SecondTrip

1. One-liner

Challenges the part a technician is about to order, while he is still standing in the customer’s kitchen.

2. Trend signal — why now?

Three things moved, and they moved in the same direction.

The failure rate is measured and it is bad. Consumer Reports’ subscriber survey puts the odds of a major appliance being repaired right the first time at 58% overall — 63% for independent shops, 51% for factory-authorized ones. Field-service benchmarks put “best in class” at 88%. So the independent shop’s baseline is roughly 25 points below where a well-run service org lands. That gap is not a mystery workflow problem. It is mostly the wrong part showed up in the van.

The documentation the tech needs to avoid that is locked up — and the law just started prying it open. The PIRG / iFixit / Repair.org survey of 50 manufacturers and 37 technicians found 86% of appliance manufacturers do not publish service manuals, 89% of techs report difficulty locating manuals, and 93.5% struggle to find schematics. GE charged $919/year for diagnostic access that authorized servicers get free. That survey is from February 2023 — it establishes the problem is structural and old, not that it is new. What is new: Washington and Colorado right-to-repair laws took effect January 1, 2026, and Texas HB2963 takes effect September 1, 2026, each obliging manufacturers to supply parts, tools and documentation to independent repair providers on fair and reasonable terms. Texas additionally mandates documentation retention for 7 years on products over $100. The corpus an independent tech is legally entitled to is expanding this year, for the first time.

The workforce that used to carry this knowledge in its head is retiring. ServiceMag (March 31, 2026) reports ~37,300 US home appliance repairers, average age 43–44, with 45% of repair businesses facing technician shortages and 60% of applicants lacking formal training. Meanwhile a 2026 washer has a variable-frequency drive, hall-effect sensors, a control board running firmware, and Wi-Fi. The veteran who could diagnose by ear is leaving; the kid replacing him orders the control board because the control board is always the answer when you don’t know.

The economics of the resulting second trip are documented: $150–$220 all-in per second truck roll ($95 trip charge usually non-billable, $60–90 labor, plus the burned schedule slot). Industry first-visit fix sits at 60–65%; top operators hit 82–87%.

Provenance:

• Signal 1 (demand): Consumer Reports survey — appliances repaired right the first time only 58% of the time overall, 63% at independent shops — https://www.consumerreports.org/appliances/washing-machines/services-with-best-washer-repair-rates-a1178747914/ — observed 2026-08-02
• Signal 2 (feasibility): Right-to-repair statutes compelling manufacturer parts/tools/documentation to independent repairers — Washington and Colorado effective 2026-01-01, Texas HB2963 effective 2026-09-01 — https://www.morganlewis.com/pubs/2026/06/navigating-the-right-to-repair-landscape-in-2026-a-refresher-on-basics-and-best-practices and https://h2compliance.com/texas-right-to-repair/ — observed 2026-08-02
• Signal 3 (economic): Second truck roll costs $150–$220 all-in; moving a 4-tech shop from 65% to 82% first-visit fix is worth $150K–$225K/year; 37,453 US appliance repair businesses, $7.4B market, 78% single-owner — https://www.deelo.ai/blog/how-to-track-appliance-repair-parts-inventory-and-reduce-truck-rolls and https://www.ibisworld.com/united-states/industry/appliance-repair/1710/ — observed 2026-08-02
• Signal 4 (supporting): 86% of manufacturers don’t publish service manuals; 89% of techs can’t find manuals; 93.5% can’t find schematics; GE charged $919/yr for access — US PIRG / iFixit / Repair.org survey of 50 manufacturers and 37 technicians, February 2023 — https://www.ifixit.com/News/71502/most-appliance-manufacturers-wont-give-the-public-repair-information-pirg-and-ifixit-find — observed 2026-08-02 Category: Underserved niche

3. The opportunity

Everyone is building the wrong product for this market.

There is now a small pile of “AI diagnoses your appliance” tools. Jenova ships a free consumer-facing appliance diagnostic that reads error codes and photos. Repflow AI does the same thing with enterprise-only custom pricing aimed at OEM call centers and warranty networks. Both answer the question “what’s wrong with it?”

