A service call is not a single event. It is a 38-step process that starts when a homeowner picks up the phone at 6:47 a.m. on a Tuesday and does not end until the renewal touch lands in the customer's text inbox at month 11. Inside those 38 steps live the booking, the dispatch, the on-my-way text, the diagnostic, the repair-or-replace conversation, the financing soft-pull, the close, the install, the warranty registration, the review request, the recall-or-callback triage, the membership pitch, and the renewal upsell. AI plugs in cleanly at 19 steps, clumsily at 11, and actively makes things worse at 8. The manager who does not know which 8 has a workflow that leaks margin every Friday and a recall percentage trending the wrong way. This lesson is the end-to-end map. Every step named, every AI plug-in tagged with where it wins, where it stays out, and where it breaks the customer experience. The manager who finishes this lesson can defend their AI workflow design in front of an owner, a Nexstar peer call, a Wrench Group portfolio review, or a PE board on slide 4.

Why End-to-End Mapping Matters Before Tool Selection

Most shops that fail at AI in 2026 fail the same way. The owner reads the Avoca case study and the Rilla close-rate lift, calls the vendors, signs three contracts, and tells the manager to make it work. Three months later booking percent has moved 2 points, close rate is flat, and the dispatcher is yelling about why the AI sent Marco to the recall again. The tools are not the problem. The map is. The shop bought tools that fit individual steps without knowing how the steps connected, so the tools optimize locally and break globally.

The discipline is inverted from what most owners try first. Map the call end-to-end before naming a single vendor. Tag each step with the role that owns it, the metric that moves with it, the data that crosses into the next step, and the failure mode if it breaks. Then — and only then — overlay AI plug-ins on the map. The Avoca seat is not a feature purchase; it is a step-1-through-step-5 replacement that requires step-7 (CSR handoff verification) to be redesigned. The Rilla seat is not a coaching expense; it is a step-18-through-step-22 instrumentation layer that requires step-30 (service manager scorecard review) to be added to the manager's daily routine. The map comes first; the tools fit the map.

This is what L2 trained operators to do at the role level — a CSR mapping daily prompts, a tech mapping the job-notes flow, an advisor mapping kitchen-table pivots. L3 is where the manager maps across roles, across data systems, and across time — from the inbound ring to the eleven-month renewal touch. Without the cross-role map, AI is a pile of seats. With it, AI is a workflow.

The 38 Steps of a Service Call

Here is the map. Every step is numbered, named, owned by a role, and timed. The numbering is the manager's reference frame for the rest of L3 — every chapter from this point on names steps by number when discussing where AI plugs in and where it doesn't. Memorize the numbering. Tape it to the wall above the dispatch board.

Phase One: Inbound and Booking (Steps 1-7)

Step 1: Inbound ring. Homeowner dials — Google LSA click-to-call, referral, or the sign on the truck. Owned by phone system. Metric: ring-to-answer time. Step 2: Answer or miss. Live CSR or AI receptionist picks up, or voicemail. Owned by Avoca / Jobber AI Receptionist / Housecall Pro AI Agents / live CSR. Metric: answer rate (target 100%, baseline ~72-78%). Step 3: Intent classification. Service call, sales inquiry, price-shopper, warranty, emergency? Owned by AI or CSR ear. Step 4: Triage. Emergency to dispatch hot list, routine to slot, sales to advisor scheduler, warranty to ticket. Owned by CSR or AI router. Step 5: Slot offer and booking. Offer window, capture address, name, equipment, problem, dispatch-fee acknowledgement. Owned by CSR or AI receptionist. Metric: booking percent (target 80-85%, baseline 65%). Step 6: Confirmation send. Text with appointment time, tech photo placeholder, dispatch-fee disclosure, portal link. Owned by ServiceTitan / Sera / HCP automation. Step 7: Customer record update. Ticket created; CSR pastes AI call summary (15 seconds with AI, 4 minutes without). Owned by CSR. Metric: after-call work time.

