Feedzai launches Farol agentic AI to cut fraud investigation times

Feedzai, the AI-native financial crime prevention platform headquartered in Lisbon and New York, has launched Farol, an embedded agentic AI system designed to reduce the time fraud analysts spend assembling case data and drafting regulatory reports. The launch, announced at the company's Feedzai Fusion London event on 24 September 2026, arrives as banks face a compounding cost squeeze: mandatory reimbursement rules in several jurisdictions are expanding loss exposure, while headcount in fraud operations continues to grow faster than budgets allow.

Farol sits inside Feedzai's existing RiskOps Studio platform rather than operating as a separate tool, which the company says resolves a structural problem that has plagued enterprise AI pilots. According to Feedzai's own research, 68% of financial institutions are actively testing agentic AI, yet most deployments fail to generate operational efficiency because third-party AI models lack access to real-time transaction data. By embedding the agent directly into existing risk workflows, Feedzai says Farol can deliver full contextual awareness and transparent audit trails without requiring infrastructure changes.

What Farol actually does

The agent launches with four capability clusters. The Risk Strategy module interrogates live rule sets to surface underperforming detection logic, with the company claiming that rule hygiene management can be reduced from days to minutes. The Investigations module summarises alert data on demand and, according to Feedzai, has been proven to reduce alert handling times by 20%. A Knowledge module acts as an always-on product guide embedded in the analyst's interface. Most structurally significant for compliance teams is the SAR Drafting skill, which automates the compilation of Suspicious Activity Reports and is claimed to cut drafting time by up to 12 times versus manual processes.

Justinas Rekus, Fraud Prevention Business Owner at SEB, the northern European financial services group operating across more than 20 countries, offered the launch's most substantive external endorsement: "Having a single, intelligent interface to handle data retrieval, insight generation, and production-ready rule suggestions fundamentally transforms how we refine our fraud strategies."

Sam Abadir, Research Director for Risk, Financial Crime and Compliance at IDC, framed the longer-term significance: agents that continuously flag which rules generate noise without catching fraud allow teams to tune detection logic on an ongoing basis rather than during periodic reviews, a shift from reactive to continuous optimisation.

The convergence angle: agentic AI meets regulatory pressure

The Feedzai launch illustrates a broader pattern emerging across financial services: the collision between agentic AI capability and a tightening regulatory environment is becoming the primary driver of AI adoption in operations-heavy banking functions. Mandatory reimbursement frameworks in the UK (under the Payment Systems Regulator's authorised push payment rules) and proposed equivalents in the EU are creating direct financial liability for fraud losses that previously sat with consumers. That liability shift changes the internal ROI calculus for compliance technology investment significantly.

For cross-sector investors, the story is equally instructive. Financial crime prevention sits at the intersection of AI infrastructure spend, regulatory compliance technology (regtech), and the broader agentic AI buildout now absorbing capital across enterprise software. The regtech segment has historically attracted steady institutional funding, but the addition of genuine agentic reasoning capability is drawing interest from AI-focused growth investors who previously concentrated on foundation-model layers rather than vertical application. Feedzai, which counts significant institutional backing and counts tier-one banks among its customers, is positioning Farol as a template for what embedded agentic AI looks like when it operates within a customer's own data environment, a privacy and audit-trail architecture that may prove increasingly important as regulators scrutinise AI decision-making in credit and fraud contexts.

The wider question for the sector is whether embedded, workflow-native agents displace the category of standalone AI co-pilot tools that financial institutions have been piloting in parallel. If Feedzai's fragmentation thesis is correct, and most isolated AI deployments genuinely fail to reach production efficiency, the competitive advantage may accrue to platforms that own the underlying data layer rather than those layering intelligence on top of it.