Prepared for Keith · Northeastern Appliance Service · July 2026
On July 14 you told us exactly where the money goes — repeat visits, ten callback slots a day, far jobs pushed three days out — and exactly how you want it fixed: a real person on the phone, asking the right questions, so the tech rolls once with the right part. Play #1 below is that fix — and underneath it runs something we've been developing in our own R&D: a private brain for your shop that starts from thirty years of your job history and gets smarter with every job you close. The other six plays are the extras, added around the fix to make it stronger. Every claim fact-checked, every security question answered.
owner conversations studied line-by-line (June 15 + July 14)
industry & vendor sources pulled for the opportunity research
claims adversarially fact-checked — 15 kept, 10 rejected & never used
systems in your stack mapped: Walkabout, ServiceBench, ServicePower, Service Central, ADP
product already live: the Kairos KPI dashboard we built for you
Play #1 · the problem you asked us to solve
"I want to make sure I only have to go to the house one time — the more times I go, I'm losing money."
— Keith, July 14 call
That sentence is the product. When a customer calls, your CSR keeps talking like always — and a quiet panel beside them fills with exactly what thirty years of your own jobs know about this model with this complaint: the questions to ask, the parts that actually fixed it last time, and a link to text the customer for a photo. The tech rolls once, with the right part on the truck.
Because it's yours. You called this the biggest leakage in your kind of company — so it goes first, and everything else on this page is an extra we add around it. It reads the registry (#2), stages the parts (#4), and moves the number the scoreboard (#3) measures, at the only moment that number can be moved: while the customer is still on the phone. And it absorbs #7's goal without a voicebot.
Nobody can diagnose a sealed system over the phone — that takes gauges on site, and we won't pretend otherwise. The win isn't remote diagnosis; it's staging the likely parts as a hedge and routing the call right: the EPA-certified tech, the longer slot, the recovery gear already on the truck. That kills the second trip even when the diagnosis waits for the driveway.
You said it yourself: every premier partner leaks money the same way. Proven here first — on your data, with your numbers — this is sellable across the Factory Certified network, with you as the reference customer. That's the long game; it never blocks the near-term build.
We prefilled the verified industry floor — $150–$300 per wasted trip — and a guess at your volume. Your drag replaces our guess. This is arithmetic, not a promise: trips × cost × 52 weeks.
Per-trip cost is the verified industry floor; Westchester likely runs higher. We quote no savings until this is computed from your real jobs.
Prototype commitment: a working version on your top 3–5 Sub-Zero models — call in and try to break it. We measure one thing: did the tech roll with the right part.
Start with the prototypeProprietary R&D · the engine underneath
This is our own research and development, built specifically for Northeastern — not generic AI with your logo on it. It starts from thirty years of your jobs and warranty claims, and then every job you close teaches it something new about how your business actually runs.
When your tech confirms what was really broken and which part fixed it, that answer feeds straight back in. The suggestion on the next call is sharper than the one it made yesterday — expect a visible difference over its first two to three months, and it never stops improving.
Your models, your failure patterns, your parts, your customers, your seasons. Hand the same software to another shop and it would grow into something different — because it learns the business it lives in instead of arriving pre-programmed with someone else's.
One isolated brain per business. It never shares what it knows, never trains anyone else's AI, and the longer it runs the more valuable — and harder to copy — it becomes. Kept access-controlled, that accumulated knowledge is legally protectable as a trade secret.
Today it powers the intake call. Where we're taking it: ask your business anything in plain English — "how did the team do on callbacks last month?", "which appliances hit the pay-for-labor window next quarter?" — and it answers from your own data, then flags the things you didn't think to ask.
Already live · the first piece delivered
Before any of this pitch existed, we built and shipped your KPI dashboard. It's live today, behind a login, holding your team's numbers. Everything on this page bolts onto it — it's the chassis, and it's proof we ship.
Jobs, revenue, callbacks, trends, and team leaderboards, per tech and per period — the employee-KPI sheet you built by hand, turned into a living dashboard with exports when you need paper.
The fairness detail your techs will care about: callbacks split into caused vs. ran — the tech who caused the repeat visit carries it, not the one you sent to clean it up.
Performance and pay finally sit side by side. The next update adds your raise thresholds: you set the bar — revenue, callbacks, whatever you weight — and the dashboard shows exactly who cleared it and who didn't, so raise conversations start from the same numbers every time.
Pay data is compartmentalized the way your body-shop project did it: role-based access, techs never see each other's money, and the full picture is visible to you alone.
