Can you do monthly reporting without a spreadsheet?

Yes, for collection, consolidation and layout. Data is taken directly from accounting, the bank and the CRM, and the report is sent by e-mail. The budget can stay in a spreadsheet: it becomes one source among others, no longer a place for copy and paste.

Can you trust variance comments written by an AI?

To the extent that each comment rests on figures you can check. Beyond the threshold, the line-by-line detail must be attached, and the system must flag a variance it cannot explain rather than invent a reason. A person reviews before distribution.

When can the report be available?

From the first working day of the next month for the parts that rely on data already available. If your accounts are not up to date by then, the report will show provisional figures: the delay then comes from bookkeeping, not from the report.

Do we need to change accounting software or CRM?

Not necessarily. Reporting connects read-only to your current tools. The real prerequisite is a stable set of definitions: what each indicator contains and which budget is authoritative.

How many indicators should a monthly report contain?

Few. One accounting firm recommends 8 to 12 at most, which is a rule of thumb, not a standard. What matters is that each indicator serves a decision and that the list stays stable from month to month.

Monthly reporting without a spreadsheet: what prepares itself

Monthly reporting without a spreadsheet: what can prepare itself?

Short answer: collecting the figures, consolidating them, comparing them with the budget and the previous month, and drafting a first version of the variance comments can all be prepared without human effort. Analysis, decisions and the sign-off on what goes to management or partners stay with people. The target is a report ready to read on the first working day of the next month, instead of mid-month.

In many small companies, the monthly report is a spreadsheet that grows every month. Someone exports the accounts, copies the bank balances, matches the budget, pastes in CRM figures, then writes three paragraphs explaining what moved. It takes days and arrives when the month is already forgotten. This article describes what can be automated in that chain, what should not be, and how to set it up without changing your tools.

Why the report arrives in the middle of the next month

The delay almost never comes from layout. It comes from aggregation: the figures live in different places (accounting, bank, a budget spreadsheet, the CRM) and someone has to gather them by hand before any thinking can start. A March 2026 article from the provider Tensoria makes the same point: what looks like a dashboard problem is really a problem of merging heterogeneous sources. That is a vendor’s view, but it matches what we see in small companies: the spreadsheet is not the flaw, it is the meeting point for data that do not talk to each other.

There is a second, quieter cause: waiting for the close. Until the month’s entries are posted, the report cannot go out. A guide from the accounting firm Hayot Expertise, published in January 2026, suggests a typical cycle with the report produced around day 7 and presented around day 10, and considers that beyond day 15 a report loses most of its decision value. These are the firm’s own method benchmarks with no cited data, but they give a scale: a report arriving on day 15 or later describes a month you can no longer correct.

The six steps of a report that prepares itself

This is the chain as we build it for monthly reporting (the service page is in French). Five steps are prepared by the AI, one stays with you:

Month data pull. Data is taken from accounting, the bank and the CRM on the agreed date, with no manual export.
Consolidation. The month’s result and cash position are computed from those sources, using the same rules every month.
Variances. Figures are compared with the budget and the previous month. Beyond a threshold you set, the detail is supplied line by line.
Commentary. Each significant variance gets an explanation written from the lines that make it up.
Formatting. The report is laid out and sent by e-mail, ready to read.
Analysis and decisions. That is you: you read, you challenge, you decide.

The order matters. Commentary comes after the variances are computed, never before, and it rests on lines you can check. That is what separates a useful comment from a plausible sentence.

What prepares itself, what gets commented, what stays human

Part of the report	Recommended handling	Why
Pulling balances and entries	Automatic	Mechanical task, single source, no interpretation
Result, cash, basic indicators	Automatic, fixed calculation rules	The calculation must be identical from one month to the next
Comparison with budget and previous month	Automatic, with an alert threshold	The threshold decides what deserves an explanation
Explaining a recurring variance (rent, subscriptions, payroll)	Automatic first draft, quick review	The cause can be read in the lines
Explaining an unusual variance	Detail supplied, comment written by a person	The cause is often a fact the figures do not show: a lost client, a billing delay
Adjusting entries, provisions	Decision of the responsible person	These are judgements, validated with your accountant
The month’s decisions	You	The report informs, it does not decide

This table is not an accounting rule. How entries, provisions and legal obligations are treated is a matter for your accountant or adviser: have them validate your rules before you write them into the system.

