Why does our month-end close take 12 working days, and how do we get to 5?
A close that runs to calendar day 18 delivers figures nobody needs for steering any more. The causes lie in process, data and tool, and most have nothing to do with AI. This page shows what a close in 5 working days looks like, what invoice processing and accrual routines contribute and what they do not, documented in a 2026 mandate.
How do you recognise the symptom?
The report for management, board or parent company is ready on calendar day 18, regularly only on the last day of the deadline. That is around twelve working days. The prior month's figures arrive when the current month is already half over. Whoever wants to steer with that is steering by figures that are already out of date.
In a 2026 mandate, 1.5 of 2.5 positions in the finance function were tied up with tasks a machine can take over: around 5,000 documents a year received by email, renamed, forwarded, filed in Outlook folders, uploaded by hand. Around 3,000 rows of personnel cost were typed from PDF payroll statements every month.
The late close costs more than waiting time. The bank asks for the prior month before it exists. The board decides on investments using figures that are six weeks old. And the 1.5 positions that rename documents are missing for the forecast, liquidity and the reconciliation with the tax adviser. That is the price management feels.
Which causes sit behind it?
A long close rarely has one cause. Usually there are five or six that add up, and each on its own costs only a day or two. The table sorts them by process, data and tool, with the marker by which you recognise the cause without commissioning an analysis.
| Cause | How you recognise it | What it costs |
|---|---|---|
| Process: no closing calendar | Nobody knows on the 1st of the month what has to be finished on the 3rd; every close runs differently | Waiting time between steps, because nobody triggers the next one |
| Process: accruals only at the end | Invoices arriving after the cut-off are collected and accrued by hand at the very end | Two to three days of waiting for the last incoming invoice |
| Process: document path through mailboxes | Incoming invoices travel by email, get renamed, forwarded and uploaded manually | 1.5 of 2.5 positions in one case; no timestamp, no first-pass rate |
| Data: master data in several systems | Invoice recognition has been bought and still does not recognise the supplier | Every unrecognised document is re-entered by hand, often with an altered invoice number |
| Data: personnel cost from PDF | Payroll statements arrive as PDF from the provider and are transferred row by row | Around 3,000 rows a month in one case; the file sat EUR 16.2k above the management accounts |
| Tool: report from unlinked files | The report is built from several Excel files, every transition is copy and paste | Prior-month values stay standing; controls are missing |
A seventh cause has been added since 2025. Companies in Germany must be able to receive e-invoices; in the 2026 mandate the reception was not covered by any process, and the first e-invoices were already arriving. Whoever prints them as PDF and types them in loses the only document that would be machine-readable without recognition.
Four of the six causes are process and data. Whoever starts with them gains most of the days without a single new tool. A closing calendar costs one afternoon. Accruals by order value in place of invoice receipt cost one rule. Both are part of stage 1 on the page AI automation in finance.
What does a good state look like, in numbers?
Five working days after month end. That is my target in mandates for companies with EUR 4m to 100m revenue, on a fixed calendar: day 1 bank and cash, day 2 incoming and outgoing invoices with accruals, day 3 personnel and fixed assets, day 4 reconciliations and check sheet, day 5 report. Each step has an owner and a time.
In the reporting prototype of 2 September 2026 the monthly routine stands in five sentences on the first sheet: insert DATEV exports, insert ERP files, capture raw data, check inputs, everything else calculates. 18 checks run automatically, both balance sheets reconcile to zero. Anyone with Excel skills can run the routine.
The target includes two metrics collected before the first tool: the first-pass rate of incoming invoices, meaning the share posted without rework, and the cycle time from receipt to posting. Without these two figures nobody can say whether invoice processing has shortened the close or merely moved the pile of documents.
A good close is also one that a second person can run. In that mandate the knowledge of how the report is produced sat in a single head outside the company; a written description was produced only on request. The closing calendar and the control sheet remove that dependency.
What does automation contribute?
Two places in the close pay off first. Invoice processing: incoming invoices are recognised, assigned to the vendor and matched with order and goods receipt, once the master data sits in one system. And accrual routines: what was delivered by the cut-off but not invoiced is calculated from orders and goods receipts, without waiting for the last invoice.
Both require the data base. In that mandate the bought invoice recognition failed on the master data: three systems, three spellings of the same supplier. Invoice numbers were altered when typed in, the goods receipt date posted in place of the invoice date. No later automation could have matched them. In practice, automation fails on master data, not on technology.
For the report, the machine read the 11 files, recognised the data flows and generated the links, the check sheet and the log. Which inputs deliberately remain inputs and which balancing items need clarification was financial judgement. A measured shortening of the close does not yet exist; operating values follow after the first month of the prototype.
What automation cannot contribute: the decision which accrual is correct in substance, whether a balancing item stays and whether the close is released. A set of figures can be internally consistent and still wrong. That is why nothing goes to the client before a second, separate instance has taken it apart.
What did this look like in a mandate?
Group subsidiary, around EUR 40m revenue, 2.5 positions in finance, bookkeeping outsourced to the tax adviser. The monthly report to the parent company came from 11 Excel files of the tax office without a single formula link. Delivery by calendar day 18, according to the client usually only met as the deadline expired.
In one working day all 11 files were transferred into one linked workbook: over 250 manual transfers are now formulas, each documented in the log with sheet, cell, old value and new formula. During linking, 12 items surfaced that still carried June values in the July report, and four balancing items whose origin nobody could explain any more.
That is a prototype and is called one: built and reviewed, not yet in monthly operation. What is measured after the first month of operation will appear here with a date. Until then: the close only gets shorter once the document path through mailboxes has ended, and that is process work.
The document path is therefore the first process to be automated in stage 2, with the first-pass rate as the metric and the delivery day as the evidence. Both are collected before and measured after. Only when the delivery day has fallen from calendar day 18 to the fifth working day will a measured shortening stand here.
Frequently asked questions about the month-end close
Is fast close something for mid-sized companies?
Yes, in the form of a fixed calendar with five working days. The corporate fast close with completion on the second working day needs systems and staff that a company with EUR 4m to 100m revenue does not have and does not need. Five days are enough for management to steer with the prior month.
Does this work when bookkeeping sits with the tax adviser?
Yes, when the closing calendar fixes the handover points: when the documents reach the tax adviser, when the postings come back, when the report stands. In the 2026 mandate the bookkeeping was outsourced; the prototype pulls DATEV exports and ERP files into one workbook, the rest calculates.
Which tool do we need for invoice processing?
None at first. First master data in one system, then a document path with timestamps, then a tool that recognises and assigns documents. In that mandate the tool had been bought and failed on three spellings of the same supplier. The selection comes at the end of stage 1, with a metric.
What do we start with tomorrow?
With a closing calendar for next month: five days, the steps per day, a name and a time per step. Then count how many documents arrive by email and how many files sit between accounting and the report. These two numbers tell you which cause comes first.
- Anonymised 2026 mandate (group subsidiary, around EUR 40m revenue): findings on delivery date, document path, personnel cost and master data from the concept paper of 2 September 2026; reporting prototype of 2 September 2026.
- Guide AI in finance: what really works and what does not, task "The month-end close takes too long".
- Guide Finance function after Series A: the first 90 days, result "monthly close in a fixed rhythm".
Your close is ready on the 18th?
30-minute initial call, no obligation. You tell me the delivery day, the number of documents arriving by email and the number of files up to the report, I tell you which of the seven causes comes first. Reply within 24 hours.
Request an initial call