Contents8
Monday, 8.30 a.m. Someone exports the week’s sales, opens last month’s spreadsheet, pastes, fixes a formula that broke, adds the support figures copied from another screen, formats, sends. At 11 the report goes out; at 2 p.m. a figure is challenged, and nobody remembers which export it came from. This guide from our section on process automation and business bots shows how this report can build itself, and why almost nothing in it needs AI.
In short
An automatic report reads figures in the tools that produce them, without exports or copy-and-paste.
Each indicator is calculated the same way every time, from a definition written once.
The danger is not the late report, it is the wrong report sent on time: you need safeguards.
AI can comment on changes in plain language; it calculates nothing.
The time freed goes to analysis: the person who assembled the report finally reads it.
Why copy-and-paste costs more than the time spent
Time first: one morning a week is around fifty mornings a year spent on work nobody chose. But the real cost lies elsewhere. Every copy-and-paste is a chance for error: a forgotten row, a filter left on, a formula pointing at a shifted column. And because the report is rebuilt by hand, two people never rebuild it quite the same way.
Large organisations are not immune. After the heavy 2012 losses of its London investment office, JPMorgan published an internal report, quoted by economist James Kwak and picked up by the New Statesman. It described the risk model involved as follows:
Operated through a series of Excel spreadsheets, which had to be completed manually, by a process of copying and pasting data from one spreadsheet to another.
Your Monday report does not involve the same sums, but the mechanism is the same: a figure that passes through hands and intermediate sheets ends up wrong without anyone noticing.
What an automatic report does for you
The Monday report, with nobody copying it
- Read
Sales, CRM, accounting, support, at the source.
- Calculate
Each indicator from its written definition.
- Check
An empty source or an absurd gap blocks the send.
- Send
At a fixed time, to the team’s inbox or channel.
Reading figures at the source
Most business tools expose their data: an API, a scheduled export, a readable database. The report fetches them itself, at the same time, with the same filters. When a business application offers no way in, there are others, which we compare in six ways to automate software without an API.
Writing each definition once
“This week’s revenue”: excluding or including tax? Orders placed or delivered? Credit notes deducted? As long as the answer lives in the head of the person who builds the report, two reports do not say the same thing. Once written into the automation, the definition applies to every send, and changes in a single place.
Sending where it gets read
A report that lands in a team channel at 8 a.m. gets read; a link to a dashboard you have to open gets read less. A good report fits on one screen, with the comparison to the previous week, and a link to the detail for those who need it.
Safeguards: a wrong report sent on time
An automatic report has a flaw the manual one does not: it goes out even when its data is wrong. A source that does not respond, an API that returns an empty list, and the report calmly announces zero sales.
Watch out
A green workflow is not a correct result
On Palmora Property, publishing safeguards block an empty import: without them, an empty API response would publish a site with no listings while the workflow showed green. A report needs the same rule: an empty source or an unprecedented gap blocks the send and alerts a person.
Three checks are enough in most cases:
- the source responded, with a plausible number of rows;
- the gap with the previous week stays within an expected range, otherwise the report flags it at the top;
- the report actually went out: its absence at the expected time triggers an alert.
That last point is developed in our guide to monitoring n8n workflows: an automation that goes silent is more dangerous than a forgotten manual task.
And where does AI fit in?
It has no business in the calculation. Adding, comparing, filtering: these are rules, which always give the same result, as our guide to automation without AI, when a rule is enough explains. A language model recalculating figures could get them wrong from one run to the next.
AI can, on the other hand, write the commentary: three sentences saying in plain language what moved and where to look, from figures already calculated. This text is reviewed before it goes out, at least in the first weeks, and it does not replace the analysis of the person responsible for the report. For marketing in particular, our article on automated marketing reporting details the indicators that show whether an action pays off.
What we run
From the field
Our SEO engine reads Search Console data every day and turns it into the week’s work queue: since it was created on 25 August 2026, this scan has been entirely deterministic and calls no AI model. An online plush-toy store receives its automatic digests (revenue, orders, average basket, anomalies), and Palmora Property its daily report. The principle is the same everywhere: a qualified alert when something happens, silence when there is nothing to say.
What the report frees up
The person who assembled the report, management controller, executive assistant or sales manager, does not disappear from the process: they move from making it to reading it. They have time to understand why a figure moved, ask the right department, propose a decision. That is the work they were hired for.
Where to start
- Pick one report.
The one that comes back every week and takes the longest.
- List its figures.
With, for each, the tool it comes from and its exact definition.
- Drop what nobody reads.
An automatic report does not need to carry everything over.
- Set the safeguards.
Empty source, abnormal gap, missing send.
- Run it in parallel.
Two weeks alongside the manual report, to compare.
That is the approach of our n8n automation, with or without AI: we start from the report as it is made today, with the person who makes it, and every run shows its cost. The other articles in the section deal with neighbouring processes: automated supplier invoices, employee onboarding, WhatsApp and website bots, AI voice agents on the phone, customer support sorted by AI, n8n, Make or Zapier, the n8n licence and AI and young people’s first jobs.
Frequently asked questions
Which tool should you use to automate a report?
An orchestrator such as n8n, Make or Zapier is enough to read the sources, calculate and send the report by email or to a team channel. A dashboard tool becomes useful when several people need to explore the figures themselves. The choice of tool comes after the list of figures and their sources.
Do you need a business intelligence tool?
Not to start with. A weekly report of ten well-defined figures, sent at a fixed time, is often more useful than a dashboard nobody opens. A BI tool is justified when the questions change from week to week and teams need to cross-reference the data themselves.
Can AI write my reports?
It can write the commentary, from figures already calculated by rules: three sentences on what moved and where to look. It should not calculate the figures themselves, and its text is reviewed before sending, at least at first. The person responsible for the report remains the one who draws the conclusions.
How do you stop an automatic report from being wrong?
By checking its inputs before sending: a source that did not respond or returns an empty list blocks the report and alerts a person. Also compare each figure with the previous week and flag unusual gaps. Finally, monitor the send itself: a report that did not go out on time should trigger an alert.
Sources
- The New Statesman (new tab)newstatesman.com
Jim · founder of Brain Plus AI
Also in automation and bots
Practical guide9 min
Monitoring your n8n workflows: no silent failures
An automation that fails without warning is worse than a manual task: nobody knows the work has not been done. Here is the minimal set-up that turns a silent failure into an alert, then into a rerun, using the tools n8n already provides.
Choose and price9 min
Automation without AI: when a rule is enough
Not every automation needs AI. A fixed rule costs less and always gives the same result: here is how to spot the steps where AI really pays off, and those where it only adds cost.
Practical guide6 min
Supplier invoices: automating reading and matching
A supplier invoice goes through three steps: reading it, matching it against what was ordered and received, then approving it before payment. What can be automated at each step, where AI really helps, and what stays with the accountant.