What changed
Schema-defined document extraction became a single cheap API call (FACT: Context.dev launch, id 940), collapsing the cost of the historical-records review that EPA accepts for classifying unknown service lines. Simultaneously, EPA finalized the Lead and Copper Rule Improvements (FACT: Federal Register 2024-23549), which locks in a nationwide obligation for drinking water systems to inventory, validate, and replace lead and certain galvanized service lines.
Why now
LCRI is final and its compliance clock runs into 2027 and beyond: systems must maintain and validate service line inventories, resolve 'unknown' classifications, and document replacement over a roughly 10-year program (FACT for the rule's existence and requirements; specific per-state deadlines are set by primacy agencies). Raw extraction is commoditizing fast, so the defensible window is the next ~6-12 months to accumulate a state-acceptance track record and validated template mappings (HYPOTHESIS on window length).
Converging signals
(1) One-call structured extraction from arbitrary documents (https://www.context.dev); (2) LCRI final rule forcing every affected community water system to inventory/replace lead lines (https://www.federalregister.gov/documents/2024/10/30/2024-23549/national-primary-drinking-water-regulations-for-lead-and-copper-improvements-lcri); (3) EPA-funded technical assistance pipeline to rural/small systems (https://www.grants.gov/search-results-detail/362798). Caveat (FACT from the grant title): that TA grant targets WASTEWATER systems — it proves the funded-intermediary channel exists, but the drinking-water analogues (RCAP, state rural water associations, EPA WaterTA/Get the Lead Out) are the actual channel and must be verified.
Customer pain
Tens of thousands of small/rural water systems have 'unknown' service line classifications that LCRI forces them to resolve. The EPA-accepted cheap path is historical records review — but the records are decades of scanned tap cards, plumbing permits, and meter cards, and small systems have no staff to review them. The alternatives are paying consultants per line or defaulting lines to 'lead status unknown,' which triggers replacement-program and public-notification burdens (FACT that unknowns carry ongoing obligations under LCRI; the $5–15/line manual baseline is stated in the convergence input and is HYPOTHESIS until quoted from a real engagement).
Who pays
Two buyers: (a) EPA/state-funded technical-assistance intermediaries who are paid under grant deliverables to help exactly these systems and can embed the tool in their workflow; (b) small utilities directly, per-line or per-project, often reimbursable with DWSRF/BIL lead service line funding (HYPOTHESIS on reimbursability per state — verify with 1-2 state primacy agencies). The beneficiary (the utility) and the buyer (often the TA provider or engineering consultant) are distinct; both are reachable without enterprise procurement.
Solved today
Manual record review by utility staff or engineering consultants billing per line or per hour; predictive-modeling vendors (BlueConduit) for larger systems; inventory-management SaaS (120Water, Trinnex leadCAST) that stores classifications but does not cheaply extract them from scanned paper records. Many small systems simply filed inventories full of 'unknown.'
Why current solutions are bad
Manual review is slow and expensive per line; predictive models are statistical (state acceptance for records-based classification is stronger); incumbent SaaS assumes the data is already structured. Nobody serves the 'we have 40 boxes of tap cards' problem at a small-system price point with per-row evidence citations that survive a primacy-agency audit.
Proposed product
A document pipeline: batch OCR of tap cards/permits/meter records → schema-defined extraction (material, install date, address, evidence snippet, confidence) → mapped into the state's LCRI inventory template → confidence-ranked field-verification worklist for the residual low-confidence lines. The per-row evidence citation is the audit-defensibility wedge. Sold per line processed or per utility project, white-labeled to TA providers and small-utility engineering consultants.
MVP version
Get scanned record sets from 1-3 small utilities (via a rural water association contact or a public-records request), build the OCR+extraction+template pipeline against one state's inventory template, and produce a completed inventory with evidence citations in under a week of wall-clock time. The demo IS the sales asset.
