Complying with anti-money laundering and countering the financing of terrorism (AML/CFT) regulations is reaching an operational breaking point. New regional compliance packages are being released, local regulators are ramping up their oversight, and traceability requirements are tightening. At the same time, Compliance departments are stacking up more and more tools but none of them are ever connected.
As a result, it takes an average of 47 minutes to process each AML alert that gets generated, and out of every one hundred alerts, an average of 97 turn out to be false positives, with only 3 resulting in a suspicious activity report (SAR). For a household of four linked to the same alert, the exercise consumes nearly 76 cumulative hours for the bank, the equivalent of 10 man-days. The problem isn’t a lack of tools, as banks are well equipped with such tools, but rather the lack of a bridge between them.
This is the bridge that is provided by Sentinel, the augmented investigation solution designed by Clever Age, using the Denodo Platform as the AI data layer and Ofelia as the AI orchestration engine. Here is how this combination concretely transforms the work of AML analysts and the returns it delivers.
The Hidden Cost of Silos
Take Thomas, an anti-money laundering officer (AMLO) at a major financial institution. A Level 3 alert comes in: a suspicious wire transfer of $160,000. To process it, he opens his Know-Your-Customer (KYC) tool, but this presents him with an incomplete file. He switches to the CRM to find supporting documentation, then to a third tool to extract a transactional history. He then leaves the bank’s internal information system to check commercial registries and identify the ultimate beneficial owners (UBOs) of the counterparty, before cross-referencing these names on a negative news platform, which returns dozens of articles to sort through manually, including those that merely contain homonyms of the key search terms.
Digging deeper, Thomas discovers that his client is linked to three other accounts, so in total he has to repeat this data collection four times. Each individual file consumes 30 to 60 minutes; the entire household approaches 1.5 hours, not counting the summary he has to write from memory for the Financial Intelligence Unit (FIU) that is investigating the SARs. Throughout his day, Thomas spends about 90% of his time handling tools, and only 10% exercising his expert judgment.
This is precisely the ratio that Sentinel flips upside down.
Under the Hood: Denodo and Ofelia, the Duo Making Agentic AI Governable
Sentinel doesn’t replace any of the bank’s existing systems such as those for core banking, KYC, filtering, screening, or external registries (commercial registries, sanctions lists, negative news). It connects them without duplicating them.
Denodo handles the data side. As an AI data layer, Denodo connects all internal and external sources and exposes them through a single logical model, without requiring replication: data stays right where it is, governed, tracked, and updated in real time. The Denodo AI SDK, paired with a Model Context Protocol (MCP) server, transforms this foundation into a standardized interface for AI agents: preparing business views, exposing metadata understandable by a large language model (LLM), executing queries, and formatting results. In concrete terms, an AI agent no longer needs to navigate the diversity of physical sources (core banking, third-party repositories, SWIFT/SEPA flows, OSINT databases); it reasons on a single model and enables users to query the data using natural language via the tools exposed by the MCP Server. This is the foundation that Clever Age calls the “Foundation of Truth”: without it, AI only reasons on disjointed fragments.
Ofelia handles the journey. Ofelia is a deterministic, auditable, and modular orchestration engine: it sequences the investigation methodology into eight steps and three concentric circles (individual analysis, relational circle, extended ecosystem), conditionally activates relevant analysis methods (temporal analysis, peer groups, network, competing hypotheses), and prohibits any unjustified scope expansion. Every step of consulted data, activated methods, generated signals, decisions, and justifications is recorded in an immutable evidence ledger.
The Ofelia principle: “Everything is configuration, nothing is code.” Adding an LLM provider, a jurisdiction, or a notification channel doesn’t alter the process or the audit trail. Most importantly, for identical alerts, the investigation path is guaranteed to be identical, a point that has become central with local regulators, who now require proof of the investigation process, not just the final decision.
Between the two, the roles are clear: Denodo supplies and governs the data, Ofelia orchestrates the reasoning, and the LLM only ever works with anonymized metadata and summaries and never with raw client data. The AI recommends; the analyst decides and signs off.
The Same Day Described above, Now with Sentinel
It’s 8:30 a.m., and the same $160,000 alert appears. But Sentinel has already taken over in the background: As soon as the signal was detected, specialized agents connected to the entire ecosystem via Denodo, aggregated the client profile and transactional history, queried sanctions lists and negative news (filtering out all false-positive homonym hits), scanned commercial registries to reconstruct ultimate beneficial owners, and mapped the four-person household, all orchestrated by Ofelia according to a predefined, traceable path.
By the time Thomas grabs his coffee, the file is pre-analyzed, consolidated, and sorted. He opens his dashboard, and spends 5 minutes instead of 1.5 hours. Sentinel provides an escalation and filing recommendation with six-axis explainability and a factual confidence score: Thomas sees exactly which document or article triggered the suspicion. He validates the information, exports a sourced investigation report in PDF format (integrity guaranteed by SHA-256 hashing, regulator-ready), and signs off. In less than 10 minutes, the decision is made and documented, and the solution’s memory is enriched for the next investigation.
The Gains, by the Numbers
- 90% estimated processing time saved on every AML alert
- 47 minutes reduced to 5 minutes of analyst time per alert
- 1.5 hours reduced to less than 10 minutes to investigate an interconnected 4-person household
- 6 axes of explainability and a factual confidence score for every recommendation
- Time-stamped, tamper-proof audit trail (SHA-256), compliant with local regulatory and AI Act expectations
- Zero data copying: In-situ processing within the client’s IT system, systematic anonymization before any LLM call
- Transferable gains beyond AML: Up to a 50% reduction in processing times for KYC/onboarding, with similar use cases in collections, customer complaints, or regulatory reporting
A Progressive Trajectory, Not a Big Bang
This shift isn’t achieved in a single sprint. Clever Age supports it through a four-level maturity methodology:
- A flash diagnostic across nine pillars to map priorities
- Building a single foundation of truth with Denodo by progressively virtualizing critical sources
- An iterative deployment of targeted modules (e.g. KYC Data Hub, AML Alert Optimizer, Sanctions Screening Enhancer, and Regulatory Reporting Fabric).
- Change management to scale adoption across the organization
- Every step is governed, traceable, and reversible
Sentinel in Action
To see Sentinel at work in an anti-money-laundering scenario, watch this short video.
