Methodology
How we score a company — and how to read it
Every company on Vereon carries two scores. They answer different questions, they can point in opposite directions, and the moment they disagree is usually the most interesting thing on the page. This is how they are built, how to read them together, and — just as important — what they are not.
The mental model: two questions, two scores
Most providers give you one number. We think a single number hides more than it reveals, so we keep two separate scores and never blend them into one.
Financial Health Score
Backward · today“How healthy is this company, on its latest accounts?”
A composite of long-established, independently peer-reviewed models of corporate financial distress — replicated and stress-tested across decades of academic literature. Computed from a company’s filed financial accounts and expressed 0 to 100, higher is healthier. It needs at least one set of filed accounts; where none exist, it is withheld, not estimated.
Early Warning Score
Forward · 12 months“How likely is serious trouble in the next year?”
A forward-looking estimate of the likelihood a company enters formal insolvency within twelve months, from a proprietary machine-learning model that reads a broad range of behavioural and financial signals — not any single ratio. Shown as a 0 to 100 risk percentile, higher is riskier, always alongside its band.
How to read them together
The two scores deliberately run in opposite directions. A strong company is a high Financial Health Score and a low Early Warning Score. That is not a contradiction — one looks back at what has been filed, the other looks forward at what tends to come before insolvency.
- Read the band first. The Early Warning Score is never shown without its band chip (Very Low → Critical). The band is the honest instrument; the percentile only separates companies within a band.
- Watch the tail with the base-rate line. Because formal insolvency is rare, we also show risk as a multiple of the population base rate (“× base rate”) — the honest way to convey how far into the tail a Critical company sits, where the percentile alone saturates.
- Mind the disagreements. When health looks fine but the warning score is flagged — or vice versa — that gap is the signal. The next section maps exactly how often that happens.
Where 5.5M+ companies actually sit
Plot every company with both scores on two axes — financial health across, forward risk down — and four corners appear. The two off-diagonal corners are why one number is not enough.
Clear on both fronts
Healthy accounts and few forward-looking distress signals. The scores agree — the reassuring, most common case.
Thin on paper, low insolvency risk
Weak or sparse accounts — often a micro-entity that files little — but no behavioural signals of approaching insolvency. The financial picture looks worse than the risk warrants.
Healthy-looking, but flagged
The accounts read well, yet the Early Warning Score is Elevated or Critical. This is the risk the balance sheet doesn’t show — and the single strongest reason to read both scores.
Warning signs on both
Weak accounts and elevated forward risk. Both scores point the same way — the clearest case for enhanced due diligence.
Counts are the four corners of the matrix across the 5.5M+ companies that carry both scores; companies in the middle bands are not plotted. The two shaded corners — ~314k healthy-looking but flagged and ~555k thin on paper but low-risk — are precisely the companies a single score gets wrong.
Reading a score in context
How to contextualise a score
A score is a summary, not a verdict. Read it against the company’s size, sector, age and filing history — a young micro-entity with thin accounts is a different animal from an established filer whose numbers have deteriorated. Every profile shows the public-record evidence behind an adverse read, and the date that evidence was assembled, so you can see why a score reads the way it does.
How a company can improve its score
Because the scores are built from public records and filing behaviour, the things that move them are the things any company should be doing anyway: filing accounts and confirmation statements on time, resolving outstanding charges and court or gazette notices, and keeping the register accurate. There is no gaming step — improving the inputs is improving the underlying position.
Who can see a company’s score
Scores are derived entirely from official public registers, regulatory filings and financial disclosures. They are available to Vereon subscribers for legitimate business purposes — credit and risk, sales intelligence, compliance and due diligence — and any company can ask us to review the public data a score rests on.
How well the Early Warning Score actually predicts
We hold the Early Warning Score to out-of-time testing — the honest standard, and a deliberately harder one than most providers publish. The model is trained only on information available up to a cut-off date, then tested on what happened afterwards, on companies and years it never saw in training (held-out years 2023, 2024 and 2025). That removes any chance of the model “learning the answer,” and it mirrors how the score is actually used: judging a future that has not happened yet.
Companies we rate Critical become insolvent at nearly 5× the UK average. And those we rate Very Low become insolvent at half the average rate.
We publish what share of insolvencies we catch and how many companies we flag to do it. Most providers don't. Why do we do this? Because when only 1 in 225 companies becomes insolvent per year, a score that calls everyone safe is 99.6% “accurate”.
| Band we rate | Share of next year’s insolvencies | Share of all companies | Against the UK average |
|---|---|---|---|
| Critical | 25.4% | 5.44%(about 1 in 18) | 4.7× |
| Elevated or above | 33.4% | 9.29%(about 1 in 11) | 3.6× |
| Moderate or above | 39.6% | 16.3% | 2.4× |
| Very Low | — | — | 0.5× |
Out-of-time, on held-out years 2023, 2024 and 2025, scored a year before the outcome. “Against the UK average” compares each band’s insolvency rate with the 0.44% population base rate (about 1 in 225 companies a year). Shares are of the companies we score.
In plain terms: the 1 in 18 companies we rate Critical hold 1 in 4 of next year’s insolvencies, and widening the screen to Elevated and above — about 1 in 11 companies — catches 1 in 3, a full year early. In blind out-of-time comparison the model substantially outperforms simple financial rules-of-thumb and single-ratio screens. We publish these figures rather than a single flattering headline, because the honest measure of a predictive model is how it performs on data it has never seen.
Darkfield: scoring the companies that file no profit figure
96.6% of UK companies never file a profit figure. We score their risk just as accurately. Of 6,414,220 UK companies, 6,195,071 have no profit-and-loss on file.
Darkfield reconstructs the forward risk read from what a company does, not only what it files. It is not a third score: it is how the Early Warning Score reads a company with no profit-and-loss on file, held to the same out-of-time test. Tested on those companies alone, the model keeps its predictive power — the same ≈0.74 out-of-time AUC as the full model, with no measurable loss. How it reads stays private; the accuracy is the proof.
Counts as of 4 Sep 2026. “No profit-and-loss on file” means none of the accounts we hold for the company carries a turnover figure.
What the scores are — and are not
- Predictive opinions, not statements of fact. The scores summarise public evidence and statistical patterns; they do not observe a company’s private circumstances, and they are not verdicts. A verified public-record event (an insolvency filing, a winding-up petition) is a fact; a model score is an opinion about probability. We keep those two registers visibly separate.
- Probabilistic — most companies in every band do not become insolvent. No score predicts the outcome for any individual company with certainty. A highest-risk company is many times more likely to become insolvent than the typical one, but the large majority of companies in every band continue trading normally.
- Not advice, and not a regulated credit reference. The scores are decision-support tools, not credit, financial, investment or legal advice, and Vereon is not a regulated credit reference agency. Always carry out your own due diligence before acting.
- Honest about thin evidence. Where a company has filed little or nothing, we say so on the profile and withhold or qualify the scores rather than guessing. A “Very Low” risk on a dormant company reflects limited evidence of risk — not demonstrated strength.
Corrections, currency and your rights
Company data changes constantly. Scores are refreshed as new filings and events are published, and each profile shows the date its underlying data was assembled. If you believe a score rests on inaccurate public data, or you want to understand the basis for an adverse read, contact us at hello@vereon.co.uk and we will investigate and, where the underlying record warrants it, correct it.
See both scores on a company you know
Pick a company you understand and judge us on what you missed. New accounts open with a 7-day evaluation — seven lookups, no card required.
Start your evaluation →Figures on this page are drawn from Vereon’s live scored population and the model’s out-of-time validation. Last updated July 2026.