Phase Two: Pre-Arrival and Dispatch (Steps 8-13)

Step 8: Dispatch board placement. Dispatcher assigns by skill, capacity, geography, revenue-per-truck targets. Dispatch Pro re-evaluates every 10 minutes; dispatcher overrides 5-15% with documented reasons. Owned by dispatcher with AI assist. Metric: dispatch yield. Step 9: Pre-arrival prep text. Tech name, window, photo, bio, ETA link. Owned by automation. Step 10: Reminder cadence. Day-before, morning-of, ETA update if dispatch shifts. Owned by automation. Metric: show rate (target 92%+). Step 11: Tech pre-route prep. Tech reads the AI-summarized call notes, equipment history, packs likely parts. Owned by tech. Step 12: On-my-way text. 25-30 minutes out, AI text with photo, bio, review nudge, ETA. NiceJob / Podium AI Employee / Birdeye AI Employee. Owned by automation. Step 13: Arrival and intro. Tech parks, walks up, introduces. Owned by tech. Metric: ride-along intro score (Rilla RGA).

Phase Three: Diagnostic and Presentation (Steps 14-22)

Step 14: Symptom interview. What the homeowner noticed, when, what changed recently. Owned by tech. Step 15: System inspection. Pressures, charges, voltages, ductwork, drainage. Owned by tech. Critical: licensed-only work. Step 16: Photo and reading capture. Equipment plate, condition issues (corrosion, refrigerant stain, panel deficiency). AI annotation suggested on tablet. Owned by tech. Step 17: Voice-to-structured notes. Tech dictates; AI structures into 5-section format (Equipment, Condition, Recommendation, Customer Mood, Next Step). Owned by tech with AI assist. Step 18: Repair-or-replace decision frame. Equipment age × repair cost vs. replacement cost, customer goals, rebate, financing payment. AI surfaces the 3-option (repair / partial / replace) talk-track. Owned by tech with AI assist. Step 19: Financing soft-pull at the door. Wisetack / GreenSky / Synchrony soft-pull; approved-up-to amount lands on advisor tablet before the kitchen table. Owned by tech / advisor. Step 20: Repair authorization or replacement handoff. Repair: customer approves and tech repairs. Replace: advisor schedules or arrives same-day. Critical: authorization above $1,500 is human. Step 21: Membership pitch. 30-second AI-tailored pitch by equipment age, repair cost, condition. Owned by tech with AI assist. Metric: MPR (target 35-50%, top quartile 60%+). Step 22: Kitchen-table close. Advisor presents good/better/best, financing math, 25C/25D credit stack, warranty terms. Rilla records and scores. Owned by advisor. Metric: close rate (target 45-60%).

Phase Four: Execution and Handoff (Steps 23-30)

Step 23: Execution / install scheduling. Repair tech executes; install advisor hands off to install coordinator. Owned by tech or coordinator. Step 24: Parts and inventory. Tech pulls or orders; truck inventory replenishes via AI-flagged restock. Owned by ops. Step 25: Job documentation. Equipment, serial numbers, warranty registration, photos, invoice line items. Owned by tech. Step 26: Customer signoff and payment. Customer signs; payment runs through ServiceTitan / HCP; financing portal completes. Owned by tech or advisor. Step 27: Warranty registration. Manufacturer warranty registered with serial; shop labor warranty terms in the ticket. Owned by tech with automation. Step 28: Departure notes. Tech dictates; AI structures for dispatch handoff. Step 29: Dispatcher reconciliation. Actuals vs. plan, override outcomes logged, tomorrow's board adjusted. Owned by dispatcher. Step 30: Service manager review. End-of-day pull, callback-in-the-making scan, coaching moments flagged. AI surfaces patterns. Owned by service manager.

