Counterparty due diligence for business onboarding
Built for the AML / onboarding team at a European neobank. An analyst types a company number; the agent pulls registry data, screens sanctions and PEP lists, runs adverse-media search, and cross-validates every declared director across LinkedIn, the corporate site, and OpenSanctions. Output: a dossier with verdict and reasoning, ready for the MLRO review queue. Customer name withheld — try the public walkthrough on three mocked companies.
What the agent owns
- Registry sync across Companies House, CVR, Brønnøysundregistrene, GLEIF + screening against EU FSF, OFAC SDN, UN, UK HMT, Swiss SECO
- Adverse-media relevance gate drops tangential hits before they reach the dossier
- Director-identity scoring 0–100 across registry + corp site + LinkedIn
- Single dossier surfaces RED / AMBER / GREEN with full evidence trail and source-cited triggers
The story
Before the deployment, the onboarding team was working from a checklist of seven manual data lookups per case — registry, sanctions, media, director LinkedIn, beneficial-owner trace, corporate-site sanity check, MLRO note. A single complex case could swallow an analyst's morning.
The agent now owns the lookup phase end-to-end. Analysts step in only at the dossier-review step: read the verdict, accept or annotate, push to the MLRO queue. The pipeline is deterministic where it can be (registry + list matching) and agentic where it has to be (director identity cross-validation, adverse-media relevance gating).
Every signal carries a source URL and a one-line reason — the MLRO can challenge any line item by clicking through to the underlying registry filing or news article. There is no "black box" verdict.
Impact
Verdict on a typical case in under 90 seconds of agent time. Analyst review on the agent's dossier averages 3–4 minutes against a previous full lookup-and-write workflow that ran 30–45 minutes per case.
The module the team leans on most is the director-identity verifier — it scores every declared director against LinkedIn, the corporate leadership page, and OpenSanctions, with explicit reasoning per signal. Sanitised demo workspace available on request.
What it takes to launch
- Read access to your onboarding queue (CRM, ticketing, or a shared inbox)
- Your MLRO's risk thresholds — what gets auto-passed, what needs review, what gets escalated
- A nominated AML/compliance lead for the first two-week feedback loop
- Pre-wired registry / sanctions / PEP / adverse-media data pipelines
- Dossier templates tuned to your queue format, in your branding
- Audit trail with source citations on every signal — exportable for the regulator
Multilingual lead-gen for property agencies
WhatsApp and web leads answered in seconds, in the buyer's language, with matching listings attached.
Category-manager copilot for multi-supplier coordination
One workspace, one persona per supplier. The copilot tells the category manager what to chase next.
Candidate qualifier and screening copilot for HR agencies
Inbound CVs and applicants get screened, scored, and routed before a recruiter opens the inbox.