30-day build
Week 1-2: build pipeline on sample records; measure accuracy vs a hand-reviewed gold set of a few hundred cards; produce cost-per-line and confidence-distribution numbers. Week 3-4: run the kill test — pitch 5 TA providers/state rural water associations and 3 small-utility engineering firms a paid pilot; simultaneously email 2 state primacy agencies asking whether AI-assisted records review with per-row evidence citations is acceptable documentation (their answer is the product's regulatory foundation).
60-day build
Convert 1-2 paid pilots ($2.5k-10k per utility project); iterate template mappings for 2-3 states; publish an accuracy/methodology one-pager aimed at primacy-agency reviewers; get a TA provider to include the tool in a grant deliverable.
90-day revenue plan
3-5 utility projects at per-line or flat-project pricing, or one TA-provider white-label agreement covering multiple member systems. Target $10k-30k cumulative by day 90-120 (HYPOTHESIS; depends on pilot conversion).
Distribution path
Non-obvious channel: EPA-funded TA intermediaries (RCAP network, state rural water associations, EPA WaterTA contractors) who serve hundreds of systems each and are paid to deliver exactly this help — one relationship yields many utilities. Secondary: the small-utility engineering/consulting firms already billing for records review, sold as a margin-expanding white-label tool rather than a competitor.
Pricing hypothesis
$0.50-2.00 per line-record processed (versus the claimed $5-15/line manual baseline — verify that figure), with a per-utility minimum ($1.5k-5k); white-label/bulk pricing for TA providers and consultants. Field-verification worklist prioritization included — that is where the utility saves real money (each avoided pothole/excavation verification is hundreds of dollars; HYPOTHESIS on exact figure).
Technical difficulty
Moderate and squarely in the founder's wheelhouse: OCR + VLM schema extraction + template mapping + a review UI. The hard part is not the model call — it is handwriting-heavy old cards, per-state template variance, and building the gold-set evaluation that makes accuracy claims credible. Solo-buildable in weeks with AI assistance.
Legal / regulatory risk
Low-moderate. No licensure required to process records or prepare inventories (engineering sign-off, where a state requires it, stays with the utility's engineer — position the tool as preparing, not certifying). Records contain addresses/owner names, so basic data-handling hygiene is needed but this is standard utility-records material, not sensitive PII at scale. Key non-legal risk: a primacy agency rejecting AI-assisted review as inadequate documentation — mitigated by per-row evidence citations and by asking 2 states early.
Platform dependency
Low. Extraction can run on Context.dev, any frontier VLM, or open models — the vendor is swappable. No platform owner can deplatform a tool whose output is a state spreadsheet template.
Founder fit
Very high. This is the founder's proven FMCSA shape: a federal rule compels a defined class to produce a submission to a government body, and the product is the automation layer priced per transaction. It also draws on his industrial-operations credibility (utilities respect operators, not slideware), public-records skills, and demonstrated-value sales style — the one-week finished-inventory demo is exactly how he sells.
Breakout potential
Real: 50 near-identical state markets, ~50,000 community water systems, a 10-year replacement-documentation tail after inventories (replacement tracking, annual inventory updates, funding-application paperwork), and the same scanned-municipal-records pipeline generalizes to other utility record-digitization mandates. The extraction layer commoditizes; the state-template + acceptance track record compounds.
Final recommendation
PURSUE, gated on the kill test. This is the founder's highest-fit shape — a finalized federal mandate, a defined compelled class, a paperwork bottleneck, and per-transaction pricing — with a genuinely new cost curve from cheap schema extraction. The two must-clear gates within 30 days: (1) at least one TA provider or consultant commits to a paid pilot, and (2) at least one state primacy agency confirms AI-assisted records review with evidence citations is acceptable documentation. If either fails, downgrade to a consultant white-label tool or kill.
Next action
Acquire one small utility's scanned tap-card set this week (rural water association intro or public-records request), build the extraction-to-state-template pipeline, and produce one finished inventory as the demo asset before making any sales contact.