How Northwind cut month-end close from nine days to two

Nine days of matching rows

Northwind is a logistics marketplace that pays out to about 6,000 carriers a month through Payly. Until this year, closing the books each month took its four-person finance team nine working days. Most of that time went to one job: matching every payout against invoices, fees and bank deposits, then explaining the rows that did not match.

“It was careful work, and it was the same careful work every month,” says Jonas Feld, Northwind’s controller. “We were good at it. We just did not want to spend a third of our month on it.”

The setup

Northwind built a reconciliation agent that runs on every payout as it happens, rather than in a batch at month-end. For each payout, it:

• pulls the payout record and fee breakdown from Payly;
• finds the matching carrier invoices in their accounting system;
• checks the bank deposit through their banking connector;
• marks the payout as matched, or opens an exception with a short explanation and the evidence attached.

The agent cannot post journal entries. It can only propose them. A person on the finance team reviews and posts every adjustment, usually in a daily fifteen-minute batch.

We did not automate the close. We automated the part of the close that was just looking things up.

The evals

Before the agent touched live data, Northwind built a suite from three past months of payouts: 400 cases, including 60 that had needed manual fixes. Each case had four checks, covering the match, the amount, the fee treatment and the explanation for any exception.

The first run passed 81% of checks. Most failures came from currency conversions on cross-border payouts, where the agent rounded too early. After a tool change to return exact amounts, the suite passed at 97%, with no failures on the zero-tolerance checks for wrong amounts. The agent then ran in shadow mode for a full month-end before going live.

The results

1. Month-end close went from nine working days to two.
2. 94% of payouts now match without any human touch.
3. Exceptions arrive with an explanation, so median time to resolve one dropped from 25 minutes to 6.
4. The team found $41,000 in duplicated carrier fees in the first quarter, mostly from a billing bug on one account.

What they would change

Feld lists two things. First, they would have started with exceptions, not matches. The easy matches built confidence, but the explanations on hard rows were what saved real time. Second, they would have involved their auditors earlier. The auditors liked the traces, but asked for an export format that took a week to add.

Northwind now plans to apply the same pattern to accounts receivable next quarter. If you are starting your own, your first eval suite in an afternoon is a good place to begin.