The registry (#2), first-time-fix scoreboard (#3), and truck stock (#4) all live inside this dashboard as new views — not new logins. It's where the co-pilot's results show up as a moving number, and it's one of the brain's first data feeds.
When you eventually ask "how did the team do on callbacks last month?" in plain English, this is the data the brain answers from.
Due diligence
Before proposing anything, we researched the systems you actually run on, read their documentation and knowledge bases, verified how the money flows through them, and checked the legal ground under every idea. Here is your stack as we found it.
Your operational system of record: every call, job, schedule, part used, and payment lives here. On call-close it auto-generates warranty claims and pushes them out via FTP batch.
Key finding, from Walkabout's own knowledge base: claim data is pushed one way only — nothing about status or payment ever comes back. That gap is a revenue leak, and closing it needs zero cooperation from Walkabout.
Where your Sub-Zero warranty claims land. You used to key these in by hand; Walkabout now files them automatically — which changed our recommendation from "build a submitter" to "watch the money."
Hidden asset: every approved claim names the failed part. Your claims history is a clean, self-labeling record of what actually breaks on each model — training data nobody else has.
The second network your claims flow through. We went through its ClaimWorks documentation, its claim-status retrieval interface, and its full reject-code taxonomy.
The network itself flags underpayment: reject codes 102 and 240 literally mean "requested amount much different than payable." Reconciliation is partly just reading what it already tells you.
Sub-Zero's portal — every schematic, parts manual, and service manual, up to ~500 pages per model, accessed through your Factory Certified relationship.
We researched exactly what may and may not be done with this content (copyright and your certification contract), so nothing we build ever puts your certification at risk. Details in Security below.
Verified against Sub-Zero's published warranty and corroborated with repair-industry sources — the tier table that drives your whole customer economics.
Years 6–12: sealed-system parts are free, labor is not — and the homeowner must use a Factory Certified shop. That is the single most valuable fact in this document.
Payroll likely via ADP (we'll confirm). And we studied the body-shop project you told us about — Square data pulled automatically, sanitized to "pure data," per-employee keys.
That project is our reference bar for confidentiality: your people's pay data and your customers' details get the same sanitized, compartmentalized treatment.
The insight that reframes everything
Sub-Zero's warranty is tiered. In years 6–12, a failing sealed system still gets free parts — but the homeowner pays all labor, and the work must go to a Factory Certified shop. That's you. Sealed-system labor runs $2,000–$3,000 (compressors $500–$1,500). Warranty work isn't just an obligation — it's your pipeline of future high-ticket cash jobs, if you track every appliance you touch and reach out before it fails.
| Appliance age | Parts | Labor | Who pays labor |
|---|---|---|---|
| Years 1–2 | Covered | Covered | Manufacturer |
| Years 1–5 (sealed system) | Covered | Covered | Manufacturer |
| Years 6–12 (sealed system) | Covered | Not covered | The homeowner — COD, certified shop only |
Verified against Sub-Zero & Wolf's published warranty (subzero-wolf.com/warranty), corroborated by independent repair-industry sources. Dollar ranges are cross-corroborated industry figures — we'll calibrate to your actual price book before quoting ROI.
The extras · plays 2–7
Play #1 is what you asked for. These six are what we're adding on top — everything else our research surfaced, ranked by impact × feasibility, not by hype. Three extend the dashboard you already have; three sit alongside it. Each one either feeds the co-pilot or cashes in on what it learns, and every card links to its full breakdown.
A registry of every customer and appliance, with a warranty-tier clock that tells you when each one enters the pay-for-labor window — so you call them before the compressor fails.
Full breakdown ↓Make repeat visits a measured number per tech and per reason. Best shops fix it first trip 85–90% of the time; the industry average is ~75%. Know where you stand.
Full breakdown ↓Roughly 20 part numbers cover 70–80% of repairs in a category. Your callback reasons tell us which ones belong on each truck — and which to pre-order on diagnosis.
Full breakdown ↓Walkabout files claims but never hears back. Rejected, underpaid, and stale claims die quietly. We reconcile every claim against what you were actually paid.
Full breakdown ↓Affluent homeowners check stars before letting anyone near a $15k kitchen. A post-job review request — sent only with your sign-off — makes reviews a system, not luck.
Full breakdown ↓AI phone answering was the weakest-evidenced idea before you told us the deciding fact: your customers want a real person. So #7 folds into the fix (#1) — no voicebot.
Full breakdown ↓How they fit together
The co-pilot (#1) is the play on the field: it changes the number at the moment it's created — the intake call. Plays 2–4 are the scoreboard and supply line around it. Plays 5 and 6 protect and grow the money. Play 7 folded into the fix. And every closed job feeds the loop that makes the whole thing smarter.