That is not the question that costs $200. The tech usually has a decent theory of what’s wrong. The question that costs $200 is “is the specific part number I am about to click ‘order’ on actually the right one, and will it actually fix this?” Those are different questions and they fail differently:

• Misdiagnosis — the part is genuinely the right part for the theory, but the theory is wrong. Fred’s Appliance Academy on why: “There is nothing that is more frustrating to a technician than getting that replacement part in, installing it, and finding that it did not fix the problem.” And the cause: “Ohms will lie. For this reason, you don’t just want to depend on your reader to make a diagnosis for you.”
• Part-number mismatch — the theory is right but the part is wrong for this model variant. ServiceWorks, bluntly: “Parts lists are different even between similar models. Thus, even without exact model matching, you risk ordering incompatible or useless parts.” Compounded by unreadable tags: “Hard-to-read model numbers make correct appliance model identification difficult… fuzzy, worn, or cryptic model/serial tags.”
• Supersession — the part number was right two years ago and has since been superseded, with the replacement valid only for certain serial ranges or production dates. Whether a supersession is unconditional or bounded by model revision is exactly the detail nobody checks in a customer’s kitchen.

So the gap: nobody sells a pre-order challenge step to the independent shop. Field-service platforms that do supersession alerts (item.com, Motility) are enterprise fleet systems. ServiceTitan is $245–398/tech/month plus a $5K–50K implementation fee — a rounding error to a 200-tech HVAC firm, and absurd for a two-van appliance shop. Housecall Pro at $59–299/mo does scheduling and invoicing, not parts adjudication.

The wedge is a product that assumes the tech is competent and acts as an adversary to his own conclusion for ninety seconds before money moves.

4. Target market

• Primary customer: Owner-operator of an independent appliance repair shop in the US running 1–10 technicians. The owner is usually still turning wrenches himself. He is the buyer, the user, and the person who eats the cost of a second trip.
• Why they buy: Because 78.16% of these businesses are single-owner operations and the owner personally absorbs every $150–220 mistake. He is not buying “AI.” He is buying “I stopped sending Dave back out.” The secondary buyer motive is training: with 60% of applicants lacking formal training, the owner needs his 6-month-old hire to order parts like a 10-year veteran, and he does not have time to check every order himself.
• Rough TAM reasoning: 37,453 US appliance repair businesses (IBISWorld 2026), $7.4B market. Say 60% are too small or too analog to buy software at all — that leaves ~15,000 realistic targets. At $99–299/mo, a 3% penetration of that realistic slice is roughly $7–13M ARR. I only need a fraction of that.
• Why now for them: Appliance complexity has outrun tribal knowledge exactly as the people holding that knowledge retire, and the documentation that would close the gap is only now becoming legally extractable.

5. Product sketch (MVP)

• Snap the tag. Photo of the model/serial plate — including worn, greasy, or half-scratched ones — resolves to an exact model and production variant. This is the load-bearing first step; every downstream answer depends on getting the variant right.
• Say the theory. Tech dictates one sentence: “F30 on a Whirlpool front loader, drum won’t spin, I’m calling it the drive motor.” Voice, one-handed, in a noisy laundry room.
• The challenge. The product argues back: the two or three other failures that produce this exact symptom set on this exact model, and the specific test that discriminates between them. Not a lecture — a question the tech can answer in 60 seconds with the meter already in his hand.
• Part verification. Confirms the part number is correct for this variant, flags supersessions, and states whether the supersession is unconditional or bounded by serial range / production date.
• Confidence call. A plain verdict before he clicks order: order it, or run one more test first. With the reasoning shown, because a tech who can’t see the reasoning won’t trust it twice.
• Callback ledger. Every call logged with the theory, the part ordered, and whether it fixed it. This is the shop owner’s view and it’s the thing that turns a tool into a system of record.
• Shop-level pattern report. Monthly: which techs have the worst part-hit rate, which models generate the most callbacks, which parts get ordered and returned.

