CreditClaw

1. One-liner

Files every HVAC warranty claim before the 60-day window shuts, then fights the short-paid ones.

2. Trend signal — why now?

Three things moved at once.

The paperwork burden is now quantified by the people trying to sell against it. Continuum, a vendor selling warranty automation to distributors, publishes the numbers: “Contractors spend 4–5 hours on what should be a 1-hour repair, buried in administrative work.” And: “The average HVAC warranty claim takes 20+ minutes to process, not to mention the 30-45 day wait on vendor approvals.” And the killer: “Roughly 40% of claims contain errors or missing documentation, leading to rejections and weeks of delay.” That is a vendor describing my customer’s pain in order to sell to my customer’s supplier. Nobody is selling to the contractor.

The deadline is hard and unforgiving. Manufacturer policy language is explicit: “All claims for reimbursement must be received at the factory within 60 days from date of service to be eligible for credit, and all claims outside this time period will be void.” There is no appeal for a late claim. A claim not filed by day 61 is not a delayed claim — it is a destroyed asset. Registration-side analysis reports that “15-25% of denied warranty claims trace to registration failure” and that “once the 60 or 90 day window closes, no resubmission fixes it.”

Browser agents got good enough to drive portals that have no API. 2026 is the year computer-use agents crossed from demo to narrow production. The documented sweet spot is exactly this shape: “Supplier portals, government databases, insurance platforms, and logistics systems that have a UI but no API.” Skyvern 2.0 posts 85.85% on WebVoyager and is described as best-in-class specifically “for form-filling tasks… insurance forms, government portals, and procurement workflows where forms are complex and vary between sites.” Every HVAC manufacturer runs its own dealer portal. None of them will ever ship a contractor-facing API for a 6-truck shop.

The economic leg: contractors large enough to feel this hire a human for it. “HVAC Warranty Administrator” and “HVAC Warranty Claims Coordinator” are real, currently-posted job titles. Average Warranty Administrator pay is “$50,252 a year… approximately $24.16 an hour”; HVAC-specific coordinator roles run “$21.47” an hour. And the job description contains my product spec verbatim: “Review short-paid claims and follow up to secure full payment.” Somebody wrote that duty into a job posting because the money is real and it walks out the door without a person chasing it.

Provenance:

• Signal 1 (demand): Contractors spend 4–5 hours of admin on a 1-hour repair; ~40% of claims contain errors or missing documentation causing rejection — vendor-published, selling to distributors not contractors — https://resources.gocontinuum.ai/why-warranty-claims-decide-who-wins-hvac-distribution — 2026-08-28
• Signal 2 (feasibility): Computer-use/browser agents moved to production in 2026 for exactly the “portal with a UI but no API” case; Skyvern 2.0 at 85.85% WebVoyager, best for form-filling on complex varying sites — https://alkunmedia.de/en/blog/posts/computer-use-agents-enterprise.html and https://ztabs.co/blog/ai-browser-automation-2026 — 2026-08-28
• Signal 3 (economic): “HVAC Warranty Administrator” is a staffed role at ~$50,252/yr whose posted duties include “Review short-paid claims and follow up to secure full payment”; 125,000 US contractor firms, 73% under 20 employees — https://www.ziprecruiter.com/Salaries/Warranty-Administrator-Salary and https://schedulingkit.com/statistics/hvac-industry-statistics — 2026-08-28 Category: Workflow automation

3. The opportunity

Warranty money flows in one direction and the tooling was all built facing the other way.

Look at who exists. InsightPro sells “warranty and service management” — to “HVAC manufacturers, distributors, and dealer networks.” Continuum sells claim automation to distributors, pitching that better claim handling protects “9% of dealer revenue” and dealer loyalty. Ferguson built a single digital intake form — for claims submitted to Ferguson. JB Warranties / JB360 sells extended-warranty programs and registration workflow. PipelineOn and the ServiceTitan/Housecall ecosystem handle registration — the front of the lifecycle, because registration drives the upsell to a maintenance plan.

Every one of those products is paid for by someone who benefits when the claim is small, late, or never filed. The manufacturer keeps the credit. The distributor wants throughput. The FSM platform wants the renewal upsell. Nobody’s revenue depends on the contractor getting paid the full amount.

That is the entire opportunity, and it’s the classic buy-side/sell-side inversion: tooling follows the fee-payer. The contractor — a 4-to-20-person shop — is the party who eats the loss twice. Once when a claim is never filed because the tech’s paperwork was thin and the 60-day clock ran out. Again when the manufacturer pays “for a lower rate, sometimes for less time than it actually took based off of how long they think it should take,” and nobody in a 6-person shop has the hours to dispute a $180 short-pay.

The incumbent to beat isn’t a product. It’s a $50K/year human, or more commonly, the office manager doing it badly at 9pm — which is why 40% of claims go in with errors.

