How First Mile Labs works

Every KYB application starts with authoritative company data. First Mile Labs queries Companies House directly, pulling the registered address, SIC codes, filing history, persons with significant control (PSCs), and active director list — automatically, the moment a company number is entered.

UBO resolution runs recursively: where a PSC is itself a legal entity, the platform follows the chain up through Companies House records until natural persons are identified, ensuring full beneficial ownership disclosure without manual analyst effort.

Registry data is cross-validated against information submitted by the applicant. Discrepancies — mismatched director names, stale addresses, dissolved entities — surface as case flags for analyst review rather than passing silently through the onboarding flow.

First Mile Labs uses OpenSanctions to screen every individual and entity in an application — directors, UBOs, authorised signatories, and the company itself — against more than 200 global sanctions lists, PEP datasets, and financial crime watchlists, including OFAC, UN Security Council, EU, HM Treasury, and the FATF grey and black list.

Fuzzy name matching handles transliterated names, name order variations, and aliases, ensuring that sanctions matches are not missed due to data quality issues in the source application. Configurable match thresholds allow institutions to tune precision versus recall according to their risk appetite.

Every screening run is recorded with a full audit trail: which lists were checked, which candidates were returned, what confidence score was assigned, and what decision the analyst reached. This record is available on demand for regulatory inspection.

First Mile Labs integrates Didit for individual identity verification. When a director or beneficial owner is added to an application, the platform determines — based on your configuration — whether IDV is required. A secure verification link is generated and sent to the individual by email; they complete the document scan and biometric liveness check on any device, in under two minutes, without downloading an app.

Didit returns a structured pass or fail result with supporting evidence: document authenticity assessment, MRZ data extraction, and liveness confidence score. The result is automatically written back to the case and factored into the overall risk score — no analyst needs to chase the individual or manually record the outcome.

The integration is privacy-preserving and GDPR-compliant by design. Biometric data is processed by Didit and not stored by First Mile Labs beyond the verification result and supporting metadata.

First Mile Labs uses Resend for all transactional email. Analysts and compliance staff access the platform via a magic-link login flow: they enter their email address, receive a time-limited link, and are authenticated without a password. This eliminates shared credentials, simplifies offboarding, and produces a clean access log without the operational overhead of a password management policy.

Applicants — the businesses and individuals undergoing onboarding — also interact with the platform through email-delivered links. Document upload requests, identity verification invitations, and application status notifications are all dispatched through Resend, using HTML email templates that are consistent with the institution's brand.

Every email event is logged against the relevant case, creating a timestamped record of when requests were sent, when links were opened, and when actions were completed. This supports the audit trail without requiring separate communication channel logging.

First Mile Labs uses Claude by Anthropic to power its document intelligence layer. When an applicant uploads a corporate document — certificate of incorporation, articles of association, shareholder register, or annual return — Claude classifies the document type, identifies the issuing jurisdiction, and extracts structured data: company name, registration number, incorporation date, director names, share classes, and shareholder details.

Extraction accuracy is validated by cross-referencing the output against Companies House data and the applicant's submitted application form. Fields where Claude's extracted value differs from the registered record are flagged for analyst review, with the source text highlighted. Confidence scores are exposed at field level so analysts understand which extractions are high-confidence and which require a second look.

Claude also powers the case AI review feature: complex cases are summarised in natural language, missing documents are identified, and escalation recommendations are generated with supporting reasoning — allowing analysts to investigate rather than prepare, and reducing the time spent on case assembly from hours to minutes.