6. AI angle — what’s load-bearing

Remove the AI and there is no product. Three jobs are genuinely model work:

1. Reading destroyed model tags. A rating plate behind a dryer drum after eight years is smeared, partially abraded, and photographed at an angle in bad light. This is the single highest-leverage step and it is pure vision. Deelo’s data claims disciplined model/serial capture alone moves first-visit fix 8–15 points in a quarter — that is the floor of what this feature is worth, before any reasoning happens.
2. Adversarial differential reasoning. Given a symptom set plus an error code plus a tech’s stated theory, generate the competing hypotheses and the cheapest discriminating test. This is the part a free error-code lookup app categorically cannot do — FixCode has 6,086 codes offline, and a code-to-text mapping tells you nothing about which of three components produced the code.
3. Reconciling messy parts data. Supersession chains, variant-specific parts lists, and manufacturer documentation are inconsistent across brands and formats. Normalizing that into “correct for your exact unit / not correct / correct but superseded under these conditions” is extraction and reconciliation at a scale no small team does by hand.

The honest caveat: the AI is doing retrieval and adjudication over a corpus, and the corpus is the hard part. See section 13.

7. Localization angle (if any)

N/A — this is a US-first play, deliberately. The wedge is bound to US right-to-repair statutes (WA, CO, TX and the state pipeline behind them) and US parts distribution (Marcone, Reliable Parts, encompass). The EU has adjacent repair rules and would be a later market with a different corpus and different distributors. Do not spread thin.

8. Business model — path to $1M–$5M ARR

• Pricing: $99/mo for a solo operator (up to 2 techs), $249/mo for a shop up to 6 techs, $449/mo up to 12. Priced per shop, not per seat — per-seat pricing punishes the customer for the exact behavior we want (every tech using it on every call).
• ACV: ~$2,400 blended.
• The pitch math the customer does in his head: one avoided second trip is $150–220. At $249/mo, the tool pays for itself if it prevents one and a half second-trips a month. A 4-tech shop is currently eating roughly four a day at industry-average fix rates. This is not a close call, and that is the entire sales conversation.
• Rough math to $1M ARR: ~420 shops at $2,400 ACV. Out of ~15,000 realistic targets, that’s under 3%.
• Rough math to $5M ARR: ~2,000 shops — 13% of the realistic slice, which is a stretch on this product alone. The credible route to $5M is adjacency: same buyer, same tag-scan, expanded to HVAC and commercial kitchen equipment, where the truck-roll costs are higher and the technician shortage is worse. Do not plan on it in year one.
• Expansion path: Seat tiers as the shop grows, then a parts-ordering integration where the verified part number pushes straight to the distributor cart. Long-term the interesting line is distributor-side: a distributor with a lower wrong-part return rate has a real P&L reason to subsidize this.

9. Go-to-market wedge — first 100 customers

• Appliantology / Master Samurai Tech. This is where working appliance techs actually congregate, and the forum is visibly full of exactly this pain — live threads like “Need access to parts for Samsung laundry tower WH53DBH700EFA3,” “Help finding/identifying sub zero harness connector plug,” “I need ID code for main control board.” Do not spam it. Answer 50 of these threads correctly and for free, with the product’s output as the visible artifact. The forum’s culture rewards competence and punishes marketing; earn the right to sell there over 8 weeks.
• The tag-scan free tool as the top of funnel. Ship model-tag-photo → exact model + parts list as a permanently free, no-signup web tool. Every tech has this problem daily, it is instantly demonstrable, and it seeds the corpus. Gate the differential-diagnosis challenge behind the paid tier.
• Scrape and call the shops directly. There are 37,453 of these businesses with public Google Business listings, phone numbers, and — critically — Google reviews. Filter for shops with 20+ reviews containing “had to come back,” “second visit,” or “wrong part.” Those owners already know they have this problem because customers wrote it down publicly. 500 targeted calls at a 10% demo rate at a 30% close is 15 customers; run it four times.
• Fred’s Appliance Academy and the trade schools. They train the incoming techs the industry is short of. A free shop-license for graduating students puts the tool in the hands of exactly the low-experience techs whose ordering accuracy is worst, and their employers are the buyers.
• Parts distributor route (later, not first). Marcone and Reliable Parts eat return-processing costs on wrong parts. This is a real partnership thesis but it’s a 6-month enterprise cycle — keep it out of the first 100.

