Mastercard
Company-reportedCreator to judgeA sub-50-millisecond model can enrich transaction risk before the issuing bank decides approval.
Financial crime · Global
Collections: Regulated autonomy
Executive brief
High-scale risk AI can qualify as a workflow case while its performance claims remain explicitly company-modeled and refresh-sensitive.
AI value · Fraud-detection model performance
Company-reportedMastercard's initial modeling showed 20% average fraud-detection improvement.
Internal modeling, not audited production performance.
Before
Scores transactions from account, merchant, device, and purchase features. → Approves, declines, or challenges the transaction.
After
Analyzes entity relationships across a much larger graph and improves the risk score in under 50 milliseconds. → Uses the score with its own rules to approve, decline, or challenge.
Human boundary
Issuers decide approval, decline, challenge, and customer remediation.
Why it matters
A sub-50-millisecond model can enrich transaction risk before the issuing bank decides approval.
This case is company-reported. Use it for the operating-model shift; do not treat the numbers as independently measured.
Before
How the work ran before the change.
Decision Intelligence
Scores transactions from account, merchant, device, and purchase features.
ControlBank fraud rules and authorization policy.
Issuing bank
Approves, declines, or challenges the transaction.
ControlIssuer decision rights.
What changed
A sub-50-millisecond model can enrich transaction risk before the issuing bank decides approval.
Decision rightHuman moves from creator to judge
After
How the same work runs now.
Decision Intelligence Pro
Analyzes entity relationships across a much larger graph and improves the risk score in under 50 milliseconds.
ControlReal-time bounded scoring.
Issuing bank
Uses the score with its own rules to approve, decline, or challenge.
ControlIssuer retains authorization.
Exception path
Uncertain or challenged transactions follow issuer verification and fraud-investigation processes.
Work removed
Decision authority
Issuers decide approval, decline, challenge, and customer remediation.
Before
Decision Intelligence
Scores transactions from account, merchant, device, and purchase features.
Control: Bank fraud rules and authorization policy.
Issuing bank
Approves, declines, or challenges the transaction.
Control: Issuer decision rights.
After
Decision Intelligence Pro
Analyzes entity relationships across a much larger graph and improves the risk score in under 50 milliseconds.
Control: Real-time bounded scoring.
Issuing bank
Uses the score with its own rules to approve, decline, or challenge.
Control: Issuer retains authorization.
Work that left the path
Human role before
Fraud teams designed rules and reviewed exceptions around transaction scores.
Human role after
Fraud teams govern thresholds, monitor drift and false positives, and investigate exceptions while the model enriches each score.
AI roleDecision mode: advisory real-time risk scoring; issuing banks retain authorization decisions.
Existing Decision Intelligence scoring.→Mastercard's initial modeling showed 20% average fraud-detection improvement.
Pre-launch modeling reported February 2024. · Existing platform scored 143 billion annual transactions; new model scans one trillion data points.
Internal modeling, not audited production performance.
Existing Decision Intelligence performance.→Mastercard's analysis projected more than 85% reduction.
Pre-launch analysis reported February 2024. · Global payment scoring platform.
Company analysis with proprietary methods; outcome must be refreshed against production evidence.
Anti-pattern
Publishing modeled gains as audited realized results.
Questions
Portability conditions
Reputation risk
high
Evidence and authority
Watch · updated
Watch status: verify the cited source and deployment condition before reusing this case.
1 primary; publication outcomes are reported.
Bundle 1.0.0 · reviewed 2026-08-23 · stable ID ac275cd6b5259281
Google · Creator to judge
Large migrations can combine machine-found locations, LLM edits, automated checks, and engineer review.
Rising Academies schools · Creator to judge
Students can receive adaptive, low-bandwidth math practice without a scarce one-to-one tutor.
The Permanente Medical Group · Creator to judge
Ambient tools can draft clinical notes, while physicians remain the editors and signers.
Twenty-eight South-East Queensland long-term care facilities · Creator to judge
Hinge: a responsive social robot can provide repeatable short engagement sessions as an adjunct to care, without replacing clinical assessment or human relationships.