The Trading Brain, Applied to Finance

The deal is done. The money still takes weeks.

In timber trading, closing the deal is rarely the slow part. What follows it is. A supplier invoice arrives as a PDF attached to an email, with the payment terms in the footer and the reference number in a format nobody else uses. Someone opens it, checks the amounts against the purchase order, checks the quantities against what actually arrived, works out the due date, books it, and files it. Then does it again, several hundred times a month, across currencies, countries and VAT rules.

The receivable side runs on the same fuel. A spreadsheet of who owes what, a calendar reminder to chase, a polite email, a second email, and eventually a phone call from someone who would rather be selling timber. None of this work is complicated. All of it is exact, and exact repetitive work is precisely what a small team ends up doing at the expense of everything else.

The real cost is not the hours, though. It is the blind spot. If your payables and receivables live in inboxes and spreadsheets, you never have a reliable picture of what is owed, by whom, and when it lands. You are running a working capital business on an estimate.

The same Brain, pointed at the ledger

We have written before about the Trading Brain, the AI layer that sits across the systems our desk already runs on and handles the routine work between them. Finance was the natural next place to point it. The pattern is the same: it works inside the tools the team already uses, it runs both as a chat interface grounded in real data and as a background agent that acts without being asked, and it escalates to a person only when a human decision is genuinely needed.

Four things now run without a person in the loop:

• Invoices read on arrival. Supplier and customer invoices are ingested straight from email, with amounts, dates, currencies, VAT treatment and reference numbers extracted and matched line by line against the purchase order and the goods actually received. Clean matches are booked. Only genuine mismatches reach a person, and they arrive with the discrepancy already identified.
• Payment terms as data, not fine print. Every agreed term becomes a structured due date the moment the invoice is booked. Net 30 from invoice date, 14 days from bill of lading, a discount for early settlement: parsed once, applied everywhere, and visible to everyone who needs it instead of living in one person's memory of the negotiation.
• Supplier payments scheduled, not chased. Due dates group themselves into payment runs proposed ahead of time, ordered by due date, early payment discount and relationship priority. Releasing the payment stays a human decision. Preparing it, evidencing the approval and confirming it back to the supplier does not.
• Receivables watched and followed up. The Brain tracks what is outstanding, sends reminders before the due date rather than after it, escalates ageing items to the account owner who actually knows the customer, and flags a payment pattern that is drifting long before it turns into a write-off.

We did this for the cash position, not for the admin

Taking work off the accounting team was the visible benefit, and it is a real one. It was not the reason. Timber trading is a working capital business: you pay a sawmill weeks before your customer pays you, and somebody finances that gap. When payables and receivables are unstructured, the size and timing of the gap is a guess, and the only safe response to a guess is to hold a buffer against your own uncertainty. That buffer is expensive, and it does nothing.

Once every invoice, term and due date is structured data, the cash position stops being a month-end reconstruction and becomes a forward view: what lands this week, what leaves, what changes if a container is delayed by ten days. Money that used to sit idle against uncertainty can be deployed on purpose, and the conversation with financing partners changes too, because a position you can evidence is priced differently from one you assert.

What that gives our customers

A trader who knows their cash position can do things a trader guessing at it cannot. Offer longer payment terms to a buyer with a strong record, instead of applying the same cautious default to everyone. Pay a sawmill earlier than agreed in exchange for better pricing, and know exactly what that costs. Take on a larger order without waiting for the previous one to settle.

That is the point where financing stops being a separate conversation with a bank and becomes part of the trade itself. Terms can be extended to the customers who have earned them, priced against a position we can actually see, rather than set by the most cautious case in the portfolio. For a buyer, it means working capital that follows their track record. For a supplier, it means being paid on a schedule they can plan around.

What comes next

The next step is internal, and it is about how far we can extend financing. Every customer we offer payment terms to today is one we can underwrite from experience: we know the account, we know the history, we know the exposure. That works, and it does not scale. The constraint was never appetite, it was visibility. Terms we could not evidence were terms we could not responsibly give, so a buyer with a perfectly good business but a short history with us got the cautious answer.

Once payables, receivables and the cash position run as structured data, that judgement stops being a memory exercise and becomes a calculation. Payment behaviour, order patterns and exposure per customer are already in the system, updated as invoices land rather than reconstructed at month end. That lets us extend terms to customers we previously had to decline: newer accounts, smaller buyers, and markets where we had no track record to lean on. The same discipline, applied to a much wider set of customers.

It also opens a door that was closed to us before. Buy now, pay later and trade finance providers underwrite on exactly the data this produces: verified invoices, confirmed deliveries and a payment history that can be evidenced rather than described. Connecting them directly to that data means terms can be offered on an order without every euro of it sitting on our own balance sheet, which is what turns financing from something we ration into something we can offer as standard.

The wider that service goes, the less of it can depend on a person in the middle. That is the same problem the Brain already solves everywhere else on the desk, and it is why the finance layer was worth building first. We built it because our own operation needed it, we run it on ourselves before anyone else, and every customer we can now finance is one we would have had to turn away a year ago.