Feedzai has launched Farol, an artificial intelligence (AI) agent embedded in its RiskOps Studio to analyse fraud-detection rules, retrieve case data, summarise alerts and draft suspicious activity reports (SARs).
Embedding the agent within the risk platform gives it access to transaction data and existing investigation workflows. Feedzai positions this approach as an alternative to connecting a separate AI model that lacks the institution’s operational context. Farol runs within each financial institution’s environment, with its data and outputs remaining inside the customer’s technology estate.
Farol launches with four sets of functions:
- Risk Strategy – Interrogates and analyses rule sets to surface actionable insights, including identifying rules that are generating noise without catching fraud, and recommending sharper thresholds to enhance performance. Management of rule hygiene becomes a task that takes minutes and not days.
- Investigations – Retrieves and summarizes alert data in moments, giving analysts the context they need to work cases faster and with greater conviction, proven to reduce alert handling times by 20%.
- Knowledge – An always-on product expert built into the workflow. Users can ask Farol how to do anything on the platform and get the answer instantly, without ever leaving the user interface to find it.
- SAR Drafting – Drafts Suspicious Activity Reports (SARs) up to 12x faster by cutting the manual effort of compiling and summarizing information.
Feedzai claims that Farol has reduced alert-handling times by 20% and can produce draft SARs up to 12 times faster than a manual process.
Justinas Rekus, fraud prevention business owner at SEB, said: “Having a single, intelligent interface to handle data retrieval, insight generation, and production-ready rule suggestions fundamentally transforms how we refine our fraud strategies.”
The product also provides audit trails and what Feedzai describes as autonomous execution capabilities. Feedzai said its research found that 68% of financial institutions were testing agentic AI. It argues that many projects have yet to generate operational efficiencies because third-party models remain separated from real-time transaction data.
Subscribe to our newsletter
