Skip to main content

Data / Methodology

How Veridion builds company data

Veridion operates the full pipeline in-house, from crawling and extraction through resolution, validation, and delivery. Every attribute carries three outputs: the result, a confidence score, and the source it was derived from.

The pipeline

Every record moves through four stages

Collection, resolution, enrichment, and validation. Each stage runs continuously and adds signal before the next begins.

Sources

Where the data originates from

Veridion's primary source is the deep web. Trade registries anchor legal truth. Commercial web and licensed partner feeds corroborate. The pipeline then manufactures additional data points on top of this verifiable foundation.

Source network verified: 2 hours ago
  • Company websites

    6
  • Trade registries & filings

    14
  • News & press

    68
  • Map pins

    3
  • Social media

    8
  • Satellite images

    4
  • Patent databases

    Integrating8
  • Financial data

    Integrating8

119+ sources across 6 live categories, processing 1.2B+ records daily. Additional clusters are integrating.

Scope

Public, structured, legally-obtained data only

Every attribute is derived from public sources and verified before delivery, never raw content rehosted verbatim.

Scope of collection

  1. 01

    Robots.txt compliant

    If a site disallows crawling, Veridion does not crawl it.

  2. 02

    No personal data

    Personal contacts, biometric data, and facial imagery are out of scope.

  3. 03

    In-house infrastructure

    Crawl, storage, models, and delivery are all operated by Veridion.

What ships with every attribute

  1. 01

    Result

    The structured fact, normalised to a controlled taxonomy.

  2. 02

    Confidence score

    Every attribute ships with a confidence score to allow for full control in large scale pipelines.

  3. 03

    Source trail

    The sources the attribute was derived from, plus the last-harvested date.

Change detection

The graph updates continuously

Core company profiles refresh continuously. Volatile signals refresh daily. Technographics use a rolling 90-day window.

Operational shifts

HQ relocation detected via registry update, old: Munich, Leopoldstr 42 → new: Berlin, Friedrichstr 191.

Product launches

New product line identified from website, SolarTech GmbH adds battery storage to offerings catalog.

Ownership changes

Acquisition signal from filing, EnerVolt AG acquires 100% of SolarTech GmbH effective 2025-Q3.

Validation

Every record is checked before delivery

Every value is checked for internal consistency and corroborated against independent sources before it ships.

  • Cross-source corroboration. An attribute is trusted in proportion to the number of independent sources that agree on it.
  • Consistency rules. Every record must satisfy hard cross-field constraints: a registration number has to match the checksum and format of the jurisdiction that issued it, an address has to resolve to coordinates inside the stated country and to a site that can physically host the operation it claims, and derived figures like revenue-per-employee have to fall within the plausible band for the company's industry.

Re-evaluation

Intelligence improves as new evidence appears

Conflict resolution

When sources disagree, the system escalates the decision to higher-order intelligence that weighs recency, authority, and the full evidence trail before settling on a value.

Restrictive to change

Established values are hard to move. Because models can be easily swayed, any change to an existing attribute must arrive with supporting evidence and clear a thorough set of filters before it is accepted.

Attribute decay

Stale signals are flagged and deprioritized over time.

Revision history

Every conclusion change is traceable to the evidence that caused it.

Limitations

What the data does not cover

  • Non-operational legal entities (holding companies, SPVs) may not carry operational attributes like employee count or revenue. Values are omitted rather than guessed.
  • Sparse-data regions and micro-companies ship at lower confidence. The score reflects this, and callers can inspect the threshold.
  • Private companies that do not disclose financials return modelled values, and the label says so.
  • Material signals update in near-real-time. Lower-priority changes may take days or weeks to appear in sources.

Methodology

Questions about how the data is built?

Our data engineers can walk you through source hierarchy, freshness cadence, and match logic for your specific use case.