The single worst sentence on a Monday morning service-manager dashboard is "see notes." Two words, no equipment make, no capacitor reading, no recommendation, no customer mood, no next-step. The dispatcher reads it, sighs, and calls the tech mid-job to ask what actually happened on the 11:30 a.m. stop. The tech, kneeling next to a 2008 condenser in 94-degree heat, has to recall a conversation from four hours ago. The follow-up call eats 6 minutes. Multiply by 18 trucks and a four-day service week, and a single shop burns 7-9 hours of dispatcher-to-tech callback time every week — roughly $620 of payroll and a 4-7% drag on dispatch yield that no one ever tracks. AI job-notes kill the "see notes" problem in 90 seconds. The tech speaks 60-90 seconds of curbside voice memo into the tablet. The 5-part prompt routes it through a structured-notes template. The output is a six-field card — equipment, condition, recommendation, customer mood, financing posture, next-step — that the service manager reads in 11 seconds and the Comfort Advisor reads off the tablet at the kitchen table. Three minutes total, not twelve. Dispatcher follow-up calls cut roughly in half. This lesson is the prompt build, the six-field structure, the tool stack, the verification discipline, and the 60-day deployment timeline that lifts dispatcher productivity 18-24% and average ticket $90-$180.

Why Job-Notes Are the Single Highest-Leverage Tech Artifact

The tech is the only role in the shop that creates the record of what actually happened on the property. The CSR books. The dispatcher routes. The Comfort Advisor closes. But only the tech stands next to the equipment, reads the model tag, measures the capacitor, traces the refrigerant stain, and watches the homeowner's face when they hear the price. Everything downstream — the recall flag, the membership pitch, the replacement proposal, the warranty claim, the maintenance touch 11 months later — runs on the tech's notes. The notes are the source-of-truth artifact for the next six revenue events tied to that customer.

ServiceTitan's 2025 trades benchmark flagged complete structured job-notes as one of the three operational variables most correlated with $300K+ annual revenue per service truck, alongside dispatch-yield discipline and membership conversion rate. Built on Tenth's 2026 dispatch-yield study found that shops with 95%+ notes completeness ran 18-24% higher dispatcher productivity than shops at 60-70%, because the dispatcher running an exception loop on incomplete notes spends roughly one full shift per week on follow-up calls instead of building the next day's board.

The trouble is that techs hate writing notes. The tablet is small, the keyboard is finger-fat, the homeowner is hovering, the next stop is dispatched in 14 minutes. The average untrained tech writes 15-25 words of free-text per ticket and skips the structured-fields panel entirely. The average AI-assisted tech speaks 60-90 seconds of voice memo and the AI produces a 90-130 word record across six fields. The difference is not effort; it is interface.

The Six-Field Structured-Notes Template

Every job-note in the shop comes out of the AI in the same six fields, in the same order, every time. The structure is what makes the dispatcher's read take 11 seconds instead of 90, and what makes the Comfort Advisor's read produce a useful kitchen-table recap rather than a search through paragraph soup. The six fields: Equipment, Condition, Recommendation, Customer Mood, Financing Posture, Next Step. Memorize that order — it survives the Tuesday huddle, fits on an index card taped to the inside of the tablet flip-cover, and matches the schema every shop's CRM, dispatch system, and AI voice-to-text layer can consume without translation.

Field One: Equipment

Make, model, serial when the tag is legible, age computed from the serial date or estimated from the install if the tag is corroded off. System type — 16 SEER2 single-stage condenser with a matching 96% AFUE upflow, a tank-style 50-gallon natural-draft water heater, a 200-amp Square D QO panel with two open slots. Be specific enough that the Comfort Advisor reading the notes for the kitchen-table sit-down tomorrow does not have to call the tech to ask what brand. The AI fills this field from the tech's spoken memo plus the photo of the data plate. Photo-to-text on data plates is one of the most-reliable AI patterns in the 2026 trades stack — ServiceTitan, Sera, Housecall Pro, and Workiz all expose photo-OCR for equipment tags and the accuracy on legible tags is 96-99%.

