In late 2025, a specialized media outlet published an article about a KYB vendor whose internal review found that its UBO product had a failure rate of more than 50%.
The product was marketed as a tool that automates beneficial-ownership analysis using live shareholder data from official company registers. Enter a company name, set an ownership threshold (say, 25%), and the product traverses the shareholder chain to compute the natural persons who ultimately control the entity.
The vendor serves financial organizations globally, including major banks, and operates partnerships with many KYC platforms. This situation illustrates structural problems.
“UBO data” can mean several things.
The vendor’s UBO product computes ownership from shareholder data rather than relying on UBO declarations filed at registers. At Topograph, we call this a “derived UBO,” distinct from a “declared UBO”. The approach is valid. Derived UBO is a necessary fallback when UBO registers are closed, restricted, or unreliable, and a valuable cross-check in countries where UBO registers are available.
Shareholder data itself can come directly from an official source: registers in Germany and Italy, for instance, publish shareholding data for several company legal forms. Shareholder data can also be inferred or computed by reading company filings, extracting information, and piecing it together with (AI-enabled) analysis.
The vendor’s own review revealed a high failure rate in shareholder identification, but clients had no way to learn about it. They got a polished output from a system they couldn’t inspect. Knowing where data came from is a necessary first layer. But even when you can trace data to a register source, you still need to know whether the extraction and computation layer actually worked.
Transparency is as important as data quality.
A derivation engine that can’t complete an ownership chain has two options: report the failure, or return an incomplete result as if it were complete. The Open Ownership / Tax Justice Network analysis of the UK PSC register found roughly 20% of entities have no declared individual beneficial owner, and some individuals appear as PSCs of over a thousand companies (patterns consistent with nominees and shell structures). Derivation products will struggle with these. But reporting failures honestly would still be useful, since the compliance team can then manually verify.
When evaluating a data provider, for any derived UBO, ask for the complete shareholder chain, the source document for each link, and the computation logic that produced the result. If the output is a name and a percentage with no traceability, your compliance file has a hole. Ask what happens when the derivation fails: does the product tell you, or does it quietly return an incomplete result as if it were complete? Ask whether the provider distinguishes between declared UBOs and derived UBOs, since they carry different evidentiary weight, and mixing them into a single unlabelled field makes proper documentation impossible.
How Topograph handles the transparency challenge.
At Topograph, when you buy the “UBO” datapoint, you only get information sourced from an official register, on demand. So there’s no doubt about what we deliver from this endpoint.
We also provide a separate “ownership graph” feature that traverses company ownership trees (multi-levels and multi-countries) to surface candidate UBOs based on capital accumulation calculations. Each node in the graph, a natural person or legal entity, is tagged with the source it came from and a taxonomy: direct-from-register, cached-from-register, derived, inferred. When the ownership percentage is AI-inferred, we cite the exact document(s) and rationale, and we flag that this is indeed AI-inferred and should be cross-checked for compliance.
We don’t claim to solve every case. Not all registers publish shareholder data. Some ownership structures involve nominee arrangements or control by other means that no automated tool can reliably detect. We’d rather tell you we can’t complete the chain than hand you a plausible-looking result and hope you don’t check.
For a detailed coverage of a related trust issue in company data, see our coverage of The Proof Gap in KYB.