The Data Sharing Crisis in Finance

Scenario

A coordinated fraud ring works three banks simultaneously, opening accounts with stolen identity documents, running small transactions to avoid suspicion, then conducting large wire transfers. Bank A blocks one card after 14 days. Bank B has no idea and gets hit the same week. Bank C takes another 10 days to detect. Each bank loses $200K–$500K before the fraud is caught. Worse: organized networks operate across dozens of institutions with no cross-bank signal sharing. Regulators fine banks $5M+ annually for failing to detect coordinated activity.

How It Works

All three banks upload encrypted fraud-signal streams to Placino: transaction patterns (velocity, geography, merchant sequences), device fingerprints, and account-opening metadata — encrypted with envelope encryption (AES-256-GCM). Placino performs federated matching: cryptographic hashing of key indicators (card last-4, device hash, merchant MCC sequence) identifies accounts with common underlying attributes without exposing customer names or account numbers. Private Set Intersection identifies shared signals across banks. Federated clustering algorithms detect coordinated transaction patterns with differential privacy (epsilon = 1.0). Results: alerts for detected rings, never raw customer data. Merkle-chain audit trails log all queries for regulatory compliance.

Scenario

A retail bank with 2M depositors partners with an insurance company. The bank identifies 800K customers with strong balance sheets and no recent insurance claims — prime targets for term life and property insurance cross-sell. But the bank cannot share customer names, addresses, or account balances due to privacy law (GDPR, CCPA). The insurer cannot gauge product relevance without seeing account data. Opportunity lost: the insurer sends cold campaigns to mass audiences; conversion rates stay 0.8%. Annual revenue upside: $8.5M–$12M, abandoned.

How It Works

Bank encrypts customer attributes (hashed email, account age, deposit balance cohort, stated life stage) using envelope encryption and uploads to Placino. Insurer uploads its customer base with the same hashed email format. Placino's Private Set Intersection identifies shared customers without exposing email addresses or account details to the insurer. Federated cohort analysis segments the matched audience: "45K high-networth customers age 35–50 with families are 7.2x more likely to buy term life than random audience." Insurance proposes targeted campaigns to the bank with behavioral segments, not names. The bank applies recommendations locally using k-anonymity-enforced segments (min 100 customers). Completion rates: insurer sees only aggregate conversion stats returned via differential privacy (epsilon = 1.5).

Scenario

A lender reviews a personal loan application: 24-year-old, recent college grad, no credit history, limited bank account age. Traditional credit bureaus offer no score. The lender rejects 80% of "thin-file" applicants out of caution — but 40% would become reliable customers with 0.8% default rates if risk signals existed. The lender turns away $15M in profitable loan volume annually due to incomplete data. Meanwhile, telco and utility companies possess behavioral signals (bill-pay punctuality, account stability, churn risk) that perfectly predict lending risk — but privacy laws prevent sharing raw customer data.

How It Works

Telco and utility partners encrypt anonymized behavioral records — hashed customer ID, payment-on-time rate, account age, churn propensity score — using envelope encryption (AES-256-GCM). Lender uploads applicant profiles (age, employment history, bank account age) with hashed identifiers. Placino performs Private Set Intersection to identify applicants present in both datasets. Federated probabilistic matching computes risk cohort membership: "Applicants matching your profile have 12% historical default rate and strong payment history." Results are cohort-level risk signals with k-anonymity thresholds (min 1K applicant matches per cohort) and differential privacy noise (epsilon = 2.5). Lender applies signals as override variables in risk models, never seeing individual applicant-level telco or utility data.

Scenario

A network of shell companies moves $500M annually across a dozen banks, depositing just under $10K in each to avoid reporting thresholds (structuring). Bank A detects a $9.9K pattern in one account and files a Suspicious Activity Report (SAR). Bank B sees similar $9.9K deposits across 15 accounts in the same week but attributes it to coincidence. Bank C is never informed. The network successfully moves money across the financial system undetected. Regulators fine participating banks $8M–$15M per incident for AML control failures. Banks waste $400M+ annually filing false-positive SARs on innocent customers — tying up compliance teams and damaging customer relationships.

How It Works

Partner banks upload encrypted transaction streams to Placino: transaction amounts, counterparty information (merchant name, account number, routing number), timestamps, and geographic data — all hashed using cryptographic key derivation. Placino federated matching identifies common counterparties across institutions without exposing account holder identities. Federated anomaly detection applies clustering algorithms to flag structured transactions, velocity patterns, and coordinated timing inconsistent with legitimate activity. Differential privacy (epsilon = 1.0) is applied to aggregate results so banks see network-level signals ("5 institutions detected deposits to this merchant network in 4-day window") rather than individual-account details. Merkle-chain audit ensures regulatory evidence is preserved.

Scenario

A bank spends $10M annually on mortgage and personal loan campaigns across Google Ads, Facebook, LinkedIn, and email. UTM tracking captures 40% of applications; the other 60% have no attributed channel. Attribution logic is guesswork: the bank credits most conversions to organic search or direct traffic because it lacks visibility into multitouch journeys. Marketing cannot optimize budgets across channels or justify spend to leadership. High-ROI channels go unfunded; low-ROI channels persist due to measurement blindness. The bank leaves $2M–$4M on the table annually.

How It Works

Bank uploads encrypted impression logs (campaign ID, user cohort hash, timestamp, hashed email) from Google, Facebook, and email systems. Bank also encrypts application records (application date, loan type, hashed email, approval status) using envelope encryption (AES-256-GCM). Placino performs federated identity resolution using hashed email as the join key, matching impressions to downstream applications without exposing email addresses to analytics partners. A federated attribution model applies propensity-score matching to control for selection bias (customers who clicked on ads may differ from non-clickers in unobserved ways). Multi-touch attribution computes channel contribution with differential privacy (epsilon = 1.5). Results: "Google campaigns contributed 32% of attributed conversions; Facebook 18%; email 15%; other 35%." Bank sees channel-level ROAS, not individual customer journeys.