10. Build complexity — justification

Medium, honestly leaning to the top of Medium. The app itself — photo capture, voice input, model resolution, a reasoning call, a log — is 8–10 weeks for a strong pair on off-the-shelf multimodal APIs. The work that makes it Medium and not Low is the corpus: assembling model→variant→parts→supersession coverage across the major brands, and building the ingestion path for tech sheets and service documentation. Expect 4–5 months to a version a professional tech will trust on brands beyond the top five. Start narrow — Whirlpool, LG, Samsung laundry and refrigeration only — and expand by brand. A product that is excellent on 40% of calls beats one that is mediocre on all of them.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Right-to-repair statutes expand legitimate access to documentation for independent repairers. Corpus must be built from legally obtained sources — see risk flags.
Ethical — no harm / dark patterns	✅	Reduces waste, extends appliance life, makes independent repair more competitive against manufacturer-controlled repair.
Market exists (evidence above)	✅	37,453 businesses, $7.4B, measured 58% first-time fix rate, documented $150–220 cost per failure.
1–5 person team can build this	✅	2–3 people, with a working appliance tech as a founder or first hire.
Launchable with <$50K / ₹40L	✅	Inference, data acquisition, and a paid domain advisor. Well under $50K to first revenue.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	16/20	Felt multiple times a week, costs measurable cash ($150–220 each), owner personally absorbs it. Not 18+ because shops have limped along with this forever — it’s a chronic bleed, not a fire.
Demand evidence	15	12/15	Strong quantified evidence of the problem (CR 58%, PIRG 89%/93.5%, truck-roll costs). Weaker direct evidence that shops will pay for this specific framing — no incumbent is selling pre-order verification to independents, which is upside and risk in equal measure.
Build feasibility	15	10/15	App is easy; the parts/supersession/tech-sheet corpus is the real work. 4–5 months to trustworthy coverage.
Distribution clarity	15	11/15	Named forum, named schools, a scrapeable list of 37,453 businesses filterable by public review text. Docked because appliance techs are famously slow to adopt software and the forum can’t be rushed.
Revenue mechanics	15	11/15	ROI math is trivially provable to the buyer. Docked because $99–249/mo against a shop owner who resists software subscriptions is a real objection, and $5M ARR needs vertical expansion.
Time to first revenue	10	7/10	Sellable within 6–8 weeks of a narrow-brand v1 to forum contacts; corpus depth gates broader rollout.
Defensibility	10	4/10	Honest score. Month 3: copyable. Month 12: the callback ledger — what was ordered, what actually fixed it — is a proprietary outcome dataset nobody else has, and that’s the only real moat here. A well-funded field-service incumbent could bolt this on.
Total	100	71/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy · domain-expertise-required

This idea fails without a working appliance technician in the founding team or as employee #1. The differential-diagnosis logic has to be right or techs will abandon it after two bad calls, and no amount of model quality substitutes for someone who has actually stood in the kitchen.

