Month-End Close in Two Hours: The Process Behind It

Our month-end close used to cost half a day, plus a week of follow-up questions from the tax advisor. Today it takes two hours, and the questions have become rare.

The difference barely happens at month end. It happens in the 30 days before.

I will show you what our process looks like: what has to run continuously, which steps go automatically on closing day, where the tax advisor comes in and which four mistakes slow every close down. This is a description of how we work, not tax advice.

Why the close takes so long in the first place

At its core a month-end close is a single question: is every movement on the account explained, documented and correctly assigned.

Whoever asks that question for the first time on the 3rd of the following month asks it 300 times in a row. For every transaction your head has to travel back into the context. What was that debit. What was that invoice for. Do I even have the receipt.

The most expensive part is retrieving that context. A payment to an unknown vendor from the 8th of last month costs you ten minutes of searching. The same payment on the day it was booked costs you ten seconds.

That is the whole lever. Everything else is execution.

What has to run before month end

Four things run continuously here, without anyone scheduling a meeting for them.

Transactions arrive automatically. Our business account is with Qonto, and Qonto has a clean API. Our system pulls new transactions several times a day, with amount, date, counterparty and payment reference. Every booking sits in the system a few hours later. The month-end close does not start on the 1st, it runs the whole time.

Rules first, then the AI. A payment to our fulfilment partner is always the same case. A Meta invoice is always advertising cost. Cases like that need a fixed rule, not AI. Rules are traceable, instant and free. Anything that matches no rule goes to a language model, which suggests a category with a reason. The full setup is in Automating bookkeeping with AI.

Receipts get collected and matched continuously. Every receipt lands in one of three channels: an email inbox, a Drive folder, or directly on the transaction in the bank. The AI reads issuer, date, number and amounts and looks for the matching payment. Details on that in Capturing receipts automatically.

The exception list is always visible. Two lists run permanently: transactions without a receipt, and uncertain assignments. We take a quick look once a week, usually ten minutes. Those ten minutes are the reason the close takes two hours.

Closing day: what happens automatically

At month end little new happens. Most of it ran long ago.

The only real work is in row five. We go through the list the system could not resolve cleanly: transactions without a receipt, uncertain categories, batched debits. For us that is a few dozen rows, not three hundred.

The export is one click at the end. DATEV, the accounting format German tax advisors work with, is very well documented, and the effort does not sit in the export. It sits in everything before it.

Where the tax advisor comes in

The boundary matters here, and it is drawn deliberately. Our system does the preparatory work. It does not do the bookkeeping.

The tax advisor gets a finished export with matched receipts instead of a pile of loose PDFs. He reviews, corrects, posts the final entries and produces the statements. His questions per month have dropped noticeably, because the most common one has disappeared: "where is the receipt for this debit."

These topics stay entirely with him:

• VAT and advance returns. Rates, reverse charge, intra-community supplies, OSS. Not a rulebook you automate on the side.
• Accruals and deferrals. An annual licence paid in March belongs proportionally across twelve months. That is a professional judgement.
• Fixed assets and depreciation. What gets capitalised and over how many years is his decision.
• Shareholder matters. Contributions, withdrawals, intercompany accounts. Nobody touches those automatically.
• Annual accounts. The real craft, and it has little to do with the monthly routine.

At the start he was sceptical, AI and bookkeeping sounded like risk to him. His most important objection is our most important rule today: every step has to stay traceable. For every automatically assigned category, the system records whether a rule or a model set it and for what reason. Leave that out and you save three days of development while losing all auditability.

The four mistakes that delay the close

From our own operations and from client projects, the same four keep coming up.

1. Receipts only get hunted down at month end. The most common and most expensive mistake. A receipt you file on the day of payment costs seconds. The same receipt four weeks later costs you a login to a portal whose password you forgot. Our rule: a receipt that is not in one of the three channels does not exist for the system.

2. Private and business on one account. Every private debit on the company account is a row a human has to touch. Twenty of those a month are twenty interruptions. A cleanly separated account is the cheapest speed-up there is.

3. Cash. Cash receipts have no bank transaction to attach themselves to. The automatic matching runs into nothing, every item becomes manual work. That is why we pay for almost everything by card.

4. New suppliers without a rule. Every new vendor creates an unclear case the first time around. If nobody creates a rule afterwards, the same vendor creates another one next month. That is how the review list quietly grows. For us, creating the rule is part of the review, not part of some later clean-up session.

On top of that come two cases that stay manual for us as well: batched debits where one payment covers several invoices, and foreign currency, where the invoice amount and the debit drift apart through exchange rate and fees. No matching engine solves either automatically.

What a two-hour close gives you beyond time

The time saving was our reason for doing it. The real benefit was something else, and I had underestimated it.

If transactions get classified continuously, you have your numbers continuously. We see the cost structure of the current month, not the one before last. Advertising, cost of goods, software, fulfilment: all categorised, a few hours after booking.

That turns bookkeeping into controlling on the side. Our liquidity view builds on the same data, more on that in Cashflow forecasting for ecommerce.

The second side effect is psychological. A close that takes two hours gets done. A close that takes half a day gets postponed, and one postponed month turns into two. What the postponing costs overall is in The true cost of manual processes.

The limits

So you do not get the wrong picture.

• Building it was not a weekend project. Rules emerge over months, because every new supplier is one you did not know before.
• The AI gets it wrong sometimes. A cryptic payment reference, an unknown counterparty, and the suggestion is off. Without a review step that would slip through. When money is involved, a human confirms, always.
• It does not replace a tax advisor. See above. It makes his input easier, nothing more.
• It does not pay off at every size. At 30 transactions a month you sort them by hand faster than you build an automation. Somewhere around 150 to 200 transactions a month the maths flips.
• Two hours is our number, not yours. Anyone with several entities, warehouses in multiple countries and VAT in five states is playing a different game.

And the most important sentence to close this section: this is a description of our process. It is not tax advice. What applies in your case is something you clarify with your tax advisor before you rebuild anything.

Conclusion

A fast month-end close is created during the month. Pull transactions automatically, rules before AI, collect receipts immediately, keep the exception list small every week.

On closing day what is left is one review round and one export. For us that is two hours instead of half a day plus follow-up questions.

If you are currently losing a day every month, start with a single step: pick one intake channel for receipts and use only that one from tomorrow. For most people that alone halves the effort.

And if you want a system like this without building it yourself: we at Flowhouse built exactly this for ourselves and build it for clients. Write to us, we will walk you through the process in detail.