Do UK companies show warning signs before they fail?

In our back-test, most UK companies that later failed showed no registry warning a year earlier. What the signals do show, and where they mislead.

Published 21 Oct 2026 · Last updated 22 Oct 2026

Sometimes, but not often in the public registry. In our back-test, 254 of 299 UK companies that later went into liquidation or administration showed no registry warning sign 12 months before. A few signals were more common before failure, but they also appear on healthy companies, so they are a reason to look closer and not a verdict.

This guide sets out what we tested, what we found, and how to read a signal when you meet one.

What did we actually test?

We asked whether the Companies House signals separate UK companies that later failed from comparable healthy ones, using only what the register showed 12 months before the event. The full method and tables are on the back-test page, and the plain-English summary is in the methodology.

• Failed companies: UK limited companies and PLCs whose liquidation (compulsory or creditors' voluntary) or administration began between 2023 and 2025. Members' voluntary liquidations are excluded, because those companies are solvent.
• Healthy companies: one for each failed company, active today, of the same type, in the same industry section, incorporated within a year either side, and evaluated at the same date.
• Two samples: a first, exploratory sample of 100 failed and 100 healthy companies, used to set the first rules. Then an independent validation sample of 299 failed companies and 299 healthy ones, with no company from the first. The figures below come from the validation sample unless we say otherwise.

For each company we rebuilt its profile as it stood 12 months before the event (the reference date), counting only filings delivered by then.

Did most failed companies show a warning sign?

No. Of the 299 failed companies, 254 (85%) were Low overall a year earlier. Another 38 were Medium and 7 were High. Of the 299 healthy companies, 283 were Low, 14 Medium and 2 High.

So the registry was quiet for most of the companies that went on to fail. If you rely on registry signals alone, expect to miss most of them.

Which signals were more common before failure?

Three of the six signals we tested were clearly more common among failed companies in the validation sample, meaning the 95% interval for the ratio excludes 1. The ratio is the share of failed companies with the signal divided by the share of healthy ones.

• Accounts overdue was found on 14.0% of failed companies and 5.4% of healthy ones. In the exploratory sample the figures were 20% against 10%, a ratio of 2.0 with a wide interval (0.99 to 4.05). Splitting by how late the accounts were did not hold up: accounts 90 or more days overdue were 7.0% against 4.0% in the validation sample (ratio 1.75, interval 0.88 to 3.49), so we adopted no lateness threshold.
• A Gazette notice for strike-off was found on 11.0% of failed companies and 4.0% of healthy ones. Of the 33 failed companies with one, 26 had a notice that was withdrawn (discontinued) by the reference date; the other 7 had not (9 of 10 were withdrawn in the exploratory sample). Our guide to a Gazette strike-off notice explains what these notices mean and what to check.
• Repeated registered-office changes appeared on 11 failed companies and no healthy ones, but only 2 against 2 in the exploratory sample, and the interval is very wide. We keep this at Low and will re-check it on a third independent sample.

Which signals told us nothing?

Three signals showed little or no difference in the validation sample, so they are context only. They never count towards the overall level.

A name change was too rare in the sample to judge, which is why it sits here and not with the signals above.

What does a signal look like in a report?

The report on the Diligence page is for a fictional company, Example Meridian Holdings Ltd. It shows four rows, each with its source:

• Accounts overdue (Medium). The latest annual accounts were not filed by the due date. Source: Companies House. This is the one signal in the report that we back-tested.
• Domain much younger than the company (Low). The company has existed for years, but its website domain is new. Source: domain registry (RDAP) and Companies House.
• LEI has lapsed (Low). Many companies let a Legal Entity Identifier lapse when they no longer need it. Source: GLEIF.
• Sanctions screening (Info). No name match on the OFAC SDN list or the UK Sanctions List. Info never counts towards the level.

Overall, the report says Medium concern. Here the level is Medium because there is one Medium signal and Info signals are ignored. In general, one or two Medium signals, or three or more Low signals, make the level Medium; any High signal, or three or more Medium signals, make it High (full rules in the methodology). The level is a way to sort what to read first. The rows behind it are what matter, and each one shows where it came from and when it was retrieved.

Why do healthy companies show the same signals?

Because a signal that is more common before failure is still common among companies that never fail. In the validation sample, 16 of 299 healthy companies (5%) were Medium or High overall a year earlier, and about 1 in 20 had overdue accounts.

There is a second reason to be careful. Our sample is built half failed and half healthy so that we can compare them. Real failure is far rarer than that, so even a signal twice as common before failure will mostly appear on healthy companies. The back-test shows that some signals were more common among failed companies. It is not a model of how likely any one company is to fail.

Is this a prediction or a credit check?

No. Diligence does not predict insolvency, and it is not a credit check or a credit rating. A report lists public signals with their sources, for you to review.

What the back-test does and doesn't cover:

• Back-tested: only the Companies House registry signals above. Their levels are set from that evidence.
• Not back-tested: the website footprint (domain age, email setup, certificates), lookalike domains, sanctions name matches, SEC EDGAR signals and LEI consistency. Those sources show only today's state and cannot be rebuilt for the past. We cannot say whether any of them would have flagged the 254 failed companies that looked clear in the registry. Their levels are cautious starting points.
• Small samples: the intervals are wide, and the validation sample was also used for the registered-office decision.
• Who is missing: failed companies come from those still listed as in liquidation or administration. Companies already dissolved are missed.
• Narrow scope: one lead time (12 months), and UK limited companies and PLCs only. The results describe those companies, not every company.
• Registers can be late, incomplete or wrong. A source can also be unavailable, and the report says how many sources were checked.

What should you do when you see one?

Treat a signal as a reason to look closer, not as a conclusion.

1. Read the filing. Open the company on Companies House and look at what was filed, when, and what it says. A Gazette notice can be read in The Gazette.
2. Ask the company. An overdue filing or a new registered office often has an ordinary explanation that a short question will settle.
3. Use a credit report when you need a credit view. How likely a company is to pay is a different question from the one registry signals answer, and a credit report is the tool for it.
4. Keep the signal in proportion. One Low or Medium signal, on its own, is common among healthy companies.

The checks in Diligence read public sources such as Companies House, GLEIF, the OFAC SDN list and the UK Sanctions List. You can run a free check with the form above or below. Full details are in the methodology.