LUMA – Contextual AML Screening and Monitoring Agent
LUMA is CleverChain’s autonomous AML Screening and Monitoring Agent which moves AML screening and monitoring from name matching against static watchlists to contextual risk analysis, and works with or without the client’s existing list providers. LUMA screens individuals and legal entities across languages and scripts, at onboarding and through event-triggered and scheduled monitoring, and delivers each alert with its match rationale, cited sources and a timestamped audit trail.
Core functionality
End-to-end AML contextual screening and ongoing monitoring (sanctions, PEP, adverse media, reputational risk, litigation and enforcement) for individuals, legal entities and their connected parties
Intended use case
Customer onboarding (physical and legal entities); Ongoing monitoring, event-driven and periodic reviews, on-demand re-screening of a customer cohort
Typical customer profile
Financial institutions and other AML-regulated firms (banks, insurers, payment companies), industrial companies
Typical workflow it supports
Onboarding, adverse media checks, ongoing monitoring, periodic and event-driven reviews
Due diligence that understands the business.
What it does.
LUMA evaluates sanctions, PEP, enforcement and adverse media signals in context (geography, sector, role, chronology and network), in real-time and with multi-attribute entity resolution across languages, scripts and transliterations. The output is a graded, dated, cited narrative that includes positive and exculpatory evidence, not a hit or no-hit flag. LUMA also builds intelligence on entities for which no watchlist entry exists and can be configured to assess reputational factors that cannot be reduced to a list or table.
Scope of screening.
Beyond sanctions and regulatory lists, LUMA covers PEPs with their family members and close associates; litigation and enforcement; corporate registries, leaked-document databases, disclosures and filings; adverse and reputational media in the language and script of publication; professional background and source of wealth; and the network of directors, controllers, beneficial owners and related parties, so that risk attached to a connected party surfaces even when the screened name is clean. Where evidence is missing or uncorroborated, the alert says so.
What each alert contains
The match rationale (which attributes matched, which did not, and why the match was kept or discounted); a risk grade with dated, cited sources and a chronology of events; positive and exculpatory evidence alongside adverse findings; information gaps stated explicitly; and a timestamped audit trail of sources queried and decisions taken, ready for the CDD or EDD file.
What industry problems, challenges or unmet need does it address
LUMA addresses the most structural problem in AML screening and ongoing monitoring, which is related to the limits of static watchlist screening. Name-against-list matching is necessary, but it is not an investigative outcome, and two of its weaknesses are structural. First, a watchlist is a restricted universe compiled by design: the provider's analysts decide which entities merit a high-risk profile, so the list is only as complete as that editorial process allows. Risk that no analyst has profiled is invisible to every institution that relies on the list. That is a false negative by construction, not a data-quality lapse. Second, the profiles themselves are static. A high-risk profile built on, for example, a name and a past offence is a problem for the user, who cannot disambiguate it or interrogate its basis, and it is equally a problem for the provider, who is not positioned to keep it current. The profile ages and becomes stale. Watchlist screening also cannot read context: geography, sector, role, chronology, or network.
Two further gaps follow from the same model based on static watchlists. Adverse media checks are often run by typing a name and risk terms into a search engine, by hand or through a search API; the analyst then has to work out which of many namesakes each story concerns, whether it is relevant, and how to evidence the decision, using results whose provenance the institution does not control. Meanwhile, ongoing monitoring is typically calendar-driven, with periodic review cycles treating a live business relationship as a series of snapshots, so risk that emerges between reviews goes undetected until the next scheduled date, while alerting on every registry update floods review queues with changes that do not alter risk. Regulation is moving towards event-driven review: the EU AMLR (Regulation (EU) 2024/1624, Article 26) sets maximum intervals for updating customer information from 10 July 2027, and the AMLA draft guidelines on ongoing monitoring, consulted on between June and September 2026, list changes in identity, ownership or legal status, adverse media and PEP status among the triggers for review.
THE LUMA DIFFERENCE
What sets our innovation apart.
01 — Entity resolution
Multi-attribute resolution (name, date of birth, role, address, corporate identifiers) combined with exact and fuzzy matching, cross-language transliteration and alias consolidation, linking the same entity across sources, jurisdictions and scripts.
02 — Contextual, scenario-based discounting
Each sanctions, PEP or adverse media signal is re-scored against context and network, suppressing look-alikes and surfacing genuine risk that list logic alone misses. A finding confirmed by independent sources is weighted above a single uncorroborated mention, contradictions between sources are recorded, and positive or exculpatory evidence lowers residual risk.
03 — Policy-wiring and audit layer
The user’s policies, data blocks and regulatory mapping are pre-wired into the workflow; every step is logged and timestamped; and each report concludes with a control checklist mapping requirements to the evidence collected, with completion status per item.
04 — Materiality-based continuous monitoring
For legal entities, LUMA monitors registry and corporate status, nature of business, ownership, beneficial owners and controllers, sanctions and enforcement, and litigation and adverse media; for natural persons, role and network changes, multilingual adverse media, sanctions and watchlists, and PEP status with family members and close associates. Monitoring is both event-triggered and time-based, configurable by segment, jurisdiction and product, and available on demand for a single case or a cohort.
05 — Adverse media with provenance you can file
In place of search-engine results read and disambiguated by hand, LUMA starts from the entity and the institution’s policy, retrieves primary and open sources under governed crawling with data minimisation and retention limits, resolves the entity across languages, weighs each signal by geography, role, chronology and network, and returns a graded, dated and cited finding that can be filed as evidence.
IMPACT
Qualitative benefits
· Calibration to policy. Thresholds, risk weights, escalation rules and review frequency follow the institution's policies, risk framework and risk appetite, and the configuration is documented for testing, consistent with the Wolfsberg Group's Guidance on Sanctions Screening (2019).
· Fewer false positives and fewer blind spots (false negatives). Namesakes are discounted with a recorded reason, while network reasoning and sources beyond lists surface risk that no watchlist holds.
· Defensible alerts. Every alert carries its match rationale, cited sources and a timestamped audit trail, so a supervisor or internal audit function can follow the decision.
· Monitoring proportionate to risk. Changes are routed by materiality, so analysts review what alters a customer's risk while the rest is updated silently with an audit entry.
· Regulatory readiness. For example, in Europe LUMA supports ongoing-monitoring obligations under UK MLR 2017 regulation 28(11) and (12) and EU AMLR Article 26 (from 10 July 2027), and the event-driven reviews set out in the AMLA draft guidelines on ongoing monitoring.
Quantitative benefits
Based on client feedback, CleverChain estimates that contextual screening removes around 90% of false positives. Where a round input is available, comprising for example of entity details and corporate associations, the reduction in false positives is almost complete. For perspective, CleverChain does not review what clients screen, so the figure comes from client-reported results.