The monthly report that prepares itself

Monthly reporting is the most predictable exercise on the finance calendar: same deadlines, same tables, same recipients. Yet in most SMEs and mid-sized companies it is still assembled by hand, in a sprint the whole team knows by heart: the exports, the chasing, the consolidation workbook, version 7 of the pack sent at 11:40 pm.

This guide starts with the substance: what a good reporting pack contains, how it differs from a dashboard, the real steps of a monthly close. Then it takes on the uncomfortable question: why reporting automation so often disappoints, and under what condition an agent can genuinely prepare the pack while your team keeps the final call.

What a good monthly reporting pack contains

A useful financial report comes down to five blocks. The rest is decoration.

• The analysed P&L: the month and year-to-date actuals, against budget and prior year, by activity or cost centre. Not twenty tabs: the lines that moved, and why.
• Cash: position at the closing date, movement over the month, three-month forecast. It is the block the CEO reads first.
• Working capital: customer and supplier payment terms, outstanding balances, inventory. This is where cash is actually won or lost.
• Business indicators: the five to ten figures specific to your activity - order book, occupancy rate, margin per project.
• The management commentary: the part that carries the value. A figure without an explanation postpones the question to the next meeting; an explained variance closes it.

Dashboard and reporting: two different objects

The two words are often used interchangeably. They do not serve the same purpose.

A dashboard steers: a handful of indicators, refreshed continuously or nearly so, to act within the month. It tolerates approximation - a 98% accurate figure today beats an exact figure in three weeks.

Reporting gives an account: a snapshot frozen at a date, produced on a fixed rhythm, on a stable scope, for management, shareholders, the bank. It has no right to approximation: investment decisions rest on it, and so does the trust of the people financing you.

Confusing the two costs money both ways. A dashboard treated like a report arrives too late to be useful; a report treated like a dashboard ends up containing figures nobody dares to stand behind.

The real steps of the monthly close

A monthly close is a rapid close, focused on the accounts that matter - a much lighter exercise than the annual one. Eight steps come up everywhere:

1. Cut-off: attach sales and purchases to the right month - invoices to issue, invoices not yet received.
2. Bank reconciliations: every bank line explained, every open item identified.
3. Matching of customer and supplier accounts, for clean balances.
4. Recurring entries: payroll, depreciation, prepaid expenses, routine provisions.
5. Intercompany accounts, where the group has several entities: reconciling them is the first step of any consolidation.
6. Analytical review: compare against budget and prior year, isolate the significant variances, get the explanation from the people running the business.
7. Producing the pack: tables, charts, commentary.
8. Distribution and the review meeting.

On paper, five working days are enough. In real life, teams often take twice that, and the difference plays out in everything around the accounting entries.

Where the days evaporate

Closing night rarely looks like accounting. It looks like this: export the trial balance from the accounting system, collections from the bank, revenue from the invoicing tool, payroll from the HR software. Paste everything into a workbook whose consolidation tab breaks at the first inserted column. Chase three managers for missing supplier invoices, and a sales rep for a credit note promised ten days ago. Redo the charts because the scope changed. Then reread, cell by cell, because a wrong pack costs more than a late one.

Every finance team knows the split: most of the time goes into collecting, rekeying and formatting; the analysis, the only part that creates value, gets whatever is left.

Automation has a blind spot: the verification tax

AI promises to give those days back. Look first at what it does with the time it frees. An IDC study commissioned by Sage in February 2026, surveying around 3,000 senior finance decision-makers and covered by IT Social, puts a number on the unease: French finance leaders spend on average thirteen hours a week verifying the results produced by AI. The study calls the phenomenon a "verification tax". 78% of French CFOs would reject a tool that is 99% reliable if it remains unable to explain the logic of its answers, and for 18% of French respondents verification exceeds thirty hours a week - the point where the tool creates more work than it removes.

The study was commissioned by a software vendor, so it should be read with distance. But the mechanism it describes can be observed without any study: in finance, a wrong figure has consequences - before the tax authority, the statutory auditor, the bank. A figure produced by an AI that cannot show where it came from does not save time: it moves the work. The team no longer keys data in; it reconstructs the proof the tool should have supplied. That is the worst of both worlds: the liability without the control.

From re-checking to arbitrating

The verification tax is not a law of nature: it is the consequence of a design choice. Two properties make it disappear.

Every figure cites its source. A figure in the pack comes out of your systems, never out of a model, and it arrives with its reference - the journal entry, the statement line, the invoice. Verification shrinks to click, see, decide.

Every action leaves a trail. The audit log records what the agent collected, computed and prepared, when, and from what. What the study describes as after-the-fact reconstruction becomes a simple read.

Concretely, the close changes shape. Through the month, the agent collects entries and statements from your existing tools - the accounting system, the bank, the spreadsheet all stay in place - runs the reconciliations, chases the missing documents. On the evening of the last day, it assembles the pack, compares against budget and prior year, and above all flags the variances, each with its source and its size. In the morning, your team does not open a collection project; it opens a list of documented variances, arbitrates, completes the commentary, signs off. The agent prepares, the human decides.

That is the work our agents for finance departments take on, running on root Workspace: your own instance, in France, connected to your tools, with citations and an audit log. And the raw material is getting better by the month: from September 2026, electronic invoicing turns the supplier flow into structured data - exactly what an agent handles well.

Frequently asked questions

How do you automate monthly reporting?

In order: stabilise the reference data (chart of accounts, analytical dimensions, scope), replace manual exports with direct connections to the source tools, automate the preparation (reconciliations, variance calculations, a first draft of the commentary), and keep a human sign-off before distribution. The condition that governs everything else: every automated figure must remain traceable to its source, otherwise manual verification takes back what automation gave.

What is the difference between a dashboard and financial reporting?

A dashboard is for steering: a few indicators followed continuously to act within the month, where freshness matters more than precision. Reporting is for giving an account: a frozen, periodic, standardised snapshot on which management, shareholders or the bank take decisions. The first tolerates estimates; the second commits you.

Which tools for financial reporting?

Four families cover most needs: the accounting software or ERP (the source), the spreadsheet (the workshop), treasury and BI tools (the presentation layer), and, for groups, a consolidation tool. The decisive factor is less the brand than the chain: one single source per figure, no rekeying between two links. An agent can orchestrate that chain on top of your existing tools, without replacing them.

Can an AI get a financial report wrong?

Yes. A model can misread a data point, match two lines that do not belong together, or write a confident commentary on a fragile figure. Reliability is therefore a matter of design: figures read from your systems rather than generated, a citation per figure, an audit log, and human validation before anything goes out. The useful question to ask a vendor is "how do we see that it was wrong, and how fast?" - because sooner or later it will be.