Executive Summary
Benefit-program fraud frequently involves a business that bills a volume of claims its actual physical location could never have serviced - a "provider" operating out of a mailbox store, or a home-health agency billing thousands of visits from a 1,400-square-foot strip-mall unit. JIL Sovereign's premise-capacity detector resolves each registered, billing, and service address into a classified premise type and an estimated square footage by fusing five independent classification sources, then checks the entity's claimed business type and actual flagged billing volume against a versioned compatibility table keyed on (business type, premise type) that carries an explicit volume ceiling per pairing.
When the claimed volume exceeds the ceiling for a premise that would otherwise be considered a compatible location for that business type, the verdict escalates from "compatible" to "volume impossible" and the excess factor - claimed volume divided by premise capacity - is captured as structured evidence.
Problem Statement
Address verification in most payment-integrity tooling stops at confirming that an address exists and is deliverable - it does not ask whether the physical premises at that address could plausibly generate the billing volume attributed to it. That gap is precisely where a class of durable-medical-equipment, home-health, and hospice fraud schemes operates: register a legitimate-sounding business at a real, deliverable address that happens to be far too small (or the wrong type entirely - a residential address, a mailbox-service suite, a vacant lot) to support the claimed patient volume, and bill anyway.
Confirming this by hand requires pulling property records, checking street-level imagery, and cross-referencing a business-type-appropriate capacity assumption - work that does not scale across a large provider or vendor population, and that most automated eligibility and payment systems never attempt at all.
Why Standard Address Verification Falls Short
- Deliverability-only checks: confirm an address is a valid mail-deliverable location, not whether the premises can support the claimed activity.
- Single-source classification: a single imagery or registry source can misclassify a premises; JIL's detector requires an aggregated, highest-confidence classification across five independent connectors before a verdict is reached.
- No capacity ceiling: most systems that do classify premise type stop there - they don't cross-reference the classification against a versioned, business-type-specific volume ceiling to catch a physically-impossible claim.
Technical Architecture
Multi-Source Classification
| Connector | Signal Contributed |
|---|---|
| Postal-service address classification | Residential vs. commercial designation |
| Commercial-places categorization | Business-type signals for the listed location |
| Street-level imagery + vision classification | Visual features - storefront glass, medical-grade entrance, signage, porch/lawn residential cues |
| Public property records | Estimated square footage, zoning |
| Mailbox / virtual-office service registry | Detects UPS Store, PMB, suite-number, and similar non-physical-operation patterns |
Each connector returns a classification with a confidence score; the highest-confidence non-unknown classification across all connectors becomes the premise's premise_type, one of a fixed taxonomy (residential single-family, residential multifamily, commercial office, medical office building, hospital campus, retail strip mall, retail standalone, industrial warehouse, mixed use, mailbox service, virtual office, gas station, restaurant, vacant/undeveloped, other/unverified). Estimated square footage is inferred from whichever connector reports it. Classification results persist keyed on an address fingerprint, so the same physical premises is classified once and reused across every entity and engagement that references it - a compounding asset rather than a per-engagement lookup.
Compatibility and Volume-Ceiling Cross-Reference
For each entity's registered, billing, and service address, the entity's claimed business type and the resolved premise type are looked up against a versioned business-premise-compatibility reference table, which returns a base compatibility verdict (compatible, soft mismatch, hard mismatch, or unverified when no rule exists) plus, where applicable, a volume ceiling and its unit. If the base verdict is "compatible" but the entity's actual flagged payment count exceeds the premise's volume ceiling, the verdict is escalated to volume impossible, and a structured volume-check record is attached:
| Field | Meaning |
|---|---|
claimed_volume | The entity's actual flagged payment count |
premise_capacity | The versioned volume ceiling for this (business type, premise type) pairing |
excess_factor | claimed_volume / premise_capacity - how many multiples over capacity the claim represents |
Prior Art Differentiation
| Approach | Verifies Address Exists? | Classifies Premise Type? | Checks Volume Against Physical Capacity? |
|---|---|---|---|
| Standard address-verification APIs | Yes | No | No |
| Single-source imagery/registry classifiers | Yes | Yes, single-source | No |
| JIL premise-capacity detector | Yes | Yes, five-connector aggregate | Yes, versioned volume ceiling per business/premise pairing |