FINANCE

WORKING CAPITAL

Reduce working capital from 67 to 49 days

How a €450M EU industrial distributor released €18M of trapped working capital — and reframed the CFO’s strategic options for the year ahead.

EU Industrial Distribution

€450M Revenue

18 Weeks Engagement

€18M

Cash revealed

67→49

DSO in days

−27%

Collection cycle

14×

Year-1 ROI

Illustrative case based on engagement patterns from EU industrial distribution. Specific business details, names, and exact figures have been adapted to preserve client confidentiality.

Situation

A €450M European industrial distributor — operating across six countries, serving roughly 12,000 B2B accounts — was carrying a working capital problem that had been creeping upward for four years.
Days Sales Outstanding (DSO) had drifted from 53 days four years earlier to 67 days currently. The aggregate effect: €26M of additional cash trapped in receivables relative to the historical baseline. The CFO had financed this expansion of working capital partly through the company’s credit facility and partly through delayed payments on the AP side, neither of which was a sustainable answer.

The standard explanation, repeated for several quarters, was that customers were paying more slowly because the macro environment had tightened. This was directionally true but operationally unhelpful — it framed the problem as exogenous and made the AR team feel like passengers rather than operators. The CFO suspected the diagnosis was incomplete but did not have the decomposition to challenge it.

The AR team, meanwhile, was running on the same playbook they had used for fifteen years: a monthly aging report, a hierarchy of dunning letters at 30/60/90/120 days, and a small team of collection specialists who worked the largest overdue balances by phone. The system was orderly but undifferentiated. Every overdue invoice was treated through the same workflow regardless of customer profile, dispute risk, or willingness to pay.
The engagement question that the CFO posed to DataQuant was direct. “Is this DSO drift macro, or is it operational? And if it’s operational, what would it take to recover it?”

Diagnosis

The DSO drift was 70% operational

When we decomposed the drift by customer segment, by product mix, by terms structure, and by collection-process behaviour, only about 30% of the increase was attributable to genuine slower customer payment patterns. The remaining 70% — roughly 10 days of DSO — was internal:

Invoice processing delay (3.4 days): the average gap between order fulfilment and invoice issuance had grown from 2.1 days to 5.5 days. The increase came from a manual approval step that had been added during a 2023 ERP migration and never removed.

Dispute resolution time (2.8 days): invoices flagged with disputes were being held in queue an average of 21 days before resolution. The disputed amount was usually small relative to the total invoice value, but the entire invoice was held off collection until the dispute closed.

Payment matching backlog (2.1 days): cash receipts from customers were arriving and sitting unmatched against invoices for an average of 4 days because the matching rules in the AR system did not handle partial payments or invoices paid via consolidated wire transfers.

Dunning misalignment (1.8 days): the dunning sequence treated all overdue invoices identically, which meant high-value commercial customers received the same treatment as low-value retail accounts. A subset of high-value customers were paying late simply because no one was working their account specifically.

None of these were dramatic individually. Together they were 10 days of DSO — €13M of working capital — generated entirely by internal process drift.

The aging report was hiding the real bottleneck

The standard AR aging report bucketed receivables by days-past-due. Current, 1–30, 31–60, 61–90, 90+. The CFO and the AR team had been managing against this report for years.

The bucketing was misleading. When we re-cut the data by cause-of-overdue rather than days-past-due, the picture changed materially. The 90+ bucket — €4.2M of overdue receivables, the focus of executive attention — was 78% disputed invoices waiting on internal action, not delinquent customers.

The non-obvious diagnosis:
The 90+ bucket was not a collection problem. It was a dispute-resolution problem disguised as a collection problem. Sending more dunning letters to those customers was actively counterproductive — they were waiting for the supplier to resolve the issue, not refusing to pay.

Conversely, the 31–60 bucket — considered “well managed” because aging was relatively short — contained a cluster of high-value customers paying consistently late by 14–18 days because no one specifically owned their account. These customers were not flagged in the aging report because they were never deeply overdue — they were chronically slightly overdue.

Customer payment behaviour was highly predictable

A behavioural analysis of 18 months of payment data showed that customer payment timing was strongly predictable from a small set of features: customer size, payment terms, historical payment lateness, dispute frequency, and recent order velocity. The variance in payment timing across accounts was much wider than the AR team realised — a typical account paid within ±3 days of its individual mean, but the means ranged from 8 days early to 45 days late.

This meant a predictive model could forecast, for any given invoice on the day it was issued, the customer’s most likely payment date with reasonable accuracy. That forecast could then be used to prioritise the AR team’s outreach — calling the customers who were forecasted to pay late, before they paid late, instead of calling everyone uniformly after the due date had passed.

Approach

Workstream 1 — Process recovery (weeks 5–9)

Three of the four operational drift sources had purely process answers, not analytical ones. They were sequenced first because they returned cash quickly and built credibility for the longer analytical workstream.

