at the seam where AI output meets human action and at the seam where human output meets the next AI step. The seam is where context evaporates, where fields go missing, where timing slips, where the receiving party catches no drift because nothing prompted them to look. The map says AI happens at Step 7 and the human verifies at Step 11 — but if no one designed what data the tech actually receives in their pre-route brief, the tech opens the ticket and sees the same generic five lines they saw before AI was in the workflow.

Handoff design is contracts. Each handoff has a sending party (AI or human), a receiving party (AI or human), a payload (the fields), a delivery timing (when it lands, where it lands), and an escalation trigger (what fires if the payload is missing or wrong). The contract is what makes the handoff reliable. The contract is also what makes the handoff auditable when the workflow drifts. The service manager who reviews the Step 30 handoff and finds the tech's structured note missing the "Customer Mood" field traces it to the Step 17 prompt — the constraint did not require Customer Mood, so AI omitted it under load. Fix the constraint, audit the next 50 handoffs, the field reappears. Without the handoff contract, the missing field is invisible until a Comfort Advisor walks into a Tuesday-evening kitchen table cold because the prior tech's notes did not capture that the homeowner was already hesitant.

The handoff design discipline operates at three altitudes. Vertical handoffs move down the org chart — owner to manager, manager to tech, tech to customer. Horizontal handoffs move across roles at the same level — CSR to dispatcher, tech to advisor, advisor to install coordinator. Diagonal handoffs cross between AI and human — Avoca to CSR, ServiceTitan AI summary to tech, Rilla scorecard to service manager. All three need contracts. The manager designs all three. The team operates within them.

The Handoff Contract: Five Fields

Every handoff in the workflow has the same five-field contract. Memorize the structure; the content inside the fields varies by step. Sending party. Receiving party. Payload. Delivery timing. Escalation trigger. Five fields. On every handoff. The contract is the artifact the team operates against and the manager audits to.

Field One: Sending Party

Who or what sends the payload. AI (Avoca, ServiceTitan AI summary, Dispatch Pro, Rilla, Hatch, NiceJob, Birdeye, Podium AI Employee, an AI classifier prompt). Human (CSR, dispatcher, tech, advisor, service manager, marketing manager, owner). Specify the role, not the person — Marco may be the tech today, Carlos tomorrow; the contract is with the tech role. For AI senders, specify the tool and the prompt — "Step 17 voice notes prompt v2.3" — so the audit trail traces back to a versioned artifact. For human senders, specify the role and any subordinate authority — "service manager (or designated CSR floor lead on weekend coverage)." The sender is the accountability owner for the payload.

Field Two: Receiving Party

Who or what receives the payload and acts on it. Same specifications as the sender. The receiving party is the verifier — they are responsible for catching drift, missing fields, or contract violations. The receiver's name is in the contract because verification responsibility flows to a named role. A CSR who pastes an AI call summary into ServiceTitan without verification is in violation of the Step 7 contract; the contract names the CSR as the verifier and the 4 p.m. daily review as the verification cadence. The dispatcher who acts on Dispatch Pro recommendations without logging override reasons is in violation of the Step 8 contract; the contract names the dispatcher as the verifier and the Friday standup as the audit cadence. The receiver is named so the audit can find the verification step.

Field Three: Payload

The fields that must be present. This is the substantive part of the contract. The payload for Step 7 (AI call summary to CSR) includes: customer name, phone, address, equipment named, issue described, slot booked, dispatch-fee acknowledgment, sentiment tag, any callback or follow-up flagged. The payload for Step 17 (tech voice notes to service manager) includes: equipment make/model/age, condition findings (with verified readings only), recommendation (repair/partial/replace), customer mood, next step, any photos referenced. The payload for Step 33 (recall classification to service manager confirmation) includes: ticket number, "go back" reason, classifier tag (recall/callback/warranty), classifier confidence, SLA start time, any related tickets in the cluster.

The payload is the field list the team agrees on. Missing fields fire the escalation trigger. The discipline is to enumerate every field before the workflow goes live — not after the first failure exposes the gap. The manager builds the payload list during workflow design, validates with the receiving role, and locks it into the constraint of the AI prompts (if AI is sending) or the structured-note template (if a human is sending). The payload list lives in the workflow map appendix.

