Percy by Footprint

• Clearing false-positive AML watchlist and sanctions hits with a cited explanation, and escalating true matches to an analyst
• Running enhanced due diligence on a high-risk customer and producing a documented report for the case file
• Researching a business, its registry records and beneficial owners during commercial onboarding
• Investigating transaction monitoring alerts and documenting the findings in a cited report for an analyst to approve
• Screening adverse media for customers flagged by a KYC or KYB playbook
• Backtesting a new or changed procedure against historical resolved cases before rolling it out

Percy is aimed at the case work itself in bank compliance, not at a chat assistant beside it: it learns your written procedures, backtests them on your own history, and leaves a cited trail an examiner can follow. That design suits teams whose problem is queue volume and documentation, not a lack of policy. The open questions are the ones the public material does not answer: what it costs, how much of the agent-building is done by you versus Footprint's team, which models sit underneath, and whether you need Footprint's onboarding stack to use it. Treat the speed figures as a starting point for a backtest on your own cases, not as a forecast.

About Percy by Footprint

Footprint built its business on identity verification, KYC/KYB onboarding flows, an encrypted PII vault and a review platform for risk teams; it now presents itself as an AI operating system for risk, with Percy as the agent layer on top. (It is unrelated to BrowserStack's Percy visual testing tool.) The workflow on the product page has five steps: you describe the standard operating procedure your team already follows, Percy drafts an agent from it as auditable steps, the agent is backtested on your own resolved cases, you refine it with plain-language feedback, and it then runs in production, checking findings against data sources Footprint says span 100+ sources in 100+ countries. Footprint names LexisNexis, Experian, ComplyAdvantage and government registries as pre-built integrations. Its listed modules include AML watchlist screening, sanctions screening, adverse media, enhanced due diligence, commercial onboarding and business due diligence, transaction monitoring investigations, alert triage and false positive resolution, alongside fraud checks such as document tamper, synthetic and behavioral fraud detection. Each investigation produces a report with citations, timestamps and a task-by-task trail, and every change to an agent is versioned and tested before release. In Footprint's developer docs, a Percy agent runs as an AI Agent node inside a Footprint playbook, pinned to a released agent version, and returns structured output that later rules, branch or action nodes use, for example to clear a case or raise a manual review for a human queue. A layer called Trust Fabric stores precedents from past cases as shared organizational memory, where agents propose and humans approve, and checks whether agents follow policy; Footprint says banks can use Trust Fabric's organizational memory as a standalone product. Footprint's own figures put watchlist hits at under a minute instead of about 30 minutes, and enhanced due diligence at under fifteen minutes instead of three hours; these have not been independently verified. The tradeoffs: pricing is not published and access starts with a demo, the docs cover the surrounding platform in depth but say little about configuring Percy agents themselves, the underlying AI models are not disclosed, and no self-hosted deployment is documented.