Phase Five: Post-Call and Loyalty (Steps 31-38)

Step 31: Review request. 1-3 hours post-visit, AI-drafted text or email with 1-click link. NiceJob / Podium AI Employee / Birdeye AI Employee. Owned by automation. Metric: review velocity (target 4-7 per truck per month). Step 32: Review response. AI drafts within 48 hours; human approves and posts. Owned by marketing or service manager with AI assist. Step 33: Callback / recall / warranty triage. Any "go back" within 30 days gets classified: recall (same problem, same system — shop eats it), callback (tech missed something — internal), warranty (manufacturer part — covered with light labor allowance). AI classifier tags; service manager confirms. SLA tiers: recall 24-hour, callback 48-hour, warranty 5-business-day. Step 34: Root-cause clustering. Weekly cluster: by tech, by part, by symptom, by install date. AI heatmap in 10 minutes. Owned by service manager. Step 35: Customer recovery. AI-drafted apology + remedy; manager edits. Owner final sign-off above threshold. Step 36: Membership onboarding. Welcome email, plan-benefit summary, first maintenance touch. Owned by automation. Step 37: 11-month renewal touch. AI-built text references install anniversary, next maintenance due, the customer's specific system. Owned by automation with marketing edit. Step 38: Renewal close or churn. Renew, upgrade, downgrade, or churn. AI-flagged churners route to senior CSR retention call.

Where AI Plugs In Cleanly

Nineteen of the 38 steps are clean AI wins in 2026, documented across Avoca's HL Bowman case, ServiceTitan's 2026 State of AI in the Trades data, Rilla's 18% close-rate lift, and the Hatch nurture cases.

Steps 1-2 (Answer rate): Avoca / Jobber AI Receptionist / Housecall Pro AI Agents / ServiceTitan Voice answer at 100% — recover the 22-28% missed-call baseline. Step 3 (Intent): AI tags inbound intent in 8 seconds. Step 5 (Booking): AI books direct to FSM for routine; warm-transfers complex. Step 6 (Confirmation): Brand-voice-locked AI text. Step 7 (ACW): AI summary collapses 4 minutes to 15 seconds. Step 8 (Dispatch): Dispatch Pro / Sera / FieldEdge — 12-18% yield lift in 60 days. Steps 9-10 (Pre-arrival, reminders): AI texts at every touch. Step 11 (Pre-route prep): AI surfaces history and parts. Step 12 (On-my-way): AI text with photo, bio, review nudge. Step 16 (Photo annotation): Rust, corrosion, drain belly, panel deficiency. Step 17 (Voice notes): 5-section structured output. Step 18 (Repair-or-replace frame): AI surfaces 3-option talk-track; human delivers. Step 21 (Membership pitch): AI-tailored 30 seconds. Step 28 (Departure notes): AI structures for dispatch. Step 31 (Review request): AI-drafted, 1-click link. Step 32 (Review response): AI drafts; human approves. Step 33 (Recall classification): AI classifier tags by SLA tier. Step 34 (Root-cause clustering): AI heatmap in 10 minutes. Step 37 (Renewal touch): AI text on install anniversary and system specifics.

Every one of the 19 moves a number on the P&L — booking %, dispatch yield, MPR, close rate, recall %, review velocity, renewal rate. Every one has a 2026 vendor and a 90-day payback. The shop that lights all 19 in the first six months captures roughly 80% of the available AI margin lift in the trades.

Where AI Doesn't Plug In

Eleven steps require human judgment, licensed expertise, or eye contact that AI cannot replicate in 2026. AI these steps and the manager creates more work, or worse, creates liability.

Step 4 (Emergency triage): AI classifies intent, but bumping a routine to the hot list requires the dispatcher's read on volume, weather, and crew. AI flags; humans decide. Step 13 (Arrival intro): The first 90 seconds is human chemistry. AI cannot script around regional voice, homeowner mood, or the dog at the door. Step 14 (Symptom interview): Active listening and pattern recognition on what the homeowner is not saying. AI transcribes; AI cannot replace the conversation. Step 15 (System inspection): Licensed work — EPA 608, state contractor board. The licensed tech cannot delegate. Step 20 (Repair authorization above $1,500): Customer signature required. Above $5K disputed language, owner final sign-off. Step 22 (Kitchen-table close): Eye contact, pacing, reading hesitation. AI preps the talk-track; AI cannot close. Step 23 (Install execution): Crew coordination, lineset routing, drain pitch, breaker sizing — licensed field judgment. Step 24 (Parts decisions): AI flags restock; tech and warehouse decide what fits. Step 26 (Signoff and payment): Human-witnessed transaction; Reg E and PCI dictate human dispute handling. Step 27 (Warranty signing): AI fills the form; licensed installer signs. Step 35 (Customer recovery above threshold): AI drafts; owner signs above refund threshold. Disputes, online-review-risk complaints, and regulatory complaints require owner judgment.