Field Two: Condition

What the tech actually found. Capacitor reading, refrigerant pressures or weight, amp draws on the compressor and fan motor, contactor condition, evaporator coil condition with photo references, drain line condition, electrical-panel deficiencies, water-heater anode condition, gas-pressure readings, flue draft. The condition field is the part of the note the service manager and the next tech read most often, because callbacks and warranty claims live in this field. The AI never invents a number — every capacitor microfarad reading, every PSI, every amp draw is either dictated by the tech ("compressor pulled 18.4 amps locked-rotor") or pulled from a meter photo. The constraint baked into the prompt is explicit: do not fabricate numeric measurements. If the tech did not dictate a number and there is no meter photo, the AI leaves the field blank and surfaces it as "tech to measure on return." A fabricated capacitor reading that triggers a warranty replacement is exactly the kind of liability moment that ends a shop's relationship with the manufacturer's distributor.

Field Three: Recommendation

What the tech recommends the customer do next, in the customer's frame, not in tech-jargon. Repair the capacitor and the low-voltage transformer for $487. Replace the 19-year-old condenser before next summer; the compressor short-cycling pattern indicates 12-18 months of remaining life. Schedule a Comfort Advisor for a three-option replacement quote. Add the surge-protect at $189 because the panel has three burned-out neutrals and the homeowner already lost a fridge in last month's lightning storm. The recommendation field is what the Comfort Advisor reads when they walk in the door for tomorrow's sit-down. It is also the field that drives membership upgrade conversations and the recall touches.

Field Four: Customer Mood

The single field most undervalued by techs and most valuable to the rest of the shop. One sentence. "Homeowner is frustrated — second time the same furnace has tripped this winter, husband on fixed income, mentioned the previous shop oversold them in 2021." "Homeowner is excited — just bought the property, wants the whole panel replaced and is asking about EV charging." "Homeowner is uncertain — wife wants replacement, husband wants repair, both mentioned the upcoming knee surgery and budget pressure." Customer mood drives every downstream conversation. The Comfort Advisor walking in to a "frustrated, fixed income" mood writes a different proposal than the one walking into "excited, EV charging." Mood is the field that distinguishes shops that retain customers from shops that burn through them.

Field Five: Financing Posture

Open or closed. If the tech floated financing — Wisetack, GreenSky, Synchrony — what did the homeowner say. "Open to financing — said the monthly payment matters more than total." "Closed — said they have cash but won't decide today." "Open with a soft yes on Wisetack — wants the 0% promo confirmed before sit-down." The Comfort Advisor reads this field and walks into the kitchen knowing whether to lead with payment, total, rebate stack, or cash discount. The financing posture field, captured at the diagnostic, is the single biggest determinant of whether tomorrow's $14,000 replacement quote closes in 45 minutes or in 95. Skip the field and the Advisor wastes the first 25 minutes of the sit-down probing what could have been a single sentence in the job-note.

Field Six: Next Step

One sentence. Booked the Comfort Advisor for Wednesday 4-6 p.m. Quoted the $487 capacitor repair, customer said call back tomorrow. Need to return Thursday with the 30-amp double-pole breaker. Recall in 30 days to confirm the temporary R-22 charge held. The next-step field is the action item the dispatcher works off. It is also the field most strongly correlated with shop revenue retention, because the recall loop, the membership upgrade loop, and the replacement nurture loop all originate here. A shop with 95% next-step capture runs 11-14% higher 12-month customer retention than a shop with 50%. AI surfacing this field — by directly asking the tech "what is the one specific thing the dispatcher needs to put on tomorrow's board?" — is the single highest-revenue prompt move in this lesson.

The 5-Part Prompt the Tech Uses Curbside

The tech does not type seven fields between stops. The tech speaks 60-90 seconds of voice memo into the tablet at the curb after the job, while the homeowner watches from the porch. The 5-part prompt from Chapter 1 is preloaded as a saved template in the shop's AI tooling (Avoca tech-companion, Housecall Pro AI, ServiceTitan Titan Intelligence, Workiz Genius, Sera AI, FieldEdge Voice, or the shop's own Claude/ChatGPT workspace wired into the CRM). The tech opens the template, taps the mic, and talks. The five parts of the prompt are baked in.

