Bureau scores are built from filed accounts, and a private supplier in England can lawfully file up to nine months after year end, so the file you are reading may describe a business that stopped existing in its current form a year ago. The signals that move earlier sit in your own transactions: the tail of each order shipping late, partial deliveries rising, a request to shorten payment terms, engineering changes going unanswered. Screening the whole vendor master produces a queue nobody works, because with a failure base rate near one per cent even a good screen is mostly false alarms, so run it only where a failure would stop production. When a signal fires, secure position on tooling and title before you talk about volume, and take advice before paying anything early.
The administration notice arrives on a Monday morning. The supplier makes a moulded housing that goes into four product families, they had scored green on the quarterly scorecard three weeks earlier because their delivery performance was 96 per cent, and the tooling for all four housings is bolted to their machines.
You have five weeks of cover. Retooling at an alternative moulder is a twenty-six week job assuming the administrator releases the tools, and whether they release them is now a legal question rather than a commercial one.
Going back through the previous nine months, the signals were all there. The last twenty per cent of nearly every order had started arriving three to six days after the rest. Two schedule changes had been refused that would previously have been absorbed. Their engineer stopped replying to change requests in March. Somebody in finance had approved a move from sixty day terms to thirty in April as a goodwill gesture, without anyone asking why it was requested. None of that appeared on the scorecard, which measured what was delivered against what was promised, and by that measure they were fine until the day they were not.
What the credit file knew, and when
The statistical prediction of corporate failure is a well developed field and the models are genuinely useful. Altman published the Z-score in the Journal of Finance in 1968, a five-ratio discriminant function that classified failures in his original sample with around 95 per cent accuracy one year before the event and about 72 per cent two years before. Ohlson followed with a nine-variable logit model in the Journal of Accounting Research in 1980. Beaver had shown in 1966 that a single ratio, cash flow to total debt, carried much of the signal on its own.
Two properties of that literature matter to a buyer.
The first is that accuracy travels badly. Begley, Ming and Watts re-estimated both the Altman and Ohlson models on 1980s data in the Review of Accounting Studies in 1996 and found classification error rates considerably higher than the originals reported, which is the ordinary fate of a model fitted on one period and applied to another. Altman himself published variants for privately held and non-manufacturing firms, dropping the market value term that a private supplier does not have, and those variants are the ones relevant to most of a direct materials base.
The second property is the one that decides timing. All of it runs on filed accounts. In England and Wales a private company has nine months from its accounting reference date to file, under the Companies Act 2006, and small companies file abridged accounts with no profit and loss detail. A score you read in November may be computed from a year ended the previous December, describing trading that finished eleven months ago, from a business whose largest customer left in March. Bureau scores also carry a heavy weight on payment behaviour reported by other creditors, which is real information and lags the operational reality by a quarter or more, because a company under pressure pays its critical suppliers and stretches everyone else in a deliberate order.
The credit file is worth having. Treating it as an early warning system is what fails, because it was never built to be one.
The signals that live in your own transactions
Your own systems record a distressed supplier's behaviour weeks or months before anyone files anything, and the records are already there. The useful ones share a shape: they are what a business does when cash is short and it is trying not to say so.
The tail of the order goes late. A supplier short of raw material or paying its own suppliers on stop ships what it can and holds the rest. Aggregate on-time performance barely moves, because most lines are fine. Measure lateness on the final quarter of each order line separately from the first three quarters and the divergence shows up early.
Partial and split deliveries rise. Same mechanism, visible in the receipt count per purchase order line. A rising average number of receipts per line is one of the cleanest indicators available and almost nobody measures it.
They stop absorbing changes. Flexibility is the first thing a plant loses when it is running hand to mouth on materials. A supplier that used to take a pull-in at two weeks' notice and now refuses at six is telling you something about their input position.
Somebody asks about money. A request to shorten terms, to be paid on despatch rather than on receipt, for a deposit against a tooling change, or for a price increase justified by input costs that have not moved. These arrive through commercial channels and rarely reach whoever monitors supplier risk.
Quality drifts in a maintenance-shaped way. Deferred maintenance shows up as dimensional drift, surface defects and process capability decline before it shows up as a breakdown. A separate post covers what to do about that on its own terms, and read as a financial signal it is one of the earliest available.
People leave. Turnover in the account team, a finance director departing mid-year, the plant manager replaced by an interim. Sales staff at a struggling supplier are usually the first to go looking.
None of these is conclusive alone. Two or three of them arriving in the same quarter at the same supplier is a different matter, and the combination is what to monitor.
The base rate problem, in numbers
The temptation is to run this screen across the whole vendor master. Work out what that produces first.
