PoD®: a 12-month distress probability you can rely on

Credit and risk teams need two things from any risk score: a reliable ranking of which accounts deserve attention this week, and enough warning to act while there are still options. 

Corporate distress has become a persistent feature of the UK business landscape. Insolvency levels in England and Wales have remained at recession-era highs since late 2022, and one in every 199 registered companies entered insolvency in the year to July 2026.

For credit and risk teams running a ledger of a thousand accounts, that means business failure is not a possibility lurking on the horizon. It is already present within the portfolio. That figure describes a handful of customers who will fail this year, sitting somewhere in the book, currently paying to terms.

PoD® is built for that problem. It is a machine learning model that outputs a single figure for any UK company: the probability that it enters financial distress within the next twelve months.

PoD® in action on the Company Watch platform.

What PoD® gives a risk function

For anyone who owns risk policy rather than individual accounts, PoD® does four things at once. 

• It puts every customer, supplier and prospect on one comparable scale, so exposure decisions across credit, procurement and finance are argued from the same number. 
• It moves the point of detection earlier, because the model reads court filings, charges, filing behaviour and group relationships as they happen, which buys the weeks of notice that separate a renegotiated payment plan from a write-off. 
• It makes policy enforceable, since a probability threshold can be written into a credit policy, applied the same way by every analyst, and audited afterwards. 
• And it makes the risk function measurable: you can report expected loss across the book, track how it moves quarter to quarter, and test last year’s scores against what actually happened.

That last point tends to matter most to heads of risk. A score that can be checked against outcomes is a score you can improve your policy around, defend to a board, and use in provisioning discussions with finance. The sections below cover how each of those benefits works in practice.

A calibrated probability

PoD® returns a calibrated probability, shown in the platform as a percentage. A company sitting at 14% means the model puts the likelihood of distress in the next twelve months as a 1 in 7 chance.

That looks like a small distinction from a risk grade or a score band. But in day-to-day credit work it changes what you can do with the output. 

A probability can be multiplied by exposure, which turns a list of risky-looking accounts into an expected loss figure and a defensible order of priority. A £400,000 exposure at 6% carries more potential loss than a £30,000 exposure at 23%, and a graded band will not tell you that. A probability can also be tied to a policy threshold that means the same thing to everyone who reads it. “Refer anything above 4%” is a rule a team can apply consistently and a credit committee can sign off. “Refer anything in band D” depends on how each person interprets band D.

It also makes the score testable. You can look back at everything that sat above 4% last year and count how many entered distress. Very few risk outputs used in UK B2B credit can be checked that directly.

Warning before the accounts show it

The main limitation of purely accounts-based scoring in the UK is timing. A small company’s filed accounts can be 12-18 months old when a credit team is reviewing them. Risk, however, does not stand still. A business can deteriorate rapidly in the period between filing and assessment, meaning traditional scores may reflect where a company was, rather than where it is heading.

PoD® draws on both financial and non-financial inputs that move well ahead of the next set of accounts, and they are updated continuously, for example:

• County Court Judgments, weighted by trajectory as well as volume. A company accumulating CCJs at an accelerating rate is telling you something quite different from one carrying an old, stable count.
• Winding-up petitions. A petition represents a creditor who has run out of patience and other remedies. The threshold for filing one is high, and the signal is correspondingly strong.
• Mortgages and charges. The useful content is in composition and timing. Security on its own says little. A new charge in favour of an invoice finance or asset-based lender, where none existed before, often means cheaper facilities are no longer available.
• Statutory filing behaviour. Persistent late filing of accounts and confirmation statements is a documented precursor to distress. Companies under strain deprioritise compliance.
• Distressed debtor exposure. Company Watch matches a company’s trade debtors against businesses already in administration or liquidation. If a customer’s largest customer has failed, that pressure is real and it is visible long before it reaches their own accounts.

The practical effect is that the score reflects the company’s current condition, and it keeps moving between filing dates.

Risk that a set of accounts cannot show you

Two of the core inputs behind PoD® cover ground that no accounts-only assessment reaches.

The first is group structure. Parent, subsidiary, shareholder and PSC relationships change how a set of accounts should be read. A subsidiary can look adequately capitalised on its own while depending entirely on intra-group funding, and a sound trading company can be pulled under by a distressed parent. PoD® resolves the group and treats those as connected risks. Ownership changes carry signals in their own right, including the pattern where activity moves in succession between related entities.

The second is director history. Company Watch resolves director identities across company records to build individual-level histories, which lets the model use prior association with insolvent entities as a feature.

Explainable risk scores: seeing what is driving the number

A probability with no explanation behind it is difficult to use and harder to defend. PoD® computes Shapley values for every scored company, breaking down the probability into the positive and negative contributions of each input. The output shows which categories are pushing the number up and which are holding it down.

This matters in three specific situations that credit professionals will recognise. Presenting a limit reduction to a credit committee, where “the model says so” is not an answer. Having the conversation with a customer or supplier who wants to know why terms have changed. And building an audit trail that stands up when a decision is reviewed months later.

The break-down is surfaced at category level, above the raw features, which keeps it readable for non-technical users and protects the underlying feature engineering. It is less granular than the coefficient-level breakdown available from a regression model such as the H-Score®, and it is an honest account of what is driving each individual prediction.