Role, Context, and Task

Role: "I am a residential HVAC service tech at a 14-truck Texas shop, just finished a no-cool diagnostic on a 2008 split system for a 67-year-old homeowner whose wife was hospitalized last week and is back home recovering." Real role plus real other-side-of-the-conversation framing. The AI's vocabulary, tone, and inference about what matters in this job snap into focus.

Context (the heart of the dictation): "Outdoor condenser is a 16-SEER R-22 system manufactured June 2008, compressor amps were 23.1 LRA on a 19-rated nameplate, capacitor measured 28 mfd on a 40+5 rated, low-voltage transformer reads 18 volts on a 24-volt circuit. Drain line cleared. Indoor coil has rust streaking on the south face. Homeowner mentioned the system has been short-cycling for two months. Husband is on Medicare, wife just home from surgery, both said they 'cannot go without cool again this summer.' Wisetack is the shop's primary financing partner." Voice-spoken, 30-40 seconds. The AI has the equipment, the condition, the customer story, and the financing context.

Task: "Write a structured job-note in the six-field format: Equipment, Condition, Recommendation, Customer Mood, Financing Posture, Next Step." The verb is "write," the shape is "six-field structured." No essay, no paragraph soup.

Format and Constraint

Format: "Output as six labeled fields, each 1-3 sentences, total under 130 words. Do not include preamble. Do not include sign-off. Direct copy-paste into the ServiceTitan job-note panel." The dispatcher reads it in 11 seconds. The Comfort Advisor reads it on the tablet at the door tomorrow.

Constraint: "Do not invent any numeric measurements. Use only the values I dictated. If a field requires a number I did not give you, write 'tech to measure on return' in the Next Step field. Do not promise a specific repair price. Do not cite a SEER rating I did not dictate. Do not promise a Wisetack approval — only note 'financing posture' as the customer's response. Match the shop's voice — direct, no exclamation points, no marketing language." The constraint is what keeps the job-note from drifting into fabricated readings, invented prices, or chatbot-flavored prose. The constraint is the difference between a job-note the warranty department signs off on and a job-note that triggers an audit.

Total tech effort: open the template, tap mic, talk 60-90 seconds, tap stop, tap generate, skim the six fields, edit one or two words, tap save. Three minutes from end-of-job to dispatcher-readable record. The pre-AI baseline was 12-15 minutes of thumb-typing, and roughly 35% of techs skipped fields entirely. The AI version produces 95%+ field-completeness and the tech is back in the truck cab in under three minutes.

The Tool Stack and Where the Voice-to-Text Actually Runs

Every CRM and FSM platform the trades use in 2026 has either shipped or is shipping a voice-to-structured-notes feature. The shop does not have to build this from scratch; it has to configure the prompt and the field map.

ServiceTitan Titan Intelligence is the largest deployment surface. It ships with a tech-notes assistant that wraps the existing ticket form. The shop configures the prompt template at the company level and the tech sees a voice-to-notes button on every ticket. Field mapping into ServiceTitan's invoice line items, equipment records, and customer history is native. Typical deployment: 4-6 weeks for prompt calibration, 2 weeks of pilot with three to five techs, then full-truck rollout.

Housecall Pro AI is more lightweight — fewer configuration knobs, faster time-to-deploy. Good fit for 1-6-truck shops. Workiz Genius AI, Sera AI (a 2026 release tied to the profit-aware scheduling stack), and FieldEdge Voice all support a structured-notes workflow with comparable shape. The six-field template ports across all three with minor field-name remapping. Sera in particular ties the notes back into the dispatch-yield optimization layer, surfacing the next-step field directly into the next-day's auto-route logic — the highest-integration option in the 2026 stack.