Touch it: hover or tap any box to light up what it feeds — click through to its breakdown.
The breakdowns
Every number below survived adversarial fact-checking. Where the research couldn't verify something, we say so out loud instead of rounding up.
Your certification obligates you to warranty work — and quietly hands you a compounding book of future high-ticket business. Every appliance you service under warranty is a customer who, in years 6–12, must come to a Factory Certified shop and must pay labor out of pocket. Today that future revenue lives in Walkabout's history and nobody is watching the clock.
Verified: sealed-system jobs run $2,000–$3,000 in labor (compressors $500–$1,500). The standard, verified conversion tactic — a diagnostic fee credited toward the repair when work is approved same-visit — is how the best shops turn the first call into the full job.
You told us the pain yourself: a repeat visit to a far job gets rescheduled three days out, and ten callback slots a day are ten new customers you couldn't serve. This play makes that loss a measured number — per tech, per failure reason — instead of a feeling.
Verified (Aberdeen Group): best-in-class shops complete 85–90% of jobs on the first visit vs. a ~75% industry average, with callback rates under 15% (best under 10%). Each avoidable repeat trip costs $150–$300 as a conservative industry floor — and a built-in Sub-Zero an hour away in Westchester is worse than the floor.
Your July call opened on exactly this: a truck can't carry a thousand parts, so it should carry the ones that actually get used. Instead of gut feel, we join your callback reasons and parts usage to Sub-Zero/Wolf/Cove-specific parts and recommend what each truck carries.
Verified practitioner rules: roughly 20 part numbers cover 70–80% of repairs in a category, and a sound par level is average daily usage × 3 days of supply. Your data decides which 20.
Our research corrected our own first idea here. You don't need a claim submitter — Walkabout already auto-files to ServiceBench and ServicePower on call-close. What nobody does is listen for the answer: Walkabout's own documentation says data goes one way only. Status and payment never come back. So rejected claims, underpaid claims, and stale claims age silently in a 30-day bucket until someone keys them in by hand.
The three dollars this recovers: underpayments (filed $520, paid $340 — caught and appealed), rejections that were never re-filed (money earned, then abandoned), and recurring leak patterns ("these job types always short-pay on labor rate"). ServicePower's own reject codes 102/240 explicitly flag "requested amount much different than payable" — part of this is just reading what the network already says.
Your work is premium; your star rating doesn't say so. For a homeowner choosing who to let into a $15k built-in kitchen, that gap costs bookings. Happy customers rarely review unprompted — the shops that look as good online as they are in person ask, every time, automatically.
Verified mechanism: post-job review requests by SMS + email, triggered on job-complete or paid, are a proven, standard pattern (confirmed against Housecall Pro's documented flow). We deliberately do not promise a review count — the one magnitude claim we found failed fact-checking, so it's not in this pitch.
The original idea was AI phone answering — a missed call at a high-ticket shop is a lost $2,000 job. Directionally true. But we're telling you plainly: this was the weakest-evidenced opportunity in our research. The one vendor case study with a concrete booking-rate number failed our fact-check, so we won't put numbers on it. And then you settled it better than the data could:
Your words: customers call Northeastern because a real person answers. Your clientele — affluent, often older — prefers humans by roughly three to one in the published research, and regulators are moving toward mandatory "this is an AI" disclosure on synthetic voice calls. An answering bot is the wrong tool for a premium brand built on trust.
Security & safety
We researched the legal cases, the regulations, and the failure stories — before writing a line of code. These aren't reassurances; they're design decisions already made. Tap any question.
You do — all of it, always. Your job history, customer list, claims, spreadsheets, photos: they're Northeastern's property before, during, and after any work we do. We build on your data; we never acquire rights over it.
It goes further: your curated job history is legally protectable as a trade secret as long as access to it is controlled. Locking it down isn't just safety — it's what keeps your thirty-year advantage yours. Part of our job is keeping it that way.
Encrypted in transit (between you and the system) and at rest (where it's stored). Every screen sits behind a login, and row-level security in the database means each account can only reach the rows it's entitled to — the same architecture already running in your live KPI dashboard.
One deliberate honesty note: we don't advertise "end-to-end encryption," because for a system that has to compute on your data, that label would be marketing, not truth. We'd rather be precise than impressive — that's the standard for every claim on this page.
No. No third-party model training, period. Your instance is yours alone — one dedicated, isolated deployment per client, so nothing about Northeastern's operations can ever surface in anyone else's tool. If this is ever offered to other shops, each gets its own sealed instance the same way; data never crosses between tenants.
