Premise-Business Compatibility Rule Engine

Executive Summary

JIL Sovereign's premise-business compatibility rule engine detects fraud signals arising from mismatch between the business type billed on a claim or transaction and the physical premises the bill was generated from. A rule library declares compatibility rules, each comprising a billed-business-type pattern, a premise-feature pattern, and a compatibility verdict of allowed, suspicious, or prohibited.

Premise features are derived from a combination of street-view imagery classification, county-records square footage, business-license types registered to the address, and co-presence of unrelated business types at the same address. The engine evaluates every (billed-business-type, premise) pair against the library and emits a finding for any rule that fires with verdict suspicious or prohibited - capturing patterns such as hospice billing from personal-services premises, durable-medical-equipment vendors operating from residential addresses, and high-value procedures billed from premises lacking the physical infrastructure to perform them.

Problem Statement

Healthcare fraud, contractor fraud, and small-business loan fraud frequently involve incompatibility between the billed business type and the physical premises a business operates from. Detecting this requires correlating billed business type with premise characteristics - an analysis rarely performed by claim-only fraud platforms, even though the underlying premise data (street-view imagery, county records, business-license registries) is individually public and accessible.

Why Existing Solutions Are Insufficient

• Claim-coding anomaly detectors: examine CPT/HCPCS/ICD combinations, billing-rate outliers, and temporal patterns, but never look at where a bill physically originated.
• Individually-accessible premise data sources: street-view imagery, county records, and business-license databases exist but are not consolidated into a declarative compatibility rule library anywhere in production fraud tooling.
• Manual site-visit investigation: catches premise mismatch but does not scale to a claim population, and provides no systematic, versioned rule set an organization can improve over time.

Technical Architecture

Premise Feature Extraction

An external data integrator fetches and caches these sources per premise, so multiple billing entities sharing a premise amortize the same fetch. A premise feature extractor assembles the cached outputs into a single structured feature record per premise.

Rule Library & Evaluation

Each rule declares: rule_id, a billed-business-type pattern (matched via NAICS code or free-text match), a premise-feature pattern, a compatibility verdict (allowed / suspicious / prohibited), a regulatory-basis identifier, a remediation-template identifier, and a severity. Patterns support equality, regex match, numeric range, set membership, and conjunction, for example:

billed_business_type IN {hospice, home-health}
  AND premise.licensed_types CONTAINS personal-services   -> suspicious

billed_business_type = ambulatory-surgery
  AND premise.sqft < 1500                                 -> prohibited

A rule evaluator iterates the library in priority order for every (billing-entity, premise) pair and appends a finding on any match. Findings include the rule identifier, the billing entity, the premise feature record, the verdict, and source-data citations for every premise feature that contributed to the match - so every finding is independently auditable back to its street-view frame, county record, or license filing.

Rule Lifecycle & Validation

New rules can be added by an authorized operator at runtime without code redeployment; every addition is written to an append-only audit log. Rules support time-bounded scoping, so a rule applies only to claims dated within a specified period - allowing the library to track regulatory changes such as new zoning rules taking effect on a given date. Each rule also declares a confidence score that findings inherit, letting investigators triage the highest-confidence matches first.

The initial ~40-rule library detects hospice-from-personal-services, DME-from-residential, and ambulatory-surgery-from-undersized-premise patterns at high precision, validated against a synthetic multi-entity test scenario spanning tens of thousands of claims with a substantial share of confirmed premise mismatches - demonstrating that the architecture surfaces coordinated-premise fraud patterns that claim-coding review alone does not catch.

Prior Art Differentiation

Patent Claim