Financial fraud detection: implementation comparison
Fraud detection governed two ways: accuracy controls, or freeze receipts with a reason, owner, and date, wrongful freezes counted by rule, and time to restore.
Where governance breaks down
Banks tune fraud systems to catch fraud. Customers frozen out by mistake wait, without reasons, for their money.
Standard governance emphasizes compliance logging and human review of flagged activity. In the Dutch childcare benefits scandal, a fraud risk score was treated as a finding, and the people it flagged could not see why. Ethotechnics measures how fast a legitimate customer recovers access, and whether a pattern of wrongful flags reaches whoever owns the rule.
What standard AI governance implements
Controls focused on detection accuracy and regulatory reporting.
- Real-time monitoring with automated alerts for suspicious patterns.
- Audit logs aligned to AML and compliance requirements.
- Human review queues for flagged transactions or accounts.
- Explanations that list triggering factors for the freeze.
- Ownership structures for compliance escalation and review.
Ethotechnics implementation
What changes, and what to verify before launch
Each freeze has an owner and a date, and wrongful freezes are counted against the rule that made them.
- Each freeze issues a receipt with the reason, the owner, and the date by which the customer will get an answer. That is what contestability means here.
- The owner can lift the freeze. If no fraud is confirmed by that date, the freeze is lifted.
- Time to restore, from freeze to restored access for a legitimate customer, is measured alongside detection accuracy.
- Freezes lifted as wrongful are counted by detection rule. The count goes to the rule's owner, who reviews the rule and records what changed.
- A named person can halt one detection rule while it is reviewed. The halt contains the harm. The review changes the rule. Each restoration, with any fees or missed payments repaid, is entered in the repair log.
Before launch, publish the receipt fields and set a time-to-restore target for customers frozen in error. Decide how the person who can halt a detection rule is alerted. The repair log also records whether downstream losses were repaid.
Where this fits
Credit, payments, and account holds
Whether a person gets credit, at what limit, and whether they can use the money or the account they already have.
- Scored cases
-
- Apple Card credit limits: STD-02 §8.1 would have caught it
- Standards to adopt
-
- STD-01 The Temporal Bill of Rights · Draft v1.0.3
- STD-02 Contestability and recourse · Draft v1.4.1
- Run first
- Delegation audit. Take one hold or limit workflow through six questions. See which actions nobody can justify, and whether each can be undone.
Cite this implementation example · v1.0.0 · updated Feb 1, 2025 · CC BY-SA 4.0
APA
Ethotechnics Institute. (2025). Financial fraud detection: implementation comparison. Ethotechnics Institute. https://ethotechnics.org/use-cases/examples/financial-fraud-detection
MLA
Ethotechnics Institute. "Financial fraud detection: implementation comparison." Ethotechnics Institute, 2025, https://ethotechnics.org/use-cases/examples/financial-fraud-detection.
Chicago
Ethotechnics Institute. "Financial fraud detection: implementation comparison." Ethotechnics Institute. Feb 1, 2025. https://ethotechnics.org/use-cases/examples/financial-fraud-detection.
BibTeX
@misc{ethotechnics_use_cases_examples_financial_fraud_detection,
title={Financial fraud detection: implementation comparison},
author={Ethotechnics Institute},
year={2025},
howpublished={Ethotechnics Institute},
url={https://ethotechnics.org/use-cases/examples/financial-fraud-detection},
version={v1.0.0}
}
RIS
TY - WEB TI - Financial fraud detection: implementation comparison AU - Ethotechnics Institute PY - 2025 UR - https://ethotechnics.org/use-cases/examples/financial-fraud-detection ER -