Kyc aml · Production

Feedzai TrustScore: Enabling Network Intelligence to Fight Financial Crime

The problem

Traditional fraud detection relies on rule-based systems or institution-specific custom AI models that are slow to deploy, require constant maintenance, and take a reactive and siloed view of fraud because each is limited to data from a single institution, while fraudsters operate and collaborate across borders.

First attempt

Custom AI models trained on a single institution's historical data are blind to emerging fraud patterns that have not appeared in their own environment, and rule-based systems require months of analyst work before deployment with constant manual updates thereafter.

Workflow diagram · grounded in source

Transaction event input

trigger

“this happens within a live and dynamic environment, where financial behaviors and technologies are changing over time”

Data standardization

integration

“the format and the meaning of the data must be standardized, such that the patterns that the experts learned can be seamlessly integrated into new financial institutions”

Specialist expert scoring

ai_action

“Each expert model is developed to become a 'specialist' in a specific geography and use case (e.g., banking, payment processing, or anti-money laundering)”

MoE risk score aggregation

ai_action

“Feedzai TrustScore is built on a Mixture of Experts (MoE) architecture, a modular approach where multiple expert models are developed on different environments or fraud scenarios, and then are combined to produce the final risk decision”

Stable risk score output

output

“each score consistently reflects the same level of transaction risk, making it more interpretable”

Continuous model refreshing

feedback_loop

“Feedzai TrustScore is monitored and tuned regularly at both the expert level and the MoE aggregation level. This ensures robustness and allows us to detect model drift early”

Reported outcome

Real-world results from Feedzai TrustScore show a boost in fraud detection, a reduction in false alerts, and a faster time to market, with new clients able to start making predictions immediately without needing historical data collection.

Reported metrics

Fraud as fraction of financial transactionsless than 0.1%

Average loss per fraud victimover $2,000

Fraud detection improvementboost in fraud detection

False alert reductionreduction in false alerts

Show all 5 reported metrics

fraud as fraction of financial transactionsless than 0.1%

average loss per fraud victimover $2,000

fraud detection improvementboost in fraud detection

false alert reductionreduction in false alerts

time to marketfaster time to market

Reported stack

Feedzai TrustScoreFeedzai IQ™Mixture of Expertsfederated learning framework

Source

https://medium.com/feedzaitech/feedzai-trustscore-enabling-network-intelligence-to-fight-financial-crime-9ce7fcff84fb

Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Real-world results from Feedzai TrustScore show a boost in fraud detection, a reduction in false alerts, and a faster time to market, with new clients able to start making predictions immediately without needing histo…

What tools did this team use?

Feedzai TrustScore, Feedzai IQ™, Mixture of Experts, federated learning framework.

What results were reported?

Fraud as fraction of financial transactions: less than 0.1%; Average loss per fraud victim: over $2,000; Fraud detection improvement: boost in fraud detection; False alert reduction: reduction in false alerts (source-reported, not independently verified).

What failed first in this deployment?

How is this kyc aml AI workflow structured?

Transaction event input → Data standardization → Specialist expert scoring → MoE risk score aggregation → Stable risk score output → Continuous model refreshing.