Business process automation: first we assess which step is worth delegating.
First you delete what exists out of habit, then you write the procedure, and finally you automate the part worth automating. A useless step, once automated, is still useless, and it needs maintaining too.
- 35
- 2026
- 0
Business process automation · Choosing the first workflow
Which step do you repeat by hand?
Business process automation connects data, tools and work steps. Before choosing software, we document how a request arrives, who handles it, which exceptions arise and what result must be verified.
Collecting and checking dataOpen an example
A recurring report or a comparison between exports: when formats and rules are stable, a script or workflow can collect the data and flag anomalies.
Reading text and draftingOpen an example
Email, documents or requests with varying content: AI can extract information and prepare a draft. We define result checks and the cases that need a person.
Coordinating steps across toolsOpen an example
A request must pass through several systems. When the path is known, we code the workflow; when steps depend on content, we consider an agent with explicit access and limits.
Examples to assess, not results already measured at your company. Access, compatibility and value are checked before implementation.
What we define before building
A bounded process, the tools involved, who checks the output and what happens when something fails. The quote separates implementation, platform usage and maintenance.
Automating isn’t adding software. It’s removing work.
Automating a process means assigning part of repetitive work to a system, with verifiable inputs, rules and results. A person keeps the checks and handles the agreed exceptions.
The right candidate is almost always the boring one: repetitive, with stable rules and a verifiable output. If every run needs a different decision, the automation becomes fragile.
Boring processes are the best ones. Interesting ones almost never are.
Recurring reports and statements
The data is already there, in the management software or the shared spreadsheet. What’s missing is the piece that collects it every week and writes to you only if something has moved.
Content and publishing with quality control
Not generating text, but the chain: gathering, claim checking, style control, publishing and logs. Without checks you just make mistakes faster.
Monitoring and alerts on what breaks
An automated process without monitoring stops working silently. This is where we decide who checks, when it alerts and when it stays quiet.
Data handoffs between systems that don’t talk to each other
The email that becomes a row in a table, the form that ends up in the CRM. Not glamorous, but the hours saved show up right away.
Four steps on the process, and the first doesn’t touch any tool.
Delete
Every step that exists out of habit leaves the flow. Sometimes the cleaned-up process is no longer worth a project: that is a result too.
Write the procedure
Inputs, steps, exit criterion, who checks. Readable by a new person and by an agent.
Automate the part worth automating
The part with the best ratio of hours saved to fragility. The work picks the tool: a script, a visual orchestrator or custom code.
Add a watchdog and watch it run
Every automation has a declared way of failing: what happens if the source doesn’t respond and how you notice. Then we observe it for a few real cycles.
Need to pass cost, payback and maintenance on to whoever signs off? The sign-off brief (in Italian) fits on one page, with no form.
The same order applies to my own system, and sometimes it leads me to switch off automations I wrote myself. Deleting is the step nobody sells.
If the team doesn’t use the automation, it doesn’t pay off. For the part that concerns people, there is team training.
Two real processes, measured while they ran, not from memory.
Twenty-one scheduled tasks in place of three hours a day (in Italian)
Writing, publishing, commenting, measuring, every day. Automated in blocks, with human review where it was needed. It’s my own case, measured on me, not an industry average.
Taking the vendor out of the path of every change (in Italian)
A firm that had to call someone to change a single line, with a site that was slow on mobile. Rebuilt on a stack the client updates on their own.
A client case with an agent, step by step and with its declared numbers, is in the case study. If the process has to interpret documents and decide, the right page is the one on agentic systems.
Eight questions on cost, maintenance and what happens when it breaks.
Which business processes should you automate first?
Repetitive ones, with stable rules and a verifiable output: reports, data handoffs, periodic checks, conditional notifications. The test: it repeats at least once a week, two people would do it the same way, and there is an objective way to say the result is right. If one of the three is missing, fix the process first.
How much time do you really get back?
It depends on the process: be wary of anyone who gives you a percentage before looking at it. My best-documented case is my professional presence: 21 scheduled tasks and 60+ hours saved in three weeks, measured on me. For your process, the number is estimated in the scoping call.
Do you use n8n, Zapier or Make?
If you need a visual orchestrator that someone at the company must be able to open, the case picks the tool. My personal system doesn’t run on n8n, Zapier or Make: it’s Claude across several runtimes (scheduled jobs on the Mac, cloud routines, a local GPU box, Vercel cron jobs) plus Python scripts, kept in sync by a signal bus in Markdown.
What happens when an automation breaks?
Sooner or later it happens: sources that change format, new APIs, expired credentials. That’s why every automation I deliver has in writing how it fails, how you notice, what it does in the meantime and who gets alerted.
Do we need to replace the management software we already use?
Almost never. We work on top of what’s there, via APIs or scheduled exports. Changing management software is a separate project, with its own costs and risks.
Who maintains the system after delivery?
You can do it yourself: files, procedure and code stay in your repository. If you’d rather I look after it, we agree on a scope with realistic response times. I work alone: no 24/7 support.
How much does it cost to automate a business process?
It depends on how many systems the process touches and how much monitoring it needs, not on the number of steps. We start with the 30-minute scoping call at €97, deducted from the quote if the project goes ahead. Consulting and project prices are on the booking page.
What about compliance? Does automating with AI expose us to anything?
If an AI system enters the process, Art. 4 of Regulation (EU) 2024/1689, applicable since February 2, 2025, requires taking AI literacy measures for the people who use it: it is an obligation to take measures, not to achieve a result. For those who use AI, Art. 50 covers specific cases: deepfakes, text published to inform the public on matters of public interest, emotion recognition and biometric categorization. The EU database concerns high-risk systems. For literacy, the measure I offer is AI training for teams. This is not legal advice: validation is up to your lawyer.
Where the models run, how I handle data and what the AI Act says about an internal process: Data, models and the AI Act (in Italian).
Bring the process that weighs on you most. We look at it before touching any tool.
Even if you describe it badly. In 30 minutes we open it up step by step: what to remove, what to write down, what to automate first. You keep the map. The call runs in English or Italian, remotely or on site.