What it changes in practice:

• Onboarding and limit setting. A probability gives you a limit calculation you can show the workings for, and a consistent basis for approving marginal accounts on tighter terms.
• Portfolio monitoring. Rank the book by probability against exposure and the review list writes itself. Because inputs update continuously, movement in the score becomes the trigger for review.
• Collections prioritisation. Where two accounts are equally overdue, the one with a materially higher distress probability is the one to chase first. Recovery rates fall sharply once a formal procedure begins.
• Supplier and procurement vetting. The same score works in the other direction, on suppliers whose failure would interrupt production or delivery.
• Portfolio and pipeline screening. Scored at population level, PoD® supports sector-level and segment-level views of where risk is concentrating.

How PoD® sits alongside the H-Score®

The two work harmoniously together. The H-Score® is a regression-based assessment grounded in audited accounts, fully explainable, and well validated over a long period. Where a structured, transparent read of statutory financial data is what a decision needs, it remains the right tool.

PoD® answers a second question: how likely is failure, and how soon. It takes a broader feature set, updates continuously, uses a gradient boosted tree framework that captures non-linear relationships and interactions a linear model will miss, and returns a calibrated probability. Many teams will use both: the H-Score® for the underlying financial position, PoD® for the likelihood and timing of failure.

What counts as distress

Definitions matter when a number is going to drive decisions, so it is worth being explicit. PoD® treats a company as having entered distress if it subsequently enters administration, receivership, a Company Voluntary Arrangement, a Creditors’ Voluntary Liquidation or a Compulsory Liquidation.

Strike-offs and Members’ Voluntary Liquidations are excluded. A solvent company can choose to dissolve itself, and counting that as failure would add noise to the training labels and inflate the score. PoD® is trained specifically on financially driven failure.

Getting started

PoD® officially launches on 21st October 2026, following an extensive beta period that put the model in front of experienced practitioners across a range of use cases and industries. The model will be retrained annually, so the feature set will develop as the underlying data does.

We surveyed our beta users on how PoD® performed against their existing process and published the findings in full. It sets out what credit and risk teams did with the probability day to day, and where it changed a decision they would otherwise have taken differently. If you are weighing up where PoD® fits in your own workflow, this is a useful read, because it comes from peers doing the same job.

If you already use Company Watch, PoD® sits alongside the H-Score® on the companies you are scoring today. The most useful first exercise is to run it across your existing ledger and see where the new probability rating adds additional value to our risk assessments.

FAQS

How effective is predictive distress scoring compared with a standard credit reference agency score? 

Most agency scores are built primarily from filed accounts and payment data, and are refreshed when those inputs change. A predictive distress model adds court-filed and behavioural signals, updates continuously, and outputs a probability that can be tested against outcomes.

Company Watch’s model of this type is PoD®, trained on financial accounts alongside CCJ data, winding-up petitions, charges, statutory filing behaviour and director-level histories. The gain comes from that additional data and the update frequency as much as from the model architecture.

How do companies monitor the credit risk of other businesses? 

In practice, through a combination of a risk score, alerting on adverse events such as CCJs, petitions and charges, and periodic manual review of the largest exposures. The efficient version replaces most of the manual review with a continuously updated probability, reserving human attention for accounts where the number has moved or the exposure is large. In Company Watch, PoD® provides that probability across a monitored portfolio, with the adverse event data that moved it visible on the same record.

What checks should be done before lending to a company or awarding a contract?

Verify the entity and its directors, read the most recent accounts alongside the filing history, check for CCJs, winding-up petitions and recently registered charges, resolve the group structure, and look at the exposure the company itself carries to distressed customers.

PoD® by Company Watch takes all of those inputs and summarises them into one probability, and the Company Watch record behind it holds the underlying detail for anyone who needs to see the working.

How can I check a UK company’s financial stability quickly? 

Search the company in Company Watch, read its PoD® probability alongside the H-Score® and the factors driving each, then switch on monitoring so material changes reach you without another manual check. That sequence takes a couple of minutes and covers the ground a manual review of the accounts and filing history would take an hour to reach. You can find out more about checking a UK company’s financial stability by monitoring key warning signs below.

Signs a UK company is heading for failure

Which services offer transparent and explainable risk scores? 

Explainability varies considerably between providers, and many publish a score with no account of what produced it. Company Watch publishes the H-Score® with a full coefficient-level decomposition, and PoD® with Shapley-based attribution showing which categories of evidence are raising or lowering the probability on each individual company. Both are designed so a credit or procurement decision can be explained to a committee, a customer or an auditor.

How can I proactively manage business credit risk? 

Score the whole ledger, including accounts onboarded years ago. Rank by probability against exposure, act on score movement, and monitor your customers’ own customers where concentration risk is high. 

Can PoD® be used for supplier and supply chain risk? 

Yes. Supplier failure is the same event, assessed from the other side of the transaction, and procurement and supply chain teams use PoD® in the same way credit teams do. It also picks up the contagion case where a supplier is exposed to a distressed customer of their own, using Company Watch’s matching of trade debtors against businesses already in administration or liquidation.