Don't just search. Investigate.

The investigation begins by working out who you actually mean. It takes the name plus whatever discriminators you have — country, registration number, year of birth, address, website, a known alias — and resolves the entity before any research starts, collecting transliterations and alternative spellings along the way. If the identity stays genuinely ambiguous, the report splits the candidates or stops for manual review. It will never silently merge two people or two companies that happen to share a name, because a dossier that quietly blends two lives is worse than no dossier at all.

Before the open-web research begins, the resolved entity is screened against the official sources the platform already covers: sanctions lists across every major jurisdiction, criminal and wanted-person watchlists, and the dedicated PEP domain. These results are deterministic and come from the registers themselves — the AI layer can explain a retrieved match, but it can never invent one or substitute its own judgment for the source. The exact sources screened and the point-in-time timestamp are recorded in the report, so the screening section of the dossier is evidence in its own right.

This is what separates Deep Research from a search. Specialist research threads run in parallel across adverse media, regulatory and enforcement records, court and corporate filings — and they re-plan as they go. A lead that turns up nothing is dropped; a lead that turns up something is pursued. Coverage is multilingual by design, because enforcement action and investigative reporting on a cross-border counterparty are usually published first, and often only, in the local language. The investigation reads sources in the language they were published in rather than waiting for an English summary that may never come.

Risk rarely sits on the name you typed. Deep Research builds a broad map of the network around the subject: parent and subsidiary companies, direct and indirect owners, directors and officers, controlling customers and material associates. A dedicated ownership specialist works the control graph — resolving direct versus indirect holdings, percentages, effective dates, historical changes and conflicting cap tables — while adverse-media and commercial-risk specialists extend the same graph from their own angles. Every finding is merged and deduplicated into one ledger, so the network keeps growing as the investigation runs rather than stopping at a fixed number of hops.

Discovering a connection is only half the job — the other half is checking it. When the investigation surfaces a company or a person who was not in your original query, that party is written into the report as a named screening referral, with the evidence that put it there, rather than being buried in a paragraph. Your analyst can run it through the platform's sanctions, PEP and criminal-watchlist screening as its own check. This is what turns a network map into an actionable worklist: an owner two steps behind your counterparty arrives on your desk as a name to screen, not as something you had to notice yourself.

Findings are reconciled against each other before the report is written. Every material claim lands in an evidence ledger tying it to the sources that support it, and where independent sources disagree the contradiction is reported as a contradiction — not averaged away into a confident-sounding sentence. Confirmed facts, reported allegations, and unverified claims are kept visibly separate, so a reviewer can see at a glance which parts of the file would survive a challenge and which parts are still someone's assertion.

The report closes with what it could not establish: unresolved identity questions, jurisdictions with no accessible public register, threads that ran out of sources — and the specific follow-ups a human should take. Most tools present absence of evidence as a clean result. Naming the gaps is what makes the rest of the document trustworthy, and it tells your analyst exactly where their time is worth spending. The whole dossier is stored as point-in-time evidence and exported as a PDF in the report language you chose.