Field Four: Delivery Timing

When the payload lands, where it lands, and how the receiver is notified. Real-time during the call (AI summary in the ServiceTitan ticket panel by the time the CSR hangs up). End-of-call (AI Avoca summary in the CSR's review queue within 30 seconds). End-of-day (tech voice notes in the service manager's review queue by 5 p.m.). End-of-week (Friday recap email to owner by 4 p.m. Friday). Pre-arrival (tech pre-route brief in the tech's tablet 30 minutes before the first stop). Post-event (review request text 1-3 hours after the customer signoff).

Timing is the variable that protects the next step. If the tech pre-route brief lands 30 minutes after the tech's start-of-day, the tech rolls without the brief and the AI plug-in fails. If the Avoca call summary lands 8 minutes after the call ends, the CSR has taken the next call and the summary's context is gone. The timing field is not negotiable — it is the timing the next step requires to function, not the timing AI happens to deliver. The manager designs timing backward from the receiver's workflow needs and forward-engineers the AI delivery to meet it.

Field Five: Escalation Trigger

What fires if the payload is missing, late, or wrong. The escalation trigger is what makes the contract enforceable. Trigger one: missing field — a payload that arrives without a required field (Customer Mood absent from Step 17 voice notes) routes to the manager's exception queue with the ticket flagged. Trigger two: late delivery — a payload that arrives after the timing window (Friday recap not sent by 5 p.m.) auto-notifies the owner that the recap is delayed and surfaces the marketing manager's exception status. Trigger three: wrong content — a payload that contains a contract violation (AI summary that invented a customer name or address; AI proposal that drafted an unverified APR) routes to the verifying receiver with a "verify and correct" flag. Trigger four: receiver did not verify — a payload that the receiver pasted without review (CSR who skipped the 4 p.m. verification window) surfaces to the service manager's coaching queue.

Triggers are not punitive; they are informational. The manager reviews triggered exceptions weekly and updates either the AI prompt (if AI is missing fields under load) or the human training (if the CSR is skipping verification). Triggers without follow-up action are noise; triggers with weekly review cycles are how the workflow improves quarter over quarter. The shop with disciplined trigger review has a workflow that gets better; the shop without has a workflow that drifts toward generic AI output and untrained human verification.

AI Hands the Tech: Step 11 Pre-Route Brief

The first major handoff in the day. AI hands the tech a pre-route brief — what the tech needs to know before opening the truck door at the first stop. The contract:

Sender: ServiceTitan Titan Intelligence (or HCP AI Team, FieldEdge mobile). Receiver: field technician. Payload: ticket number; customer name and address; equipment (make, model, age, install date if available); call notes from Step 7 (the CSR's verified AI summary); prior service history on this system; outstanding warranty status; likely parts to pack (based on call description and equipment); financing tier the customer pre-approved for if a soft-pull was triggered on a prior visit; any open complaints, recalls, or warranty issues on the customer's account. Delivery timing: tech's tablet 30 minutes before first stop; refreshed when the dispatcher reshuffles. Escalation trigger: missing critical fields (no call notes from Step 7, no equipment info, no parts recommendation) routes to dispatcher for manual completion before tech rolls.

What this handoff prevents. Without it, the tech opens the ticket on arrival at the home, reads two sentences ("furnace not heating, said it started last night"), starts diagnostic from scratch, asks the homeowner questions Step 14 should have already partially answered, and packs parts on a guess. With the handoff, the tech knows the system is a 14-year-old Goodman, the customer mentioned a smell, prior service in 2024 was an evaporator coil replacement, the customer is on a payment plan for an unrelated debt, and the prior advisor flagged the system as a likely replacement candidate. The tech walks in with context. Step 14's symptom interview becomes 3 minutes instead of 12. The diagnostic is sharper. The repair-or-replace conversation at Step 18 lands at the right pace because the tech has already seen the system's history.

The Tech Hands the Manager: Step 30 Structured Note

End of day, every tech submits structured notes for every stop. The service manager reviews these in the morning routine the next day. The contract:

Sender: field technician (with AI assist on structuring at Step 17). Receiver: service manager. Payload: ticket number; equipment with make/model/age; condition findings with verified readings (static pressure, refrigerant charge, amp draw — actual measurements, no fabrications); recommendation (repair/partial/replace) with reasoning; customer mood (engaged, hesitant, frustrated, ready); financing pre-approval status if soft-pull was triggered; next step (advisor referral, install coordinator handoff, recall follow-up, customer recovery); photos referenced by tag; coaching moments the tech wants the manager to know about (struggled with a homeowner question, missed a membership pitch opportunity, found a system issue I didn't expect). Delivery timing: ticket closed within 90 minutes of the last call signoff; service manager queue receives by 6 p.m.; next-morning review by 9 a.m. Escalation trigger: missing Customer Mood or Next Step routes to tech for completion before close-of-day; ticket open past 90 minutes auto-notifies service manager.