Invoice issuance lag. The 2023 manual approval step was reviewed and removed for invoices below €15K (roughly 87% of all invoices) where the existing automated controls were sufficient. Average time-to-invoice fell from 5.5 days to 1.8 days within four weeks.
Payment matching. The AR system’s matching rules were reconfigured to handle partial payments and consolidated wire transfers. A simple fuzzy-match layer was added to handle small variances in payment reference fields. Auto-match rate rose from 68% to 91%; manual matching backlog cleared in three weeks.
Dispute workflow. A separate “dispute hold” queue was created so that disputed amounts (rather than entire invoices) were the units held from collection. Average dispute resolution time fell from 21 days to 7 days because clear ownership was assigned to operations rather than left ambiguous between AR and customer service.

Workstream 2 — Cause-coded receivables view (weeks 7–11)

The single most-used artefact of the engagement turned out to be a re-cut of the AR aging report. Every overdue invoice was tagged with a cause code:

CAUSE-A: genuinely delinquent customer (collection action required)
CAUSE-B: disputed invoice awaiting internal resolution (operations action required)
CAUSE-C: system or process error on supplier side (process fix required)
CAUSE-D: customer pays consistently late but reliably (account-specific arrangement required)

The cause-coded view immediately changed the AR team’s daily work. Collection effort was redirected from CAUSE-B (which collection effort could not solve) to CAUSE-A and CAUSE-D (which it could). Roughly 31% of collection-team time was reclaimed and reallocated.

Workstream 3 — Predictive AR model (weeks 9–18)

The longer build was a customer-level payment-timing forecast model. For every active customer, on the day of invoice issuance, the model produced a predicted payment date with a confidence band. This output was wired into the AR team’s case management tool so the daily workqueue was sorted not by aging bucket but by highest expected impact of intervention — invoices with high predicted payment delay where outreach was historically effective.
The model used a gradient-boosted regressor (LightGBM) trained on 24 months of customer-level payment data, with features spanning customer characteristics, invoice characteristics, recent payment behaviour, and macro/seasonal signals. The model achieved a mean absolute error of 4.1 days on the test set, sufficient for operational prioritisation.

Outcome

Metric	Before	After (2 seasons)	Change
End-of-season gross margin	44.6%	48.8%	+4.2 pts
Margin recovered (annualised)	—	€7.8M	+€7.8M
Stock units in 50%+ markdown	Index 100	Index 69	−31%
Avg first-markdown timing (weeks from launch)	Week 7.4	Week 6.1	−1.3 wks
Merchant override rate	—	22%	—
Season-end inventory carry-over	Index 100	Index 91	−9%

The €18M of released working capital reset the CFO’s strategic options for the year. The credit facility utilisation — which had been climbing — was paid down. The delayed-payment posture on the AP side (which had been straining supplier relationships) was reversed and used to negotiate early-payment discounts that captured another ~€600K of value annually.

Key Lessons

"It's the macro environment" is rarely the full story

Most multi-year DSO drift is at least 50% internal. Decompose the drift into operational components before accepting an exogenous explanation. The decomposition is the diagnosis — and it usually surfaces fixes that the existing reporting was structurally unable to expose.

Aging buckets are the wrong organising principle for AR work

Days-past-due tells you that an invoice is overdue but tells you almost nothing about whether collection effort will resolve it. Cause-coded receivables — distinguishing delinquency from disputes from process errors from chronic late-but-reliable accounts — reframes the AR team’s daily work and immediately reclaims 25–40% of collection capacity.

Process fixes precede analytical fixes

Three of the four DSO drift sources had purely process answers — no model required. Sequencing the process fixes first returned cash within weeks and built executive confidence for the longer analytical workstream. A predictive AR model deployed before fixing the obvious process drift would have produced fewer outcomes and less credibility.

Customer payment timing is highly predictable

Most B2B customers pay on a tight individual distribution — ±3 days of their personal mean. The variance is across customers, not within them. This makes predictive AR models viable and usable, even on modest data volumes, because the signal is strong relative to the noise.

Working capital is a strategic instrument, not a tactical metric.

€18M released changed what the company could do, not just what its balance sheet looked like. Frame DSO programmes in terms of strategic optionality — facility utilisation, supplier leverage, M&A capacity — not just cash conversion cycle. The CFO conversation is more productive at that altitude.

Related Case Studies

Markdown Optimization at Scale

May 29, 2026

How a €620M European apparel retailer recovered €7.8M of margin across 320 stores — by replacing calendar-driven markdowns with sell-through-driven optimisation.

From Spreadsheets to SKU-Level AI

May 29, 2026

How a European confectionery manufacturer lifted forecast accuracy by 17% across 12,400 SKUs — and recovered €5.2M of margin in the process.