Key takeaways

The starting problem: monitoring that ate hours every week

Following AI news seriously used to take hours of reading every week. Dozens of outlets publish daily, a large share of the articles overlap, and the genuinely useful signal drowns in the volume. The observation is banal, and that is exactly what makes it interesting: nearly every company has a version of this problem.

A manufacturing SMB tracks competitors' prices and product releases. A professional services firm follows regulatory changes in its sector. A sales team watches for tenders and buying signals. In every case the pattern is identical: collect scattered sources, cut the noise, summarize what matters, distribute to the right people. And in every case, that work is either half-done or done at the cost of recurring hours that produce nothing else.

IA Brew (page in French) was born from that observation. The goal was not to read less, but to stop spending a single minute on the mechanics: reading should start where the selection is already done.

What was built: a complete chain, from collection to delivery

The result is a 93-node n8n workflow that runs every week. It collects 20+ feeds, merges articles telling the same story, scores and selects the most relevant topics, writes the summaries, lays everything out and sends the email. Nobody touches anything between the trigger and the recipient's inbox, and every step is logged: when something goes wrong, you know where and why.

A word on the tool, because it matters to the equation. n8n is a workflow automation platform of German origin, with 1,000+ integrations, that can run on your own server; your data then stays with you. It is no longer a niche project: on May 12, 2026, SAP made a strategic investment valuing n8n at $5.2bn and plans to embed the platform in its Joule Studio environment by Q3 2026. Seven months earlier, in October 2025, n8n had raised $180 million at a $2.5bn valuation. When a vendor like SAP builds an automation tool into the core of its offering, it says something simple: these building blocks are becoming standard infrastructure for business processes, SMBs included.

Why does a "simple" task need 93 nodes?

Because the version that works in a demo and the version that runs unattended for months are two different pieces of software. The happy path, where every source responds and no article is duplicated, accounts for only a small part of the workflow. Most of the nodes handle everything else: a source that stops responding, a feed that changes format, two outlets publishing the same story an hour apart, a failed send, a step that must resume without restarting the whole chain.

That is the most transferable lesson of this project: the difference between a gadget and a dependable tool is failure handling and step-by-step logging. A workflow without logs or alerts always ends the same way: it breaks silently, someone notices three weeks later, and trust is gone. By contrast, a workflow that warns its owner before recipients notice anything becomes a piece of infrastructure you forget about.

The same pattern runs in client settings too: for the French cheesemaker Ermitage, an automated monitoring pipeline built on n8n collects and distributes the information their teams need, with the same logging discipline. The subject changes, the mechanics stay.

When automation is not the right choice

This needs saying plainly, because it is the question a business owner should ask before signing anything. Three situations make automation premature or pointless.

If the task takes 20 minutes a week, the return on investment is not there: keep it manual. If the process changes every month, automation will freeze a way of working that has not stabilized yet; stabilize first, automate second. And if an off-the-shelf tool already covers the need, a standard feed aggregator or simple keyword alerts, take the off-the-shelf tool: custom automation earns its keep when you need to cross heterogeneous sources, apply your own ranking criteria, and distribute in your own formats.

A good automation candidate looks like this: a repetitive task, weekly or daily, stable in shape for months, spanning several tools, and costing hours to someone whose actual job is something else.

What you can check in your own company

The exercise fits in three questions. Which recurring task does your team redo identically every week? How many hours does it consume, cumulatively, over a quarter? And is the process stable enough to describe on one page? If all three answers exist, the task is probably automatable, and the profitability math takes minutes.

If you want to run that math on your own case, the simplest route is a 30-minute scoping call: we identify your most expensive process and what a documented n8n automation would concretely change, numbers included. And if the honest answer is "keep your manual process", that is the answer you leave with.

Frequently asked questions

How much does an n8n automation like this cost?

Scoping takes 30 minutes. A first, focused need, such as a monitoring pipeline that collects and summarizes a few sources, often fits in a Sprint starting at €400. A full pipeline with deduplication, error handling and automated sending is quoted per project, depending on how many tools need connecting and how many failure cases need covering.

What happens when a source goes down?

A well-built workflow includes explicit error handling: every step is logged, failures trigger alerts, and broken sources are skipped cleanly instead of blocking the whole chain. The owner finds out before the readers do, not after.

Do I keep the workflows if the engagement ends?

Yes. The client receives the exported workflow files, the administrator access and the documentation. They can take over in-house or hand maintenance to another team at any time, with no dependency on the original contractor.

The takeaway. IA Brew's 93 nodes are not a technical feat; they are honest accounting: roughly one third visible work, two thirds reliability. Before automating anything in your company, ask a contractor one question: "what happens when it breaks?". The quality of the answer is worth more than any portfolio.