What this handoff enables. The service manager opens the queue at 9 a.m. Wednesday morning and sees 14 tickets from Tuesday across 6 techs. Three are flagged as Comfort Advisor referrals (next step = advisor); two are flagged as recall risks (recommendation reasoning surfaced a concern); one has a coaching moment the tech raised (Marco missed a membership pitch on a 12-year-old system). The manager prioritizes the three advisor referrals into the morning queue, schedules a 15-minute coaching with Marco at the 11 a.m. huddle, and reads through the rest in 35 minutes. Without the structured note, this routine takes 2 hours of unstructured reading; with it, 35 minutes and the manager catches every signal.

The Manager Hands the Owner: Friday Recap

End of week, the service manager (or operations manager, depending on shop org) sends the owner a Friday recap. AI drafts; manager edits; owner reads. The contract:

Sender: service manager with AI assist. Receiver: owner. Payload: 7 numbered lines covering the week's metrics — booking %, MPR, financing close %, recall %, RPT (service and replacement), average ticket (service and replacement), GLSA ROAS; 3 wins (specific tech, specific call, specific customer); 2 problems (specific recurring issue, specific risk surfacing); 1 ask (one decision or resource the manager needs from the owner); Monday priority (one item the owner can act on); celebration line (one team member or customer moment worth naming). Delivery timing: Friday 4 p.m. via email; owner reads within 90 minutes of receipt. Escalation trigger: recap not received by 5 p.m. Friday auto-emails the marketing manager for status; metrics outside the alert thresholds (recall % above 5%, financing close % below 18%, RPT below baseline) trigger a yellow flag in the recap subject line.

What this handoff enables. The owner reads in 4 minutes. The owner has signal on every operating metric, three specific wins to celebrate Monday morning, two problems to acknowledge in the Monday huddle, one decision to make over the weekend, and one priority to come back to. Without the structured recap, the owner gets a 35-minute conversation with the service manager every Friday afternoon and the conversation drifts into anecdotes. With it, the conversation is informed and short. The owner asks better questions Monday because Friday's signal is in their head.

The Customer Handoff: The Fifth Edge

The handoffs above all stay inside the shop. The customer-facing handoffs are the fifth edge — where AI output meets the homeowner directly, with no internal intermediary. The on-my-way text. The review request. The proposal email. The renewal touch. The contract structure is the same; the audit discipline is sharper because blast radius is highest.

Sender: AI tool with brand-voice system prompt (NiceJob, Podium AI Employee, Birdeye AI Employee, Hatch, ServiceTitan automation). Receiver: the customer (homeowner). Payload: bounded to verified inputs only — tech name and photo, ETA from dispatch, customer's first name (or Mr./Ms. depending on shop voice), the address, the equipment named in the prior conversation, the financing payment from the portal, the membership benefit from the plan document, the install anniversary date from the customer record. No fabrications. No invented promises. Delivery timing: SMS or email at the specified offset (25 minutes for on-my-way, 1-3 hours post-visit for review request, end-of-day for proposal, 11 months 2 weeks for renewal touch). Escalation trigger: customer reply outside the AI's confidence threshold routes immediately to a human (CSR for in-hours, on-call dispatcher for emergencies). Unsubscribe or opt-out honored within 1 message cycle. Any complaint flag surfaces to the service manager and owner.

The customer handoff is the highest-blast-radius interaction in the workflow. One wrong on-my-way text touches one customer; one wrong review-response template touches every recent reviewer; one wrong renewal-touch template touches 1,200 install anniversaries. Test 6 of the decision framework (irreversibility and blast radius) applies most aggressively here. The handoff contract has template-level review before any AI-drafted customer message goes live, plus per-instance review by the human approver named in the contract. The verification cadence is daily at low volume (under 50 messages/day), hourly at high volume (200+ messages/day during a marketing push or seasonal peak).

Building the Handoff Contract Library

The deliverable for L3 Chapter 1 Lesson 3 is the handoff contract library. One page per major handoff — the 12-15 contracts that matter most. Sender, receiver, payload, timing, escalation. Lives next to the workflow map and the Decision Rule Quick Card in the shop's Notion, Google Doc, or pinned Slack post. Together the three artifacts — map, rules, contracts — are the manager's operating playbook.

