Program fraud detection for Unify
A staff-facing integrity console that turns Unify's historical application corpus into reviewer-ready signals for coordinated fraud, synthetic identities, and document/payment anomalies without automating denials.
What we can show today
benefit applications available for pattern learning
program history to ground cross-program fraud signals
automatic denials from model flags
Risk Overview
Flagged vs. clean application ratio
Recent Runs
Latest analysis activity
Top Vectors
Highest-yield detectors by flags raised
Fraud detection that protects legitimate applicants
Use nearly 450,000 historical applications to learn how fraud behaves, then turn that into evidence-driven detection that minimizes friction for legitimate households.
Historical behavior modeling
Build baselines from prior applications so the platform knows what normal, anomalous, and previously confirmed fraud patterns look like.
Coordinated fraud detection
Look for linked submissions that reuse entities, documents, devices, and payout rails across apparently separate applicants.
Synthetic identity vectors
Detect applications that use realistic but fabricated identities by measuring internal consistency across identity, payment, and household details.
Linked applications sharing payout, device, address, or document signals.
Identity records with realistic details but inconsistent internal evidence.
Historical examples feeding watchlists, similarity search, and vector tuning.
Flags create review work, not automatic adverse action.
High-risk escalation should require more than a single weak indicator.
Set a target ceiling for reviewed legitimate applicants caught in fraud queues.
Unify Program Integrity Pilot
A presentation-ready slice of the broader fraud-intelligence model: analyze historical patterns, surface coordinated attempts, and protect legitimate applicants from blunt rules.
Rental assistance review stages
The first production slice for a broader fraud-intelligence platform.
Upload rental assistance batch and normalize mapped fields.
Perform four-point profile review and obvious document checks.
Apply payment, identity, and template-based vectors across the batch.
Escalate suspicious combinations to specialized fraud review.
Confirm fraud, dismiss, or hold for manual follow-up.
Promote confirmed signals into watchlists and the fraud corpus.
What this frontend now covers
Datasets
Upload program datasets, normalize identity and payout fields, and prepare historical application records for fraud analysis.
Fraud Vectors
Create deterministic and similarity-based detectors from documented vectors, confirmed fraud patterns, and synthetic identity indicators.
Analysis Runs
Launch batched analysis across mapped applications, process coordinated-risk vectors, and inspect reviewer-ready outcomes.
Application Review
Investigate rental assistance applications, review case stage, request explanations, and disposition cases.
Synthesis
Explain why an application was flagged with a selectable model through OpenRouter or local Ollama.
Configured fraud vectors
Rule-based and similarity detectors ready to run.
Shared Address + IP Reuse
Flags rental assistance applications when the same service address and submission IP appear repeatedly within a review window.
Organized Fraud Similarity Cluster
Compares normalized rental assistance application fingerprints against a confirmed fraud corpus of template reuse and payment-risk cases.
Digital Gift Card + Tax Return
High-confidence vector derived from audit findings: digital gift card payment combined with tax return documentation.
Go2Bank / Green Dot + Utility Template
Escalates applications using high-risk routing numbers paired with utility-bill documentation or duplicated utility templates.