No — your rule, our design constraint. We actually researched AI voice thoroughly rather than dismissing it, and the research agreed with you: your clientele prefers a human on the phone by roughly three to one, regulators are moving toward mandatory AI-disclosure on synthetic voice calls, and a botched robot call costs exactly the kind of relationship your brand runs on. Humans answer. The AI is the quiet panel beside your CSR — it never speaks, and it never contacts a customer on its own.
No message goes out without a human decision. Anything outbound — a review request, a photo-upload link — follows draft-before-send: the system prepares it, a person approves it. Features that message customers ship switched off and stay off until you've approved the exact wording and timing. Opt-outs are honored permanently, and the whole flow is built TCPA-aware (the law governing texts to consumers).
This is the question we researched hardest, because your Factory Certified relationship is the business. The short version: having access to Service Central is not the same as having the right to copy it into software. Two separate gates apply — copyright law, and your certification agreement — and recent court cases (including a $1.5 billion settlement over exactly this pattern) show what happens when tools ignore that line.
So we designed around it: version one doesn't touch the manuals at all. Your own job history carries it. If you later want manual content in-scope, it happens the safe way — your copy, inside your isolated instance, citing sources rather than reproducing pages, and only after your agreement allows it or your Sub-Zero Field Service Manager gives narrow written permission. That ask is yours to make (we'll prepare it with you), it's a small ask designed to get a yes, and nothing we ship ever depends on it.
Not by design — the roadmap is read-before-write. The claims work starts strictly read-only: pulling status and payment data, which cannot alter a claim or touch the account your revenue depends on. Only after that has run cleanly — and only with your explicit go — would assisted filing exist, and then exclusively through official channels with a person approving every single submission. Never scripted portal tricks, never silent automation. Full audit log, hard kill switch.
Role-based access: each person sees what their job needs and nothing more — a tech doesn't browse company financials; pay data isn't visible to other employees. Sensitive actions are logged, so there's always an answer to "who did what, when."
You've seen this standard before: the body-shop project you described — payment data pulled automatically, sanitized to pure numbers, separated per employee. That's our reference bar here too.
It will be sometimes — so the system is built assuming it. Three protections: assist, not authority — the AI suggests, your CSR and tech decide, on every single call. It knows when to shut up — when past jobs don't give a strong signal, it says "not enough signal — human call" instead of guessing, and anything smelling like a sealed-system fault is automatically deferred to a tech with gauges. It keeps score on itself — every suggestion is logged and checked against what the tech actually found, so accuracy is a number you can see, not a promise you have to take.
Clean exit, contractual: you get a complete export of your data in standard formats, and your instance — including anything ever derived from your documents — is deleted. Your data walks out with you because it never stopped being yours.
No. Nothing here moves money, files taxes, or touches bank credentials. The claims work reads payment records to reconcile them; paying, appealing, and banking stay with your people. Payroll data, if it's ever in scope for KPIs, gets the sanitized per-employee treatment described above.
Because we showed you the ones we threw away. Every claim in this document went through adversarial verification — 10 of 25 didn't survive, and they appear on this page only crossed out (next section). Where a figure is an industry range rather than your actual number, we say so, and the first real step is replacing ranges with your real numbers before any ROI gets quoted. A pitch that never says "we don't know yet" is the one to distrust.
Straight answers
Ten claims from our research failed adversarial fact-checking. A normal pitch would have led with them — they're the impressive-sounding ones. They appear here only as what they are: rejected.
Rejected in verification — used nowhere in this pitch:
Also flagged: published dollar ranges skew to cheaper markets — Westchester/NYC-metro costs likely run higher, which makes the callback math more compelling for you, not less. We still won't quote it until it's computed from your jobs.
Next steps
We build the pre-screen co-pilot on your top 3–5 Sub-Zero models — intake screen, your-history lookup, question list, parts shortlist, photo link. You call in and try to break it. Live feedback, fast iterations, one success metric: the tech rolls once with the right part.
Exports, not integrations, to start: Walkabout job history, a claims export, your callback and revenue spreadsheets. Your data stays yours (see Security) — we just need it flowing so the co-pilot (#1) and the registry (#2) have fuel.
Ten minutes of answers replaces every industry range in this pitch with your actuals:
The registry (#2) and the first-time-fix scoreboard (#3) come online with the co-pilot, since they feed it. Truck stock (#4), the claim watchdog (#5), and reviews (#6) follow in trust order — each one switched on only with your sign-off.