What an AI-first team does 10× better: a browser agent doesn’t forget the 60-day clock, doesn’t skip the serial number, and doesn’t decide a $180 short-pay is beneath disputing. The marginal cost of chasing a small claim goes to roughly zero, which changes which claims are worth filing at all.

4. Target market

• Primary customer: US residential/light-commercial HVAC contractors, 4–25 employees, 2–12 trucks, $700K–$6M annual revenue. Dealer for 1–3 brands (Carrier/Bryant, Trane/American Standard, Lennox, Goodman/Daikin/Amana, York, Rheem). Buyer is the owner or the office manager / service coordinator. Already running ServiceTitan, Housecall Pro, FieldEdge, Jobber, or Workiz. Not the PE-rolled 300-truck platforms — those have a staffed warranty department and a procurement cycle I don’t want.

• Why they buy: Because the claim window closes and the money is simply gone. A shop this size runs a few hundred in-warranty service calls a year; at a typical labor allowance per claim, the unfiled and short-paid tail is a five-figure annual leak that never appears on a P&L as a loss — it appears as revenue that was never recognized. The pain is felt weekly (the claim backlog) and acutely at month-end (the credit didn’t show up).

• Rough TAM reasoning: ~125,000 US HVAC contractor firms; 73% are under 20 employees and “more than 78% employ fewer than 10 people.” My band — big enough to have real claim volume, too small to staff a warranty administrator — is realistically 20,000–35,000 firms. At $300/mo that’s a $70M–$125M addressable line. I need well under 1% of it.

• Why now for them: Equipment mix has shifted to variable-speed / inverter / heat-pump systems with more electronics and more expensive in-warranty part failures, and the refrigerant transition churned the installed base. Higher-value claims, more of them, same two-person office. Meanwhile the tooling that arrived to help was aimed at their distributor.

5. Product sketch (MVP)

• Claim clock. Every completed service job on in-warranty equipment creates a claim with a countdown to the manufacturer’s filing deadline. Red at day 45. Nothing silently expires.
• Tech-side evidence capture. A 40-second mobile prompt at the truck: model/serial photo, failed-part photo, nameplate, symptom. The agent reads the plate and fills model and serial itself — the two fields that get claims kicked back.
• Auto-filed claims. A browser agent logs into each manufacturer’s dealer portal and submits the claim with the documentation attached. Human confirms before the irreversible click, per honest 2026 agent practice.
• Rejection triage. When a claim bounces, the agent reads the reason, tells the office in one sentence what’s missing, and refiles once the gap is closed — while the window is still open.
• Short-pay detector. Reconciles the credit actually received against the labor allowance claimed. Flags every underpayment and drafts the dispute with the job record attached.
• Recovery statement. One monthly page: claims filed, credits received, short-pays recovered, dollars still outstanding, and dollars lost to expiry. This is the renewal argument.
• Per-brand rulebook. Encoded filing deadlines, required fields, and documentation rules per manufacturer, updated as policies change.

6. AI angle — what’s load-bearing

Remove the AI and this collapses into a spreadsheet of reminders, which is what the office manager already has and already ignores.

Three places AI does the actual labor:

1. Portal operation. Each manufacturer’s dealer portal is a different form, behind a login, with no contractor-facing API and no intention of ever having one. A computer-use agent navigating those portals is the only path that doesn’t require me to negotiate integrations with Carrier — which I will never get as a two-person startup. This is precisely the documented production use case for 2026 browser agents.

2. Vision on the nameplate. Model and serial transcription errors are a leading cause of the 40% error rate. A photo → validated model/serial, checked against the brand’s format, kills a whole rejection category at the source.

3. Reading rejections and short-pays. Denial reasons and credit memos arrive as unstructured text and PDFs in inconsistent formats. Parsing “what did they actually pay vs. what did I claim, and why the gap” is language work, done hundreds of times a month, at a per-claim value too low to justify a human.

The honest limit: the agent must not file blind. A wrongly-filed warranty claim is a real problem with a manufacturer relationship, so v1 keeps a human confirming submission. That’s a feature, not a hedge — and it’s what “production-ready” actually means for agents right now.

7. Localization angle

N/A — this is a US play. The wedge is the US manufacturer dealer-portal landscape and US warranty policy (60-day filing windows, labor allowance schedules, brand-specific registration rules). The same shape exists in Canada and Australia, and the product ports later, but there is no language or payment-rail angle here. Geography is a market boundary, not a moat.

