Your Working Capital Problem Isn't a Finance Problem. It's a Data Problem.

Every manufacturing CFO I meet has a working capital target, and most of them are missing it. What's striking is how consistently the response is organisational rather than informational: pressure the collections team, run an inventory reduction drive, renegotiate supplier terms.

Each of those addresses one component in isolation. And in isolation, each of them can be achieved while total working capital stays exactly where it was, because the cash conversion cycle is a single connected system, and squeezing one part of it usually pushes the problem into another.

The cycle is one system managed by four teams

Days inventory outstanding, days sales outstanding and days payable outstanding are owned by supply chain, commercial and procurement respectively, each with its own metric, its own system and its own incentive. Nobody owns the sum.

The results are predictable. Procurement wins a discount for early payment and DPO drops. Supply chain hits a service level target by holding more safety stock and DIO rises. Sales closes the quarter by extending credit terms and DSO rises. Every team hit its number. The CFO's number got worse.

You cannot optimise a cycle by optimising its components separately. You need one view of the whole thing, current enough to argue with.

Why the view doesn't exist

Not because the data is missing, because it's scattered and unreconciled. Receivables ageing lives in the ERP but the reason an invoice is unpaid lives in an email thread. Inventory sits in the WMS but the reason a batch isn't moving lives in a quality hold note. Purchase commitments are in the ERP but the renegotiated term is in a PDF someone saved to a shared drive.

So the monthly working capital pack is assembled by hand, is three weeks stale by the time it's discussed, and shows balances without causes. It tells you DSO went up. It cannot tell you that DSO went up because eleven invoices to one customer are disputed over a delivery documentation issue that's now three months old.

A sequence that works

Attack it in order of cash return per unit of effort. Receivables first, it's the fastest money and the data is the cleanest.

1. 1Receivables. Reconcile ageing against dispute correspondence, delivery proof and credit notes to separate genuine disputes from administrative delay. Most organisations discover a surprisingly large share of overdue balance is not a payment problem at all, it's a missing document or an unresolved short-shipment nobody chased.
2. 2Inventory. Classify slow-moving stock by why it's not moving: forecast error, quality hold, obsolete SKU, or committed against an order. The action for each is completely different, and an aggregate DIO number hides all four.
3. 3Procurement. Bring actual payment behaviour together with contracted terms and reconcile the gap. Early payments made without a discount, and discounts contracted but never captured, are both common and both pure margin.
4. 4Dispatch and order-to-cash. Close the loop from dispatch to invoice to collection, and measure where the days actually accumulate, invoicing lag is frequently a bigger contributor than customer payment behaviour.

What AI contributes

The reconciliation. Reading dispute correspondence and matching it to invoice lines. Extracting terms from contract PDFs and comparing them to actual payment runs. Classifying quality hold notes into inventory categories. Matching short-shipment claims across a delivery note, an email and an ERP record that don't share a key.

That's the work that makes a live cash conversion cycle possible, and it's precisely the work that's too laborious to do by hand at the frequency it needs doing. The financial decisions stay entirely with finance.

Working capital isn't stubborn because your teams aren't trying. It's stubborn because they're each optimising a component they can see against a whole that nobody can.