The library compounds quarter over quarter. Each escalation trigger fired produces a contract update or a prompt update or a training update. Q1: AI starts dropping the Customer Mood field on Step 17 voice notes under high-volume conditions; constraint updated, drift resolved. Q2: CSR floor inconsistent on the 4 p.m. Step 7 verification cadence; training huddle and audit log launched. Q3: Friday recap drifts into anecdote without metric anchors; payload field list tightened, AI draft prompt updated. Q4: customer-handoff opt-out compliance audited against TCPA; one disclosure language update, no violations. Each quarter improves a handful of contracts; over a year the library matures from 12 baseline contracts to 18-22 mature contracts with documented version history. The shop's AI workflow is now a versioned operating asset that survives turnover and audit.

The library is also what makes the shop's L4 and L5 work possible. An owner cannot evaluate an AI tool's procurement (L4) without knowing what handoffs the tool fits into. A multi-shop operator cannot roll out an AI strategy across 25 locations (L5) without a portable handoff contract library each location can adapt to local context. The contracts are the substrate for every higher-level decision. The manager who masters L3 has built the operational foundation that L4 and L5 strategy plug into; without it, L4 strategy is wishful and L5 rollout is fragile.

The Three Artifacts Together: Map, Rules, Contracts

This concludes L3 Chapter 1. The three lessons produced three artifacts that, together, are the manager's AI workflow operating playbook. Lesson 1 produced the workflow map — 38 steps, 5 columns, on the wall above the dispatch board. Lesson 2 produced the Decision Rule Quick Card — six tests, applied per step per sub-decision, posted next to the map. Lesson 3 produced the handoff contract library — 12-15 contracts per major seam between AI and human, lived in the team's Notion or Slack.

Map describes what. Rules describe when. Contracts describe how. The CSR who looks at the map knows where Avoca fits. The dispatcher who reads the rules knows why Dispatch Pro can be overridden 5-15% of the time. The tech who follows the contract knows what fields go into the Step 17 voice note. The service manager who audits the contracts knows what to investigate when the Step 33 recall classification drifts. The marketing manager who reads the customer-handoff contract knows what opt-out cadence is enforced. The owner who reviews the Friday recap knows the recap is structured because the manager built the contract.

Three artifacts. One operating playbook. The 60% of trades shops who hit their AI metric targets at month two have these three artifacts. The 40% who hit them at month four do not. The next 7 chapters of L3 — missed-call recovery, dispatch and smart routing, ride-along coaching, RC&D triage, marketing spend attribution, advanced workflows, advanced prompting — all build on this Chapter 1 foundation. Without the map, rules, and contracts, the workflows are improvised. With them, the workflows are designed, defensible, and durable.

Key Takeaways

• Shops do not fail at AI steps; they fail at handoff seams. AI generates competently; humans receive context that evaporates without a contract; fields go missing under load; timing slips; escalations never fire because no one designed the trigger.
• Every handoff has the same five-field contract. Sending party, receiving party, payload, delivery timing, escalation trigger. The structure is invariant; the content per handoff varies.
• AI hands the tech a pre-route brief at Step 11. 30 minutes before first stop. Ticket, customer, equipment, history, parts, financing, complaints. Missing critical fields route to dispatcher for completion before tech rolls.
• The tech hands the manager structured notes at Step 30. Equipment, condition (verified readings only), recommendation, customer mood, next step, coaching moments. Missing fields route to tech for completion before close-of-day.
• The manager hands the owner the Friday recap. 7 metric lines, 3 wins, 2 problems, 1 ask, Monday priority, celebration. AI drafts; manager edits; owner reads in 4 minutes. Yellow flags fire on threshold breaches.
• The customer handoff is the fifth edge — highest blast radius. AI-drafted on-my-way, review request, proposal, renewal touch. Template-level review before deployment plus per-instance human approval. Verification cadence scales with volume.
• Three artifacts together: map, rules, contracts. Map describes what. Rules describe when. Contracts describe how. L3 Chapter 1 ends with all three on the wall, next to each other, working in concert. Without them, L4 and L5 are wishful.
• Handoff library compounds quarter over quarter. Every fired escalation produces a contract, prompt, or training update. Year 1 matures 12 baseline contracts into 18-22 versioned contracts; the workflow becomes an operating asset that survives turnover and audit.