How to automate processes with AI in your company: A Practical Guide

1. START WITH THE PROCESS, NEVER WITH THE TECHNOLOGY

At a recent conference on AI in industrial companies, the experts all came to the same conclusion: the bottleneck has never been the technology, but rather the process and the culture.
They summed it up with a figure I use every day:

The challenge is 80 per cent business and 20 per cent technology.

The practical consequence is immediate: automating a disorganised process only exacerbates the chaos. If your order-processing system exists solely in one person’s head and across three separate Excel spreadsheets, adding an agent or an AI system on top of that won’t solve anything. 

The same applies when embarking on a digital transformation simply out of a ‘fear of being left behind’ or because of the ‘hype’ surrounding it today; AI or technology will only amplify the chaos or the order of your processes. 

💡 What you can do:

• Map the process as it is TODAY (not as it should be)
• Identify who does what, using what information and which tools
• Distinguish between decisions that require human judgement and those that do not
• It is the least glamorous part of the project, but the one that determines whether automation will deliver benefits or not.

2. WHAT SHOULD YOU AUTOMATE FIRST? THE 30% RULE

McKinsey assessed this even before the roll-out of this new generative AI began:
→ Barely 5 per cent of roles can be fully automated. 
→ BUT one-third of the tasks in 60 per cent of roles can be automated.


The unit of analysis is not the job role. It is the task.
The tasks most likely to be automated share three characteristics:
→ Repetitive and high-volume: they occur dozens or hundreds of times a month, always following the same structure. 
→ They do not require judgement, but rather a process: extracting data from a PDF, categorising an email, preparing a draft reply. 
→ The information is accessible: in an email, the ERP system or a document. Not in someone’s head. If the data required by the process is scattered across different formats, there will first need to be some data preparation work
A good AI agent or automation system works like a highly trained intern: it researches, prepares the work, and your team supervises and makes the decisions. 
Tasks that do require judgement — such as approving, negotiating or deciding on an exception — are always excluded from the initial scope of automation. 
Starting with supervised assistants, rather than autonomous agents, is exactly what the companies that are leading the way today did.

3. THE FOUR STEPS WE FOLLOW IN EVERY PROJECT

1. Map the process as it stands today: not as we want it to be. This is where the surprises come to light: duplicate steps, approvals that nobody can remember why they exist, information that is copied several times.
2. Redesign before automating: often the greatest savings do not come from AI or technology, but from eliminating unnecessary steps. First, eliminate waste; then implement the steps that add value and prioritise use cases.
3. Build the minimum system that works end-to-end: one task, one process, one use case, in production and validated. Prioritise small steps forward rather than a six-month pilot with twenty use cases: one that works and frees up hours from the very first week.
4. Monitor, measure and scale: every agent, system or automation must always start with human review of what matters. As the data shows it is performing well, its autonomy is expanded – not before.

4. THE REAL BENEFIT: SCALING UP WITHOUT HIRING

We need to be honest about what automation and artificial intelligence can and cannot do. 
Automation is not about replacing people. It’s about freeing up time wasted on repetitive, mindless tasks so that we can grow and increase turnover before expanding the workforce. 

The result should never be ‘fewer staff’, but rather greater productivity for the existing team. Doing more with less. 

5. SUBSCRIPTION OR DEVELOPMENT: THE DECISION THAT ALMOST NO ONE TELLS YOU ABOUT

There are two main ways to automate processes using AI:

→ Rent: sign up for a SaaS solution for each process. It’s quick to get started, but over time subscriptions pile up, your data isn’t really yours, and the monthly cost just keeps rising.

→ Develop: build systems on top of the tools you already use, with solutions such as self-managed n8n, where the system is documented, auditable and owned by you. When the developer finishes, the system remains yours. 

Large Catalan companies have made up their minds: they didn’t buy tools; they’ve built them in-house with governance, their own repositories and trained teams. 

The good news

💡 You no longer need those budgets to implement the same projects in your company; what you need is a method.

WHERE TO START

Choose a single process – the one that takes up the most hours or has the greatest impact on your profit and loss account – and ask yourself these three questions:

→ Is there documentation on how it currently works? 
→ Which part requires judgement and which part is purely repetitive? 
→ Where is the information needed to carry out the process or task?

If you can answer these, you’re ready to automate. If you can’t, you know what your first task is. And that, too, offers a huge return.

And in your company, which process would you automate tomorrow if you knew it would work?