The Claude and ChatGPT Workspace bridge. Shops without a native FSM AI layer run the six-field template through a shared Claude Projects workspace or ChatGPT Team workspace with the template saved as a system prompt. Tech records voice memo on phone, drops the transcript in, AI returns the six fields, tech copy-pastes into the CRM. Slower than native (3-4 minutes vs. 2.5), but operational in 24 hours rather than 4-6 weeks. Most shops should treat this as the bridge while the native integration is configured.

The Verification Discipline That Keeps Fabricated Readings Out of the Record

The 30-second verify habit from L1 applies to every job-note before it saves to the ticket. The tech reads the six fields back and runs the 5-checkpoint mental scan: numbers, names, parts, warranty/code language, financing language. Job-notes are where fabricated readings cause the most expensive downstream damage, so the verify is non-optional.

Numbers checkpoint. Every capacitor reading, PSI, amp draw, voltage, model number. The tech scans the Condition field and confirms each number matches what they actually measured. A fabricated 28-mfd reading on a capacitor that was actually 39 mfd creates a false-warranty-claim trail; the manufacturer's distributor catches it in a quarterly audit and the shop loses the warranty-replacement relationship.

Names and equipment. Customer name, address, model number, brand. The AI sometimes substitutes a similar-sounding brand from training data (Mitsubishi Electric vs. Mitsubishi Heavy, Fujitsu vs. Fujikura). The tech checks the brand string against the photo of the data plate.

Parts and warranty. AI hallucinates part numbers more often than any other field — a non-existent capacitor part number ordered to the truck means the next-day return trip fails. The AI will guess "10-year parts" on a unit that has 5-year parts unless the tech has dictated the actual warranty terms.

Financing and regulatory. Any Wisetack APR, GreenSky tier, Synchrony promotional language. The AI must not invent financing terms. The tech confirms the financing posture field reads as the customer's response — "open to financing" — and contains no APR, payment amount, or approval language. APR fabrication is a Reg Z exposure; the owner has personal liability for the language a tech's note records. The verify pass takes 30 seconds. The tech who skips it is the tech the shop loses their warranty relationship over, six months later, on a quarterly distributor audit nobody saw coming.

The Dispatcher Side of the Equation

The dispatcher is the second-largest beneficiary of structured AI job-notes after the tech who writes them. The Built on Tenth dispatch-yield study quantified the savings: shops moving from 60-70% completeness to 95%+ recovered roughly one full shift per dispatcher per week. The math: each follow-up call takes 5-7 minutes. An average dispatcher runs 14-22 follow-up calls per day in a low-completeness shop, totaling 90-140 minutes daily. At 95%+ completeness, follow-up call volume drops to 3-6 per day (the genuinely-novel ones the structured field cannot answer), saving 70-110 minutes per shift.

The downstream effect compounds. The dispatcher with 90 extra minutes a day builds a 3-5% tighter board, surfaces 1-2 additional re-dispatch opportunities, and runs the 4 p.m. CSR-floor audit with actual capacity to coach instead of just react. The 18-24% dispatcher-productivity lift is the integral of these effects over 60 days. The owner sees the lift in two reportable lines: dispatch-yield ($/truck/day) up 6-10% and average ticket up $90-$180 because the next-step field is now actually worked into the recall loop.

The 60-Day Deployment, the Failure Modes, and the Metrics

Week 0: the service manager and lead tech draft the six-field template. They open the FSM AI layer (Titan Intelligence, HCP AI, Workiz Genius, or the workspace bridge), paste in the role-context-task-format-constraint template, and run 5-10 dry-run voice memos against last week's known-good job-notes. Calibration takes 90-120 minutes of focused work. The constraint phrasing — "do not invent numeric measurements" — needs the most attention; phrasing variations alter how strictly the AI enforces the rule.

Weeks 1-2: pilot with 3-5 techs. Service manager reviews output at end-of-day for the first two weeks. Coaching points: voice-memo length (60-90 seconds — under 30 produces sparse notes, over 120 overwhelms the format constraint), microphone discipline (curbside in the truck cab, never inside the customer's house), the verify discipline (every note gets the 30-second pass). End of week 2: pilot techs are at 92-96% field-completeness and average curbside time per note is 2.5-3 minutes.