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

• Pricing: $249/mo base (up to 40 claims/mo) → $449/mo (unlimited claims, multi-brand, short-pay disputes). Plus an optional 15% success fee on recovered short-pays, which is how I get in the door before they trust the base subscription.
• ACV: ~$3,600 blended, plus $400–1,200 of success fees on recoveries in year one.
• Rough math to $1M ARR: 280 contractors × $299/mo × 12 = $1.0M. That’s roughly 1% of my 20K–35K firm band.
• Rough math to $5M ARR: ~1,150 contractors at a $360 blended ACV, which needs the product to work for the 25–75 employee tier too, plus at least one distributor or buying-group channel deal carrying a few hundred accounts. Realistically a year-3 number, and it requires plumbing and electrical adjacency (same portal problem, same deadlines).
• Expansion path: Per-truck or per-claim-volume tiering as shops grow; adjacent trades (plumbing water heaters, appliance, electrical panels) reuse the same engine; then the parts-return workflow, which sits right next to the claim and leaks money the same way.

The pricing sanity check: at $299/mo I’m asking for $3,600/yr against a $50,252/yr warranty administrator and against a five-figure annual leak. That’s an easy conversation — which is exactly why I’d rather sell it as ROI than as software.

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

• The free leak audit — this is the whole wedge. Offer: give me read access to your last 12 months of completed jobs on in-warranty equipment, and I’ll tell you free how many claims were never filed and how much you were short-paid. This is a deterministic report, not a pitch, and it produces a specific dollar number for that specific shop. A contractor who is shown “$14,300 you didn’t collect last year” converts on the spot. Run it manually for the first 40 accounts.
• Success-fee entry. Lead with “I’ll recover your open short-pays for 15% and you pay nothing else.” Zero-risk entry kills the software-evaluation reflex, and the recovery statement converts them to subscription in month two.
• ACCA chapters and state HVAC associations. Local chapters run monthly meetings and are hungry for “how to stop losing money” content. Speak at 10 chapters in year one. These rooms are 30–80 owner-operators — precisely my buyer, in person, with a peer-recommendation dynamic.
• Manufacturer dealer meetings and buying groups. Brand dealer councils and groups like Blue Hawk / AD gather hundreds of my customers annually. Not a partnership — a booth and a leak audit.
• The FSM ecosystem back door. ServiceTitan/Housecall Pro have large Facebook user groups and marketplaces where the standing complaint is that warranty tracking is a field, not a workflow. Show up as the thing that finishes the job their FSM starts.

Bluntly: this is a sales-heavy business with a technical core. The audit report is the salesperson.

10. Build complexity — justification

Medium. The FSM integrations (ServiceTitan, Housecall Pro, Jobber APIs) and the vision-on-nameplate piece are off-the-shelf. The hard part is the browser-agent fleet against a handful of manufacturer portals that change without notice, plus the per-brand rules encoding — that’s grinding, not research. Realistic v1: 12–16 weeks for two people, covering two manufacturers and one FSM integration deeply rather than everything shallowly. The manual-service version of this (humans filing claims behind a product facade) can be running in 3 weeks and should be, because it teaches the rules before I automate them.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Filing claims as the contractor’s authorized agent, with their credentials and consent. Standard BPO/delegated-access arrangement.
Ethical — no harm / dark patterns	✅	Recovers money the contractor is contractually owed. No inflated claims — accuracy is the product, and fraud would destroy the manufacturer relationship.
Market exists (evidence above)	✅	Staffed job roles, vendor-published pain metrics, 125K firms.
1–5 person team can build this	✅	Two people, 12–16 weeks, off-the-shelf agent tooling.
Launchable with <$50K	✅	Well under. Main cost is founder time plus inference.

All five pass.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	16/20	Real recurring money loss, felt weekly, with a hard deadline that destroys the asset. Not quite hair-on-fire because it’s an invisible loss — unrecognized revenue, not a bill that arrives. Contractors tolerate it precisely because it never shows up as a line item, which is also the main sales obstacle.
Demand evidence	15	11/15	Strong indirect evidence: staffed warranty-administrator roles with “recover short-paid claims” in the JD, vendor-published error/time statistics, adjacent products funded and selling. Docked because I could not obtain direct verbatim contractor complaints — Reddit and HVAC-Talk were inaccessible during research. The demand is inferred from labor markets and vendor positioning, not heard from customers’ mouths. That gap is real and it’s the first thing to close.
Build feasibility	15	10/15	Not a weekend build. Browser agents on portals that change, multi-brand rule encoding, FSM integrations. 12–16 weeks for a pair, with ongoing maintenance drag as portals shift.
Distribution clarity	15	12/15	The free leak audit is a genuinely strong, specific, dollar-denominated wedge with named channels (ACCA chapters, dealer meetings, FSM groups). Docked because it’s high-touch — this doesn’t self-serve, and 100 customers is a quarter of selling, not a two-week sprint.
Revenue mechanics	15	12/15	Pricing is anchored against a $50K salary and a measurable leak; 280 customers for $1M ARR is credible. Success-fee model de-risks entry. Docked because ARPU depends on claim volume I haven’t validated per-shop, and the $5M path needs adjacent trades.
Time to first revenue	10	8/10	Success-fee recoveries can produce cash inside 6–8 weeks, before the product is finished, because the manual version works.
Defensibility	10	5/10	Honest score. The per-brand rulebook and portal-automation maintenance is a real grind that compounds over 12 months, and the claim history creates mild lock-in. But there’s no structural moat — a funded competitor or an FSM platform deciding to own this can copy it. Execution-plus-focus moat, not a durable one.
Total	100	74/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy · sales-heavy

