Missing company registry data is not one problem but three. The register may not publish the field at all, the search may have found no record, or the source may have failed to answer. Record each case with its own label, because each one supports a different next step, and none of them is evidence that a company is legitimate or risk-free.
This guide is for compliance, procurement and finance teams who verify companies, and for developers building those checks into agent workflows. It shows how to tell the three gaps apart, what to record for each one and how to keep a gap from turning into a silent pass.
Missing company registry data has three causes
A blank cell in a verification file hides the reason it is blank. Before anyone decides what the blank means, name its cause.
- Not published: the source exists and answered, but it does not hold or disclose this field for this kind of entity.
- No record found: the source answered, and the search returned nothing for the inputs you sent.
- Source failure: the source did not answer properly, because of an outage, a timeout, rate limiting or an error.
Each label points to an action. A field that is not published needs another source or the counterparty's own evidence. No record found needs the inputs and the jurisdiction checked. A source failure needs a retry, not a conclusion.
The labels also protect the reviewer. A file that says "ownership not published by this register" reads very differently from a file where the ownership column is simply empty.
Empty cells. Wrong conclusions.
Most verification sheets flatten all three cases into one empty value, or into a default such as "no", "false" or "invalid". That is where the errors start.
A common pattern looks like this:
- A VAT check times out and the tool records the number as invalid, so a legitimate supplier is blocked.
- A name search returns nothing and the file says "not registered", when the company is registered in another country.
- A register does not publish shareholders and the ownership field is left blank, which a later reader takes to mean no owners of concern.
- A sanctions screen is never run because the registry step failed first, yet the screening column shows no match.
None of these is a data problem in the source. They are recording problems, and they make the file impossible to defend when someone asks how a decision was reached.
Know what each source covers
Before running checks, write down what each source can return for the entities you deal with. Official documentation usually says where coverage ends.
Companies House is a good example. Its guidance on searching the register says the free service holds the records of companies dissolved since January 2010, and that dissolved records used to be removed after six years. It also states that the service is not intended to be comprehensive and that many details relevant to particular circumstances have been omitted. In May 2026, Companies House announced a review of the 20-year retention period for dissolved company records and paused their destruction during the review.
Ownership is the field most often missing for legal reasons. On 22 November 2022, the Court of Justice of the EU ruled invalid the rule that beneficial ownership information in the EU had to be accessible to any member of the general public in all cases. What you can retrieve now depends on the member state and on your own entitlement to access.
The mandatory inputs for a registry check are short: the legal name or registration number, and the country. Optional inputs, such as a former name, an address or an LEI, help resolve a no-record result but should not be required to start.
Sources answer. Labels explain.
The technical goal is to keep the source's own answer intact, then attach a label your reviewers can act on.
Some sources already separate these cases. The EU's VIES web service, in its published service description, returns faults such as MS_UNAVAILABLE when the member state application is not replying, MS_MAX_CONCURRENT_REQ when too many requests are reaching that member state, and TIMEOUT when no reply arrived in time. Those are source failures. They say nothing about whether the VAT number is valid, so a workflow that maps them to invalid is recording the wrong fact.
GLEIF goes further and explains missing data inside the record itself. When an entity does not report a parent, the Level 2 reporting exceptions format gives the reason, including NATURAL_PERSONS, NON_CONSOLIDATING, NO_KNOWN_PERSON and NON_PUBLIC. An exception is a reason for absence, not a statement about who owns the entity. GLEIF's LEI data format also defines a LAPSED registration as one not renewed by its renewal date, so a lapsed LEI is a freshness issue rather than a sign the company has closed.
For every check, keep these fields together:
- The source, the inputs sent and the retrieval time.
- The source's raw status or fault code, unchanged.
- Your gap label: returned, not published, no record found or source failure.
- Whether the item is mandatory or optional under your policy.
- The next step, and who owns it.
Keep matching and interpretation separate from the record. A fuzzy name match that links a no-record result to a similar company is an algorithmic suggestion, and an AI summary of the gaps is an interpretation. Neither should overwrite what the source returned.
One label. Then a next step.
Implementing this is mostly a matter of rules that run before anyone reads the file.
- Map each source's statuses and faults to the four labels, and test the mapping with real responses, including errors.
- Retry source failures a bounded number of times with a delay, then leave them labelled as failures rather than converting them.
- For no record found, re-check the identifier format, try the registration number instead of the name, and confirm the jurisdiction.
- For fields not published, look up whether another official source holds them and whether you are entitled to use it.
- Block an automatic approval when a mandatory item is anything other than returned, and route it to a reviewer.
- Re-run open items on a schedule, so a temporary failure does not stay open forever.
A fictional example shows the labels at work. Fenmoor Tooling Ltd gives a UK company number and an EU VAT number. The register returns the company as active, so that row is returned. The VAT check fails with MS_UNAVAILABLE, so it is a source failure and is retried later. The register names a parent company incorporated outside the UK as the person with significant control, but the parent's own owners are not published in any source the team can access. That row is not published, and the reviewer asks the supplier for a declaration. Nobody approves the file until the mandatory rows are resolved.
There is a trade-off. More labels mean more items in a review queue, and some teams will be tempted to collapse them again. The cost of collapsing them is paid later, when an empty field is read as a clean one.
How we approach it
We built Apiosk so that a missing record stays visible. When a source has no record, the answer says what is missing rather than guessing, and each answer names the sources it used. You can see which registers, sanctions lists and public records are connected on our sources page, and coverage depends on the country and the source.
Company verification returns the legal name, number, registered address, status, registration date and directors on file. Beneficial ownership comes from available official filings, so where a register does not publish ownership, the gap is reported rather than filled. Identifier validation checks VAT numbers, IBANs and company numbers against the issuing source.
Before anything runs, you see the sources and the maximum cost and approve them, as described on our pricing page. For items that stay open, recurring checks can repeat an approved check on your schedule and within a budget. Apiosk retrieves and combines the evidence; a person decides, or your own rules do.
For the dates attached to each record, see our guide to company registry data freshness. For how gaps fit into a full onboarding file, see the supplier onboarding evidence checklist.
Blank fields in your files? Label them.
If your verification sheets cannot tell a failed source from a missing record, see the checks Apiosk runs and make every gap say why it is there.
This article was prepared with AI assistance using the linked Companies House, European Commission, CJEU and GLEIF pages and the current Apiosk product pages. It is not legal advice, and Fenmoor Tooling Ltd is fictional.
Frequently asked questions
Does "company not found" mean the company does not exist?
No. It means the search you ran, against the register you chose, returned no record at that time. The identifier may contain a typo, the company may be registered in another jurisdiction, or the record may sit outside the coverage of the service you searched. Record the search inputs and treat the question as open until you have checked those causes.
Should a failed VAT check be treated as invalid?
No. The VIES web service reports technical faults such as MS_UNAVAILABLE, MS_MAX_CONCURRENT_REQ and TIMEOUT separately from a valid or invalid answer. A fault means the national database did not answer your request, so record it as a source failure and retry later. Only a completed check that returns invalid tells you something about the number.
What should we do when a register does not publish ownership?
Record the field as not published by that source, with the source name and date. Then follow your own policy, which may mean requesting a declaration from the counterparty, using another official source you are entitled to access or escalating to a reviewer. In the EU, access to beneficial ownership registers varies by member state, so check the rules that apply to you.
Can an AI agent decide what a missing field means?
An agent can label the gap, retry a failed source and suggest the next check. It should not infer that a company is clean because a field is empty, and it should not fill the field with an estimate. The decision about whether the file is sufficient belongs to the person or rule your policy names.
