Home / Insights / Banking

Before an NPL Forms: Designing an Early Warning System

By the time arrears reach 90 days it is too late for early warning. An effective EWS combines financial deterioration, account behaviour, sector signals and relationship manager knowledge long before that.

The cheapest stage of problem loan management is the stage before a loan becomes an NPL. The most valuable investment in credit risk management is therefore a strong early warning system (EWS) and a clear action mechanism that runs once a signal fires. The EBA guidelines on loan origination and monitoring and on NPE management also stress the importance of monitoring, governance and early intervention.

Signal groups

Combinations of signals rather than one signal

Letting a single indicator raise an alert produces a high false positive rate. A short limit overrun, for instance, can be normal behaviour in a strongly seasonal business. But if a limit overrun, a deterioration in collection periods and a rise in tax arrears all appear together, the economic meaning of the risk changes.

In a modern EWS design, therefore, signals are weighted, trend and speed information is used, comparisons are made against customer and segment norms, and defined thresholds are tied to action levels.

What happens after an alert?

1. Verification: the relationship manager and risk check whether it is a data error or a temporary event.
2. Classification: the intensity of monitoring and the risk level are set.
3. Action plan: further information, a meeting, freezing limits, repricing, collateral, restructuring or exit options are defined.
4. Deadline and owner: every action needs a responsible owner and a completion date.
5. Feedback: the model is recalibrated by measuring regularly which signals genuinely predicted default or NPL.

Reading it in reverse, for the corporate customer

Companies are not invisible to a bank's early warning system. On the contrary, a company should see the signals the bank sees even earlier in its own management reporting. The 13-week cash flow, receivables ageing, covenant headroom, inventory turnover and bank limit utilisation all belong on the CFO's own early warning dashboard.

The best restructuring is not the one done after the problem has grown; it is the one where the funding requirement, the maturity structure and the business plan are redesigned together as soon as the risk signal appears. Early warning is therefore not only a risk mitigation tool but a customer sustainability tool.

Official sources and further reading

• EBA — Guidelines on loan origination and monitoring
• EBA — Guidelines on management of non-performing and forborne exposures
• EBA — Risk Assessment Report, June 2026

In EWS design, what happens after the signal matters more

If nothing is triggered when the model flags a customer “amber” or “red”, the early warning system is only a reporting tool. For each level, a responsible person, a deadline, a mandatory review, a customer contact plan and a credit action have to be defined.

Combine financial, behavioural and external signals

The cost of a false positive

An over-sensitive model raises alerts on hundreds of customers and relationship managers stop taking the system seriously. Too loose a model catches the problem late. Thresholds should be calibrated not only for statistical accuracy but against the case capacity the operating team can review and the cost of false alarms.

Watchlist governance

The criteria for entering and leaving the watchlist should be written down, not left to relationship manager discretion. The different views of credit risk, the business unit and, where needed, workout teams should be held in the same case record; the decision taken and the reason for it should form an audit trail.

How is model performance monitored?

• What percentage of customers that reach NPL or default were caught beforehand?
• What is the average lead time in days or months?
• How long does it take from alert to action?
• What is the false positive rate, and how frequent are relationship manager overrides?
• What is the marginal contribution of each signal type?

A sample scoring architecture for an EWS

A simple starting model might weight financial, behavioural, external and sector, and relationship manager qualitative signals at 40, 35, 15 and 10 per cent respectively. Those proportions are not universal; they should be calibrated to the portfolio type and the quality of the data. But the idea matters: different sources of information are combined instead of relying on a single arrears signal.

Override governance

A relationship manager's close knowledge of the customer is a valuable source of information; but a model that is constantly overridden is not reliable. The reason for every override should be coded, compared with the outcome that actually followed, and analysed to see where the model makes systematic errors.

The handover point between early warning and workout

The aim of the early warning team is not to send the customer to workout as late as possible but to send them at the right time. Once the problem has become structural, continuing with normal relationship management can increase the severity of loss. The trigger set should make the handover point explicit through criteria such as arrears, covenant breach, a cash gap, a change of management or a rapid deterioration in the credit rating.

Four management views in portfolio reporting

• New watchlist entries and their main causes
• Customers leaving the watchlist and the reason for exit
• Average lead time on watchlist customers that became NPL
• Concentration of alerts by sector, product and branch

Those reports feed back not only into the risk function but into credit policy, pricing, limit and sector appetite decisions. The EWS then turns from a tool for catching problem loans into a learning system that improves portfolio strategy.