Anti-Money Laundering (AML) compliance has become one of the most resource-intensive obligations facing professional service firms. Law firms, accounting practices, and corporate service providers must verify identities, screen clients against sanctions and politically exposed person (PEP) lists, monitor ongoing relationships, and maintain audit-ready records — yet much of this work is still done manually, on top of inboxes full of scanned passports and half-completed onboarding forms. The result is a compliance function that consumes time, creates risk, and rarely scales with the firm.
Artificial intelligence is starting to change that. Large Language Models in particular are well suited to the parts of AML work that have historically resisted automation: interpreting unstructured text, separating relevant matches from noise, and turning fragmented evidence into a coherent narrative. Used well, AI does not replace the compliance officer — it delivers cases that arrive already triaged, reasoned, and explainable.
This is the premise behind AI-SENSE, our recently completed R&D project and the foundation of a new capability inside the Zygos practice management platform.
What AI-SENSE set out to do
AI-SENSE (AI Assisted Client Risk Assessment) was a nine-month R&D initiative carried out by Softline Computer Systems Ltd between July 2025 and March 2026. It pursued three objectives: build an AI-assisted client risk assessment tool inside Zygos, embed a structured, research-informed product development methodology within the company, and validate commercial viability through user research and prototype testing with existing clients.
The first month of user research produced a pivotal finding. Among 15 surveyed compliance professionals and managing partners, 93% still gathered onboarding information via unstructured email, and 87% reported that clients routinely submitted incomplete information. Building AI on top of that pipeline would have produced unreliable answers, so we re-shaped the architecture into two connected components: a digital onboarding layer that captures clean, structured data, and an AI screening layer that operates on it.
What we delivered
The project produced a working prototype, integrated into the live Zygos stack and now entering pilot deployment with selected client firms. The headline outcomes:
- A digital client onboarding system that replaces email back-and-forth with a structured, validated workflow, including document upload, OCR, and AI-based document classification.
- An AI-powered AML/PEP screening engine built around the Acuris API, with an LLM-based false-positive filtering pipeline — the core innovation — exposed to the user through the “Zygos Screening Agent,” which synthesises screening hits into a natural-language assessment with confidence scoring.
- A configurable risk scoring model mapped to AMLD4–6, the Cyprus AML Law (L.188(I)/2007), and sector guidance from CySEC and the Cyprus Bar Association.
- A provider-agnostic LLM architecture supporting OpenAI, Anthropic, Google and AWS, designed to avoid vendor lock-in and remain durable as the AI landscape evolves.
All five project deliverables (D1–D5) were completed on schedule and within the approved budget. Production release is planned for Q2–Q3 2026.
Acknowledgement
The AI-SENSE project (NPD-CAPBLD/0225/007) was funded under the RESTART 2016-2020 Programmes of the Research and Innovation Foundation (RIF) of the Republic of Cyprus, co-funded by the European Union – NextGenerationEU.