Say you have 1,200 direct suppliers and the annual failure rate among them is 0.8 per cent, so about ten failures a year. Suppose your composite screen is good: it catches 80 per cent of the failures and clears 90 per cent of the healthy suppliers. It flags eight of the ten real ones. It also flags ten per cent of the 1,190 healthy ones, which is 119. You have 127 alerts a year to find eight events, and roughly fifteen of every sixteen alerts is a false alarm.
Nobody works that queue for long. Within two quarters the alerts are being closed unread, including the eight that mattered.
The arithmetic changes when you change the population rather than the model. Apply the same screen to the ninety suppliers whose failure would actually stop production, and it produces around ten alerts a year. Ten alerts is a half day of work per alert, which means each one gets a real review: a look at the transactional signals, a call to the account manager, a check of the filed accounts and any charges registered against the company. A separate post covers how to identify that ninety, and another covers ranking whatever the screen throws out by consequence rather than by score.
The general point is worth carrying beyond this topic. A screen for a rare event is only as useful as the review capacity behind it, and narrowing the population is almost always cheaper than improving the model.
What to do when one fires, in order of cost
The response that gets reached for first is usually the worst one available. Cutting volume or shortening terms at a supplier under cash pressure accelerates the failure you are trying to avoid, and if you are a large share of their revenue it may cause it outright. Work the ladder from cheap and reversible upward.
Start with a conversation. A visit and a direct question, asked by someone who has a relationship there, gets further than a questionnaire or a formal letter. Ask about their own supply position and their order book rather than their balance sheet, because those are questions a plant manager will answer honestly and they tell you most of what you need.
Then verify what is verifiable. In the United Kingdom, charges registered at Companies House, county court judgments, late filing notices and a change of auditor are public and cheap to check. Equivalents exist in most jurisdictions. A new floating charge granted to a lender is a fact rather than an inference.
Then secure your position on things that are physically or legally at risk, which is where most of the recoverable value sits and where delay is expensive.
Then build cover, on the specific parts that gate the most revenue, sized against the requalification lead time rather than against a policy target.
Then start the alternative. Qualification is slow, it is the item on this list with the longest clock, and beginning it does not commit you to switching.
The tooling, the title and the paying-early trap
Three legal facts decide how much of the ladder above is available, and they are worth establishing before a signal ever fires rather than during an insolvency.
Tooling ownership needs to be documented and, ideally, visible. Tools you paid for that are in a supplier's plant with no purchase order trail, no asset labels and no acknowledgement of ownership are a claim you will be arguing in the middle of an administration, competing with a lender that has a charge over the plant's contents. An annual confirmation from the supplier listing your tools by asset number is a fifteen minute administrative task that changes the conversation entirely.
Retention of title works in both directions. Your supplier's own supply contracts may include title retention clauses of the kind litigated in the Romalpa case in 1976, which means material sitting in their plant that you believe you have effectively bought may still belong to their supplier. If they hold consignment stock for you, or you hold consignment stock for them, get the ownership and the segregation documented while everyone is still friendly.
Paying early is the one to take advice on. Where a supplier fails soon after receiving preferential treatment, an insolvency practitioner may be able to challenge the payment: preferences under section 239 of the Insolvency Act 1986 in England and Wales, or a preference claim under section 547 of the United States Bankruptcy Code with its ninety day look-back, one year for insiders. The rules are jurisdiction-specific and fact-specific, and the practical instruction is that a decision to accelerate payment to a supplier you believe is in trouble belongs with counsel before procurement acts on it.
Where this stops
Financial distress models predict formal insolvency. The event that hurts you is a supply stoppage, which happens more often, arrives earlier and includes cases where the supplier survives perfectly well. A supplier can be solvent and still deprioritise your volume because a better customer arrived, or lose the contract manufacturer behind their own product, or be told by their bank to shrink the order book. Monitoring for insolvency catches a subset of the events you care about.
Ownership structure hides a great deal. An operating company can look healthy on its filed accounts while the debt that will kill it sits in a holding company two levels up, and a private equity owned supplier may be servicing borrowings that never appear in the entity you trade with. Where the exposure is large enough to matter, the group structure is the thing to map, and it takes an afternoon.
There is also a limit on what any of this buys you when the requalification clock is long. If a second source takes two years to qualify, as it does in some regulated categories, a six month warning changes very little about your options and mostly changes how much stock you build. In those categories the answer has to be structural and decided years earlier, which is a different piece of work from monitoring.
Pull the last twelve months of goods receipts for your top fifty critical suppliers and compute two numbers per supplier: the average number of receipts per purchase order line, and the lateness of the final quarter of each order line against the rest. Both come out of the receiving system in an afternoon, and both move before the accounts do.
More on this in resilience on the DemandLabs platform.
