Unit21, a provider of AI Risk Infrastructure for fraud prevention and AML monitoring, has launched Agentic Task Builder, a capability enabling risk and compliance teams to create custom investigation tasks using plain English prompts. The tool extends the company’s existing AI Investigation Agents—which triage alerts, investigate cases, and draft regulatory narratives—by allowing teams to define bespoke logic without relying on engineering resources or vendor roadmaps. This addresses a long-standing challenge in financial crime programs, where unique typologies and thresholds often require tailored workflows that traditional AI tools cannot accommodate. The Agentic Task Builder represents a significant evolution in how financial institutions can adapt their AI systems to meet specific operational and regulatory demands, moving beyond rigid, pre-defined functionalities to a more dynamic, user-driven approach.
Agentic Task Builder Enables Plain English AI Task Creation
The Agentic Task Builder allows users to describe investigation tasks in natural language, which the system then translates into executable AI workflows. For example, teams can prompt the tool to “search for adverse media or sanctions exposure tied to this entity” or “review transactions for deposit and withdrawal patterns just under reporting thresholds.” The platform generates two core components: one for external data searches (e.g., web research) and another for internal data analysis (e.g., transaction pattern detection). These components integrate into the AI Agent workflow, combining outputs into a unified investigation narrative. Every task execution is designed to be explainable, auditable, and compliant with regulatory evidentiary standards, ensuring human oversight remains central to decision-making. The external data component leverages generative AI to research public internet sources, returning structured answers with citations, while the internal data component analyzes transaction patterns, entity behavior, and counterparty activity directly within the alert’s summary panel. This dual approach ensures comprehensive coverage of both external and internal factors critical to financial crime investigations.
Custom Investigation Logic Without Engineering Dependency
Historically, modifying AI-driven risk workflows required technical teams to recode systems or wait for vendor updates. The Agentic Task Builder removes this bottleneck by enabling analysts to build and refine tasks independently. The tool includes a built-in quality judge that evaluates outputs, suggests improvements, and iterates on prompting logic. It also recommends additional data sources based on the user’s environment, enhancing task precision. Before deployment, teams can backtest tasks against historical alerts to validate performance and identify potential gaps. Once live, the system allows sampling of results by agreement/disagreement rates between analysts and AI, supporting progressive autonomy settings at the queue level. This self-service approach aims to align AI capabilities with each institution’s specific risk profile and operational needs. The Agentic Task Builder doesn’t simply execute instructions—it reasons over the user’s environment like an experienced analyst, clarifying intent and refining outputs through iterative feedback. For instance, a fraud ops manager can create a task to detect account-takeover and mule signals by analyzing rapid fan-in/fan-out transactions or new-device activity before large payouts. Similarly, AML investigators can design tasks to check entities for adverse media or sanctions exposure, or to identify structuring patterns in transactions. Compliance teams can deploy custom name-matching and pass-through analysis on-demand, without waiting for engineering sprints or vendor updates. Each task is tuned to the program’s specific thresholds, narrative formats, escalation rules, and review controls, ensuring consistency across teams and geographies.
Implications for Financial Crime Programs
The launch responds to growing demand for adaptable AI solutions in financial crime detection. Unit21 reports that its AI Agents have reviewed over 1.5 million alerts to date, with customers already creating custom tasks in minutes—a process that previously took hours or weeks. By decentralizing AI customization, the tool may reduce reliance on centralized engineering teams while accelerating response times to emerging threats. However, adoption will depend on how effectively institutions balance automation with regulatory scrutiny, particularly in validating AI-generated insights against compliance frameworks. The platform’s emphasis on explainability and audit trails positions it as a potential fit for programs prioritizing defensible decision-making. The Agentic Task Builder’s ability to backtest tasks against historical alerts ensures that teams deploy with evidence rather than assumptions, flagging unrepresentative samples and surfacing warnings. Once deployed, the system enables teams to monitor agreement and disagreement rates between analysts and AI, allowing for progressive autonomy adjustments at the queue level. This adaptability is crucial as financial institutions face increasingly sophisticated threats and must maintain compliance with evolving regulatory standards. The tool’s design reflects a shift toward empowering individual analysts with AI capabilities tailored to their specific workflows, potentially transforming how financial crime programs scale and respond to new risks.
Key Takeaways
- Unit21’s Agentic Task Builder allows risk and compliance teams to create custom AI investigation tasks using plain English prompts, eliminating the need for engineering involvement.
- The tool generates two task types: external data searches (e.g., adverse media checks) and internal data analysis (e.g., transaction pattern detection), both integrated into AI Agent workflows.
- Teams can backtest tasks against historical alerts and adjust autonomy levels based on AI performance, with all outputs designed to meet regulatory audit standards.
FinanceInsyte's Take
Unit21’s Agentic Task Builder addresses a critical gap in AI adoption for financial crime programs: the mismatch between standardized tools and institution-specific workflows. While the capability to build custom tasks in minutes could streamline operations, its success hinges on how well teams navigate regulatory expectations around AI explainability and validation. Buyers should evaluate whether the platform’s self-service model aligns with their compliance infrastructure and risk tolerance, particularly as financial institutions increasingly seek scalable yet defensible automation solutions.
Source: Businesswire