Needs someone who can keep a browser-agent fleet alive against hostile-by-neglect portals, and someone who can sit in an ACCA chapter meeting and talk to a 55-year-old shop owner without sounding like a SaaS founder. That second half is the harder hire.

Key assumptions to validate (3–5)

1. Assumption: A typical 6-truck shop leaks $8K–$20K/yr in unfiled and short-paid warranty claims. How to test: Run the free leak audit manually on 10 shops’ 12-month job history. If the median is under $5K, the pricing collapses and this becomes a $99/mo utility, not a business.
2. Assumption: Contractors will hand over dealer-portal credentials to a third party. How to test: Ask 20 owners directly, early, before building. If they balk, the product has to become a co-pilot that drafts and hands off — different product, weaker value.
3. Assumption: Browser agents hold up across 3–5 manufacturer portals with acceptable maintenance drag. How to test: Build against two brands and measure successful-submission rate and breakage frequency over 8 weeks. Under 85% unattended success means the unit economics need a human in the loop permanently.
4. Assumption: Manufacturers tolerate agent-driven submission by an authorized dealer. How to test: Read dealer portal terms of use for the top 5 brands, and ask two friendly distributors what happens if a dealer automates submissions.

Risk flags

1. Platform dependency (severe): The entire product runs through portals owned by companies with no incentive to help me, who can add a CAPTCHA, change terms, or block automated access on any given Tuesday. This is the single biggest risk and it has no clean mitigation beyond diversifying brands and keeping a human-assisted fallback.
2. Channel conflict: ServiceTitan or a distributor group could build this as a feature. ServiceTitan already exposes warranty fields; the gap is workflow, not data. A 12-month head start and trade credibility is the only defense.
3. Credential and liability handling: Holding dealer portal credentials and filing on a contractor’s behalf means a mis-filed claim is my fault in their eyes. Needs a tight human-confirmation step and clear terms, which caps how automated v1 can honestly be.
4. Invisible-loss sales friction: Money never collected doesn’t hurt like money spent. Some owners will look at a $14K leak report, nod, and not buy — because it was never in their bank account. The success-fee entry exists specifically to route around this, and if it doesn’t work, distribution scores drop hard.

14. Structured verdict

Score:                  74/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Technical founder who can run browser agents in production, paired with
                        someone credible in the trades. Prior field-service or contractor
                        back-office exposure is worth more than HVAC knowledge.
Time to revenue:        6–10 weeks via success-fee recoveries; 4–5 months to subscription scale
Capital to launch:      $15–25K (mostly inference + founder time)
Top 3 assumptions to validate first:
  1. Median annual leak per 6-truck shop is $8K+ — manual audit of 10 shops' job history
  2. Owners will delegate dealer-portal credentials — direct ask to 20 owners before building
  3. Unattended portal submission holds ≥85% across 2 brands over 8 weeks
Kill criteria:
  - Abandon if median measured leak across 10 audited shops is under $5,000/yr
  - Abandon if fewer than 6 of 20 owners will delegate portal credentials or an equivalent
    authorized-agent arrangement
  - Abandon if two or more major manufacturers explicitly prohibit or technically block
    agent-driven submission by dealers
  - Abandon if ServiceTitan ships native multi-brand claim filing before v1

15. Next step — 1-week validation sprint

• Day 1–2: Recruit 10 HVAC shops (4–15 trucks) through two ACCA chapter contacts. Offer the free leak audit in exchange for 12 months of completed-job data on in-warranty equipment.
• Day 3–4: Do the audit by hand. For each shop, count in-warranty service calls where no claim was filed, and reconcile filed claims against credits received. Produce the dollar number per shop. In parallel, ask each owner the credentials question flat out and read the dealer terms of use for Carrier, Trane, Lennox, Goodman, and Rheem.
• Day 5: Decide.

Falsifiable go/no-go: Go only if the median measured annual leak across the 10 shops is ≥$8,000, AND ≥6 of 10 owners verbally commit to a 15% success-fee recovery engagement on the spot when shown their own number. Anything less and this is a utility, not a business — and I’d rather know in a week than in six months.