Where AI Makes Things Worse

Eight steps actively degrade when AI is bolted on without judgment. The manager protects the shop by keeping AI out, or tightly constraining what it does.

Step 19 (Soft-pull explanation): AI triggers the soft-pull and surfaces the approved amount, but explaining what a soft-pull is, what it does to credit, and what the disclosure means is human. Reg Z and FCRA exposure follows misstatements; a hallucinated APR is a regulatory event. Constraint locks AI to portal numbers verbatim; the human explains. Step 22 (AI-scripted close attempts): Letting AI write "you should buy now because" breaks the close. AI preps; the human owns the moment. Step 25 (AI-fabricated readings): AI fills in plausible static pressure, refrigerant charge, amp draw when dictation skips them. That becomes the warranty trail. Constraint must lock AI to verified inputs. Step 27 (AI-invented warranty terms): AI drafts confident warranty language that contradicts the manufacturer. The shop is bound by what the customer signs. Manufacturer terms come from the manufacturer, verbatim. Step 32 (Auto-posted review responses): Auto-posting without human review gets flagged by Google as inauthentic, harms local SEO, and apologizes for issues the shop did not commit. Human approval non-negotiable. Step 33 (AI-only recall classification): Classifier suggests; service manager decides. Mis-classified callbacks (called warranty) become liability when the manufacturer denies; mis-classified recalls hide coaching. Step 35 (AI-drafted disputes without owner edit): AI admits fault the shop does not own. Disputes require owner judgment on what to concede and what to defend. Step 38 (AI auto-closed renewals): AI flagging churners is good; AI auto-closing renewals misses the retention conversation. High-value renewal calls are human.

The pattern is consistent. AI degrades the workflow when it produces customer-facing language without verification, makes decisions carrying regulatory or licensing exposure, or replaces the human moment that closes, recovers, or retains. The L1 Cardinal Rule applies through every step: nothing AI says about a customer goes to the customer without a human eye on it. At L3 the manager designs the workflow so the human eye is built in.

Building the Step-by-Step AI Overlay

The deliverable for L3 Chapter 1 is the shop's AI overlay map. One page. 38 steps down the left column. Five columns: role owner, primary metric, AI plug-in (cleanly / clumsily / not-at-all), vendor or tool, human verification point. The map aligns the dispatcher, CSR floor, techs, advisors, marketing manager, and owner around what AI is doing at every step and what the human is doing.

Every cell is concrete. Step 2 (Answer): role owner = AI receptionist; metric = answer rate; AI plug-in = clean; vendor = Avoca; verification = 4 p.m. daily CSR review of AI-booked calls. Step 18 (Repair-or-replace): role owner = tech; metric = replace-attach rate; AI plug-in = clumsy (AI surfaces frame, tech delivers); vendor = ResponsiBid + ServiceTitan templates; verification = advisor confirms financing math. Step 22 (Kitchen-table close): role owner = advisor; metric = close rate; AI plug-in = not at all; vendor = Rilla for instrumentation only; verification = every word at the table.

Every shop that hits its AI metric targets at month two has the map on the wall. Every shop that hits them at month four does not. The map is the difference between buying tools and designing a workflow. Without L2's prompt discipline the map has no operating substrate; without the L1 Cardinal Rule it has no safety net.

The Renewal Loop and the Eleven-Month Touch

The renewal loop is the step most shops ignore and AI is built for. The customer who bought the $14,200 replacement in March is the highest-LTV asset on the books. The membership plan attached generates $189-$359 per year for as long as the customer renews. The renewal touch at month 11 is the difference between 70% and 88% retention — at 70%, average customer worth roughly 3.3 years of plan revenue; at 88%, roughly 8.3 years. Same install conversion. Same install cost. A 2.5x LTV swing on the renewal-loop discipline alone.