Weeks 3-6: full-truck rollout. The service manager runs a Tuesday huddle each week walking through 3-5 example notes (the great, the deflected, the verify-catch). The 4 p.m. dispatcher review surfaces any cases where the next-step field was missed or the recommendation field was vague. Weeks 7-9: integration with the Comfort Advisor desk. The Advisor reads the customer-mood and financing-posture fields off the tablet before walking the kitchen-table sit-down. Close rates lift 3-6 points where the financing-posture field is consistently captured.

Three failure modes consistently kill rollouts. First, the "tech writes long" failure: techs who treat the voice memo like a free-form essay produce 3-4 minute audio that overwhelms the format constraint and the AI returns paragraph-shaped notes. The fix is huddle practice and a 90-second timer in the FSM UI. Second, the "skip the verify" failure: techs who trust the AI output and save without the 30-second pass. The fix is monthly random audit of 5-10 notes per tech against the source voice memo. Third, the "no constraint update" failure: the prompt template never updates after Week 0, and as new equipment types appear, the AI drifts. The fix is monthly prompt-template review in the manager's first-Monday operations huddle, with explicit version-stamp so the floor knows what changed.

Five numbers track the rollout, and the service manager reads them every Monday morning. Field completeness: baseline 55-70%, target 92-96%. Below 88% by week 6 means the prompt template needs recalibration. Curbside time per note: baseline 11-14 minutes, target 2.5-3.5. Above 4 minutes, techs are over-talking. Dispatcher follow-up call count per day: baseline 14-22, target 3-6 at steady state — the single most visible owner-facing number. Next-step capture rate: baseline 30-45%, target 90%+. The leading indicator that predicts membership-conversion lift 4-8 weeks later. AI-caught-error rate: the number of fabricated readings the verify pass catches per week. Healthy: 1-3 per tech. The "AI caught a hallucination" wall from L1 accumulates every catch. Zero catches per tech per week means either the constraint is too aggressive or the verify is being skipped — both warrant a Tuesday huddle correction.

Key Takeaways

• The tech's job-note is the source-of-truth artifact for the next six revenue events tied to that customer. CSR, dispatcher, Comfort Advisor, service manager, recall touch, membership upgrade — all run downstream of the note. Structured notes are the highest-leverage tech artifact in the shop.
• The six-field structure is non-negotiable: Equipment, Condition, Recommendation, Customer Mood, Financing Posture, Next Step. Same order every time. 90-130 words total. Dispatcher reads it in 11 seconds; Comfort Advisor reads it off the tablet at the door tomorrow.
• The tech speaks 60-90 seconds of voice memo, not 12 minutes of thumb-typing. AI is the interface that lets the tech talk like a tech and produces notes the manager wants to read. Total curbside time: 2.5-3.5 minutes per note vs. 11-14 minute baseline.
• The 5-part prompt is preloaded as a template in the FSM AI layer — Titan Intelligence, HCP AI, Workiz Genius, Sera AI, FieldEdge Voice, or a Claude Projects / ChatGPT Team workspace as the bridge solution.
• The constraint phrase "do not invent numeric measurements" is the single most important line in the prompt. Fabricated capacitor readings, invented part numbers, and hallucinated APRs are the highest-cost AI-error classes in field-service.
• The 30-second verify habit applies to every note before save. Five checkpoints: numbers, names, parts, warranty/code, financing. The shop that skips it loses warranty relationships in quarterly distributor audits six months later.
• Dispatcher follow-up calls drop from 14-22 per day to 3-6 per day at 95%+ field completeness. Saves 70-110 minutes of dispatcher time per shift.
• Dispatcher productivity lifts 18-24% and average ticket lifts $90-$180 at 60-day steady state, because the next-step field is consistently captured and worked into the recall, membership, and replacement nurture loops.
• Three failure modes: tech writes too long, tech skips the verify, prompt template never updates. The Tuesday huddle and monthly template review are the operating cadence that holds the lift.