The 5 Weak Signals That Predict Business Failure 6 Months Before the Balance Sheet

Companies don't fail overnight. In 73% of cases, warning signs appear 6 to 12 months before insolvency proceedings — but they remain invisible to traditional assessment methods.

Here are the 5 patterns our models detect systematically.

Supplier Payment Delays: The Hidden Thermometer

Supplier payment days (SPD) is one of the most reliable indicators of a company's cash health. Unlike the annual balance sheet, it moves in real time and reflects liquidity tensions before they crystallize.

A rise from 30 to 65 days over two quarters is systematically correlated with increased fragility. In our database of 3,000+ files, this signal alone increases the probability of default within 12 months by 2.8x.

> "The balance sheet tells you what happened. Payment flows tell you what's happening." — RocketFin Data Team

:::insight **RocketFin Exclusive Data** — In 3,000+ analyzed files, companies whose SPD exceeded 60 days for two consecutive quarters had a 12-month default rate of 18.4%, versus 3.2% for those with stable SPD. :::

Abnormal Director Rotation

Every management change registered in public records generates a signal in our engine. An unplanned CEO or CFO departure, especially combined with a short tenure, indicates organizational fragility that financial ratios don't capture.

High-risk patterns include:

- 2+ management changes in 18 months - CFO resignation less than 6 months after a fundraise - Asset pledging immediately after a new director's appointment

:::takeaway **Key Takeaway** — Director rotation is an organizational signal, not a financial one. Almost all classic models ignore it. That's precisely why it's predictive. :::

Digital Presence Degradation

A company that stops investing in its online presence is actually signaling an undisclosed budget restriction. Metrics we continuously analyze:

| Signal | Alert Threshold | Score Weight | |---|---|---| | SEO ranking loss (top 10) | -40% in 3 months | +8 risk pts | | Google Ads spend reduction | -60% | +5 risk pts | | LinkedIn inactivity | > 90 days | +3 risk pts | | Domain expiry | < 60 days | +12 risk pts |

:::alert **Regulatory** — From August 2, 2026, any score used in a commercial decision must justify its contributing variables (EU AI Act). RocketFin provides the 5 main variables per score, including digital signals. :::

Early Legal Signals

Public records contain early indicators that very few operators cross-reference with financial data. Our engine monitors:

**Asset pledging** — A sudden pledge of a business goodwill or industrial equipment often reveals a short-term liquidity tension the company is trying to finance through a secured loan.

**Confidential conciliation proceedings** — Although not published, their existence often shows up in pledge registries. Strong signal: 6x higher probability of failure.

:::insight **RocketFin Exclusive Data** — Companies with a pledge registered in the 6 months preceding analysis have 4.2x higher probability of default within 12 months, all else being equal. :::

Gap Between Declared Performance and Behavioral Signals

The fifth signal is the most subtle — and often the most predictive. It's a discrepancy between what the company communicates (figures, speech) and what its behaviors reveal.

Concrete examples:

- An SME announcing "40% growth" but whose banking flows are stagnant - A director displaying growing public optimism while multiplying personal guarantees - Customer payment delays lengthening while the company communicates a "full order book"

This discrepancy is measured by our engine through cross-referencing declarative data (balance sheets, registrations) and behavioral data (flows, legal, digital).

What This Means for Your Credit Decision

These 5 signals form the backbone of our predictive model. Individually, each has limited value. Combined and weighted by sector and company size, they produce a score with 87% accuracy at 12 months — versus 61% for traditional balance-sheet-only methods.

The practical difference: out of 100 analyzed files, our models correctly identify 87 at-risk profiles versus 61 with classic methods. For a 500-client portfolio, this statistically represents 130 avoided defaults per year.