AI is built for this because it is the loop most shops cannot operate manually at scale. Pulling 1,200 install anniversaries, segmenting by equipment, customizing the message to the system the customer owns, surfacing next-maintenance, computing renewal price with loyalty discount — 18 hours a week or never gets done. AI does it in 12 minutes; marketing manager edits and approves; automation sends. Retention moves; LTV moves; exit multiple moves (PE buyers scrutinize LTV/CAC and recurring revenue in 2026).

A shop with disciplined Steps 37-38 has a different P&L 24 months out than a shop that bolts AI onto answer rate and ignores the back half of the call. End-to-end mapping forces the manager to value Step 37 the same as Step 2 — both are AI plug-ins, both move a number, both compound.

The Manager's Deliverable for This Chapter

By the end of this lesson, the service manager or ops manager builds a one-page deliverable. Title: "Service Call Workflow Map — [Shop Name]." Format: 38 rows, 5 columns (Step / Role Owner / Primary Metric / AI Plug-In Type / Vendor / Verification Point). Constraint: every row filled, no placeholders. Cost: 2 hours the first time, 30 minutes to update quarterly. Lives on the wall above the dispatch board, in the shop's shared Notion or Google Doc, and in the appendix of the AI roadmap the owner reviews.

Three effects show up in the first two weeks on the wall. The CSR floor stops arguing about which calls Avoca should be answering — the map answers it. The dispatcher stops overriding Dispatch Pro without logging the reason — the Step 8 verification column makes the discipline visible. The service manager stops triaging recalls by gut — the Step 33 classifier flow is named, SLA tiers documented, the heatmap (Step 34) runs every Monday at 9 a.m. Three weeks in and the shop's AI workflow operates differently than before, with the same tools already on contract.

The next two lessons operationalize the map. Lesson 2 covers decision rules for AI-ready vs. human-only steps. Lesson 3 covers handoff design — what AI hands the tech at Step 11, what the tech hands the manager at Step 30, what the manager hands the owner Friday. Map is the substrate; decision rules are the operating logic; handoffs are how the workflow moves between humans and AI without dropping the ball.

Key Takeaways

• A service call is 38 steps, not one event. From the inbound ring at 6:47 a.m. to the renewal touch at month 11. Every step has a role owner, a primary metric, an AI plug-in type, a vendor, and a human verification point. The map is the artifact that names them all.
• Nineteen of the 38 steps are clean AI wins in 2026. Answer rate, intent classification, booking, confirmation, ACW, dispatch, reminders, on-my-way, photo tagging, voice notes, repair-or-replace frame, membership pitch, departure notes, review request, review response, recall classification, root-cause clustering, renewal touch.
• Eleven steps require human judgment, licensed expertise, or eye contact. Emergency triage, arrival intro, symptom interview, system inspection, repair authorization above $1,500, kitchen-table close, install execution, parts decisions, customer signoff, warranty registration signing, customer recovery above a threshold.
• Eight steps actively degrade when AI is bolted on without judgment. Financing soft-pull explanation, AI-scripted close attempts, AI-fabricated readings in documentation, AI-invented warranty terms, auto-posted review responses, AI-only recall classification, AI-drafted dispute apologies, AI auto-closed renewals.
• The L1 Cardinal Rule operates through every step. Nothing AI says about a customer goes to the customer without a human eye on it. At L3 the manager designs the workflow so the human eye is built in at the verification points named on the map.
• The renewal loop (Steps 37-38) is the most-ignored AI win in the trades. A 70% to 88% retention shift takes LTV from 3.3 years to 8.3 years on the same install. AI is built for the eleven-month touch; most shops never wire it up.
• The deliverable is the one-page workflow map. 38 rows, 5 columns, on the wall above the dispatch board. The shop that builds the map captures 80% of the available AI margin lift in the first six months. The shop that skips the map buys tools, not a workflow.