Variance commentary: the part that worries people, and why it can be controlled

An AI-written comment is reliable to the extent that it is anchored in figures you can check. A serious system does not invent explanations: it notes that a line went up, lists the entries that make it up, and states the variance in one sentence. Beyond the threshold, the line-by-line detail is attached so the reader can trace back to the source.

Two safeguards are worth demanding. First, every figure quoted in a comment must exist in the report: if the sentence says “payroll costs are up”, the amount and the line are visible alongside. Second, when the system cannot find the cause in the data, it must say so (“variance not explained by this month’s entries”) rather than invent a reason. A report that admits what it does not know is more useful than one that reassures.

Hayot recommends commenting on variations above 5% and limiting the report to 8 to 12 indicators. These are a firm’s rules of thumb, not a standard, but the idea is sound: an explicit threshold and a limited number of indicators keep the commentary readable and the checking possible. Set your own thresholds, line by line, with the person who will read the report.

A worked example: what manual reporting costs

Here is a calculation with assumptions you should replace with your own. In a 20-person company where the office manager spends 2 days every month producing the report (collection, spreadsheet, matching, comments, layout), that is 24 days a year. At an internal cost of 70 francs an hour and 8 hours a day, that comes to roughly 13,400 francs a year of internal work, not counting the managing director’s review time. This is an order of magnitude, not a measurement.

For comparison, a provider, Kolonell, claims in a September 2026 article a gain of 2 to 4 days a month for an automated report. That is a commercial estimate with no independent source: it shows the range vendors advertise, not what you will get. To replace our assumptions with your data, the five-question express diagnostic puts a figure on the time lost from your own tasks.

Setting it up without changing your tools

The starting point is not a new management-reporting package. It is your current tools: accounting, bank, the budget spreadsheet and the CRM. This is the order we follow:

Freeze the target report. One page, 8 to 12 indicators, in the order your management reads them. Do not start from an empty dashboard.
Fix the definitions. Which lines count as revenue, what “available cash” means, which budget is authoritative. This step saves more time than everything after it.
Connect the sources read-only, without altering existing tools.
Run two months in parallel. The automatic report and the usual one are compared line by line. Every difference is explained before moving on.
Set the thresholds and the send date. The report goes out on the first working day, at the same time, to the same recipients.
Name a reviewer. One named person reads the report, with a deputy, before the management meeting.

If your reporting depends on badly categorized expenses, deal with that first: we covered it in the article on categorizing SME expenses. An automatic report does not fix late bookkeeping, it simply makes the lateness visible sooner.

Common mistakes

Automating a report nobody reads. If nobody consults the current report, automating it produces the same result faster. First decide what it should help you decide.
Multiplying indicators. Forty lines drown the three that matter. Keep few, and keep them stable.
Changing definitions midway. If revenue is not computed the same way each month, comparisons are worthless.
Removing the reviewer. Sending partners a text nobody has read moves the risk onto them.
Forgetting the source of truth. Two budgets in circulation produce two different reports. Only one is authoritative.
Sending financial data to a consumer service. Check where the figures are processed and stored before choosing a solution, and have your adviser confirm this if your data is sensitive.

When this approach is not the right choice

We advise against automating reporting in four situations. First, if your accounts are not up to date before mid-month: the delay comes from bookkeeping, not from the report, and that is where to start. Second, if your definitions change every quarter (chart of accounts being rebuilt, software migration): stabilise them first. Third, if your report is one page of five figures that you fill in within twenty minutes: automation adds nothing that a well-made template does not already provide. Fourth, if nobody has time to review: a report without a reviewer is a risk, not a gain.

In the first three cases, the right investment is tidying up. In the last, you first need to decide who is accountable for the report.

What a typical month looks like afterwards

This is the intended sequence, as an illustration and not a measured result. During the month, nothing changes in your tools. On the last day, the data is pulled. On the first working day, the report arrives by e-mail with its variances and comments. The named person reviews it and corrects what needs correcting, then management makes its decisions while the month is still fresh. The change is not only the time saved: the meeting is finally about the month that just ended.

Where to start

Take last month’s report and note three things: the real time spent producing it, the sources it depends on, and how many indicators your management actually looks at. If one person spends more than a day on it, there are no more than four sources and the definitions are stable, you have a good candidate. You can also start from the list of processes the studio automates to place reporting among the other tasks in your business.