Key assumptions to validate (3–5)

1. Assumption: Wrong-part ordering — not scheduling or diagnosis-in-general — is the dominant driver of the ~40% first-visit failure rate. How to test: Get 15 shop owners to share 90 days of callback records; categorize root cause. If wrong-part is under 40% of callbacks, the wedge is aimed at the wrong thing and needs re-pointing.
2. Assumption: A shop owner will pay $249/mo for a tool that does not replace his field-service software. How to test: Pre-sell to 20 forum contacts at a founding rate before the corpus is built. Under 5 conversions kills the pricing model, not necessarily the idea.
3. Assumption: Model tags can be read reliably enough from real field photos. How to test: Collect 300 real tag photos from working techs — the ugly ones, not the clean ones — and measure exact-variant resolution accuracy. Below 85% and the whole stack is unreliable.
4. Assumption: Sufficient parts/supersession/documentation coverage is legally and practically obtainable for the top brands. How to test: Spend two weeks attempting to assemble complete coverage for Whirlpool and LG laundry only. If that’s a slog, brand expansion is a multi-year problem.
5. Assumption: Techs will actually stop and consult a tool before ordering, rather than ordering on instinct. How to test: Instrument the free tag-scan tool — measure what fraction of sessions continue past model resolution into a diagnosis question.

Risk flags

1. Data acquisition / legal risk. The corpus is the moat and also the liability. Right-to-repair laws expand access for independent repair providers — a software vendor aggregating and redistributing manufacturer documentation is a different actor with a different legal posture. Build the corpus from legitimately obtained and licensed sources; do not assume the statutes cover a redistribution business. This is the single biggest risk in the idea and it needs a lawyer before it needs a marketer.
2. Behavior change risk. The product only works if a tech pauses before ordering. Experienced techs are confident, often correctly, and confident people don’t ask for second opinions. The product may end up most valuable to exactly the junior techs who have the least authority to buy it.
3. Incumbent bolt-on. ServiceTitan, Housecall Pro, or a parts distributor could add “verify this part” to an existing product and reach the customer through a channel already installed. The 12-month answer is the outcome ledger; there is no 3-month answer.
4. Adoption pace. This trade adopts software slowly. Sales cycles will be longer than the SMB SaaS norm despite the obvious ROI, because the buyer is a man in a van who checks email at 9pm.
5. Free-tier commoditization. Consumer AI diagnostic tools are already free and improving. If a general model gets good enough at part-number verification without a specialized corpus, the differentiation evaporates.

14. Structured verdict

Score:                  71/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Technical pair with a working appliance technician as co-founder or first hire
Time to revenue:        6–8 weeks to first paying shop on a narrow-brand v1; 4–5 months to broad trust
Capital to launch:      $20–35K (inference, data acquisition, paid domain advisor)
Top 3 assumptions to validate first:
  1. Wrong-part ordering drives the majority of callbacks — audit 90 days of records from 15 shops
  2. Shop owners pay $249/mo for a non-replacement tool — pre-sell 20 forum contacts before building the corpus
  3. Real-world model tags resolve to exact variant ≥85% — test on 300 ugly field photos
Kill criteria:
  - Abandon if wrong-part/parts-mismatch is under 40% of categorized callbacks in the 15-shop audit
  - Abandon if fewer than 5 of 20 pre-sell targets convert at the founding rate
  - Abandon if exact-variant tag resolution stays below 85% on real field photos after two months of work
  - Abandon if counsel concludes the parts/documentation corpus cannot be assembled and redistributed lawfully at acceptable cost

15. Next step — 1-week validation sprint

• Day 1–2: Join Appliantology as a paying member. Pull every thread from the last 90 days about parts identification, wrong parts, and callbacks. Categorize by root cause. Produce a number: what share of documented parts problems would a pre-order verification step have caught? This is desk research and it is cheap.
• Day 3–4: Collect 300 real model-tag photos — post in the forum offering something of value in exchange, and ask the 15 shops from the outreach list. Run exact-variant resolution against them and measure accuracy on the ugly half specifically.
• Day 5: Call 20 shop owners from scraped Google listings filtered for reviews mentioning return visits. Pitch the pre-order challenge at $249/mo, founding rate $149, card on file today for access in 60 days.
• Decide go / no-go on: ≥5 of 20 taking a founding-rate card AND exact-variant tag resolution ≥85% on the ugly-photo half. Both must clear. One without the other means the idea is real but the product isn’t, or the product works and nobody’s buying — either way, don’t build yet.