D&B's identity-anchored AI needs match-level receipts

Make every resolved identity reproducible

The direct buyer answer is that an identity-anchored output still needs a match receipt. Preserve the third party exactly as submitted, including name, address, registration number, tax identifier, website, country, contact, source system, relationship role, and observation time. Then retain the candidate entities returned, matched and conflicting fields, normalization applied, match points or other reasons made available, selected D-U-N-S Number, rejected candidates, confidence or uncertainty, rule or model version where known, reviewer, override, and decision time.

A stable identifier can connect records, but the assignment can still be wrong, stale, or too broad for the intended relationship. Similar names can identify different legal persons; a brand can span parents, subsidiaries, branches, and trade styles; an address can be shared; and a company can merge, reorganize, dissolve, or change ownership. The workflow should permit unresolved, needs-documentation, confirmed, superseded, and corrected states rather than silently merging every candidate into one clean profile.

Version the graph and the delivery route

D&B describes one commercial graph delivered through embedded products, MCP connectors and agentic workflows, APIs, batch files, and warehouse feeds. Buyers should record which product or endpoint supplied each field, request and response identifiers, query terms, fields requested, graph or dataset timestamp if available, schema and connector version, pagination, filters, response status, transformed fields, delivery time, and the downstream record that consumed the result. A later profile should supplement or supersede the earlier observation without rewriting what an earlier decision actually saw.

Relationships need their own provenance. For a parent, subsidiary, branch, beneficial owner, or other affiliation, preserve the two entity identifiers, relationship type, direction, percentage or control basis where supplied, effective and observed dates, source category, confidence, conflicts, and correction history. A graph edge can focus diligence; it does not by itself establish legal ownership, control, sanctions attribution, contract party, service dependency, or responsibility for another entity's conduct.

Keep identity, signal, and decision separate

The page places identity verification beside AML, sanctions, anti-bribery and corruption screening, due diligence, risk scoring, and continuous monitoring. Those stages answer different questions. An identity match establishes which candidate record is being considered. A list or risk signal states what a dated source reports about that subject. The organization still has to determine the applicable relationship, jurisdiction, policy or authority, ownership rule, transaction or service scope, evidence needed, escalation path, and accountable action.

For every screening or monitoring event, link the resolved entity receipt to the exact list, source, publication or retrieval time, matched field, signal version, prior observation, analyst disposition, false-positive rationale, escalation, restriction or condition, affected engagement, and downstream acknowledgment. Reducing false positives, accelerating onboarding, or supplying near-real-time alerts remains provider positioning until tested on the buyer's authorized population with declared measures and independently reviewed errors.

Test identity drift before trusting automation

A representative evaluation should include two same-name companies in different countries, a parent and subsidiary sharing a brand, a branch with an old address, a recently acquired supplier, a beneficial-owner change, a misspelled identifier, a sanctions candidate with partial field agreement, and a previously accepted match that is later corrected. Run the cases through each proposed delivery route and require consistent entity selection or an explained difference, visible uncertainty, source-to-field provenance, versioned corrections, and a reversible downstream impact review.

Dun & Bradstreet's official page supports the attributed positioning about identity-anchored data, the D-U-N-S Number, entity resolution, ownership and relationships, screening, monitoring, alerts, embedded AI, MCP connectors, agentic workflows, APIs, files, and warehouse feeds. It does not independently establish population coverage, source correctness, match accuracy, ownership, list scope, latency, model performance, legal applicability, third-party suitability, or an operating outcome. Procurement, compliance, sanctions, anti-financial-crime, risk, data, security, audit, business, and legal owners retain those judgments.

What we will watch next

Third Party Current will watch for later primary-source evidence that changes the maintained company, capability, or standards record. The next useful evidence may include implementation documentation, release details, regulator findings, corrected methods, product packaging, customer-observable workflow, or a subsequent company statement. Until then, the dated source and its stated boundary remain attached to this analysis.