How to automate an SME in Barcelona: 7-step guide

Where an SME wanting to automate should start

Most SMEs and freelancers in Barcelona and Catalonia don't have a technology problem — they have a poorly defined process problem. Before talking about AI, agents or integrations, you need to know exactly what is being done, who does it and with which tools. Automating without this first step usually reinforces a bad process rather than improving it.

This guide covers the 7 steps followed by any well-planned automation project, from the initial inventory to long-term maintenance. It's not a list of trendy tools: it's the practical criteria that decide whether a project delivers a return or not.

Step 1: build a real inventory of your processes

Write down, process by process, what comes in, what goes out and who's involved: invoicing, bank reconciliation, quote responses, report generation, order entry, support ticket handling. You don't need any sophisticated tool for this: a simple spreadsheet with columns for 'process', 'frequency', 'owner' and 'tools used' already gives you the full picture.

The most common mistake at this stage is skipping it because 'we already know how we work'. In practice, once you document things, hidden manual steps always turn up — forwarded emails, manual exports and imports between systems — that nobody had counted as real work.

Step 2: measure the real time you spend

For each process in your inventory, track for one or two weeks how much time each person spends on it and how often. You don't need stopwatch precision: an honest estimate in minutes or hours per run is enough.

This data lets you turn a feeling ('we spend too much time invoicing') into a number you can compare against the cost of automating it. Without this measurement, any later decision is based on intuition, not data.

Step 3: prioritise by impact and effort

With the inventory and the measured time in hand, classify each process along two axes: the impact of solving it (time saved, errors avoided, better customer experience) and the effort it would take to automate (number of systems involved, whether an API exists, complexity of the business rules).

Step 4: choose between an existing tool or custom development

Not every process needs bespoke software. If the process is standard (invoicing, basic CRM, ticket management) and volume is moderate, it's often faster and cheaper to adopt a well-configured existing tool. If the process is specific to your business, connects systems with no native integration, or requires its own logic (business rules, validations, AI-based decision criteria), custom development tends to pay off in the medium term because it avoids paying for features you won't use and doesn't depend on a third party's limitations.

At AurIA42 we often work through this decision with SMEs in Barcelona and Catalonia via our process automation for SMEs page, where we detail which kinds of processes tend to benefit more from custom development versus a generic tool.

Step 5: test a scoped pilot before scaling

Once you've chosen a priority process, don't try to automate everything at once. Define a pilot with a small, measurable scope: one type of invoice, one order-entry channel, one type of report. A well-scoped functional pilot can usually be delivered in around 4 weeks, at an indicative cost of between €4,000 and €12,000 depending on complexity and the number of systems involved.

The pilot serves to validate two things: that the technical solution works with real data, and that the team actually adopts it. Many projects fail not because of the technology, but because nobody ends up using it day to day.

Step 6: roll out and train the team

Once the pilot is validated, extend it to the remaining cases and formalise team training: who reviews exceptions, who maintains access credentials, who decides when an automation needs to be temporarily disabled. Without this step, the automation becomes a 'black box' that nobody understands and that generates distrust when it fails.

This is also the point to document the automated process with a simple diagram and a short manual, so that dependency doesn't rest solely on whoever built it.

Step 7: measure results and plan maintenance

Automating isn't a project that ends: it's a system that needs upkeep. APIs change, data volumes grow, and business rules evolve. That's why it makes sense to have ongoing maintenance from €79/month, covering monitoring, small adjustments and incident resolution.

Periodically measure the actual time saved (comparing it against the step 2 measurement) and errors avoided. This data lets you decide whether it's worth extending automation to other processes in your initial inventory.

Common mistakes that drag out or sink a project

• —Automating a poorly defined process instead of simplifying it first.
• —Trying to automate everything at once without going through a pilot.
• —Not assigning anyone responsible for maintenance once it's launched.
• —Choosing a generic tool for a process with very specific business rules.
• —Not planning for what happens when the automation fails or receives unexpected data.

Risks, limitations and when NOT to automate

Automation has real limits. If a process keeps changing because it isn't defined yet, it's better to stabilise it first. If volume is so low that the time saved doesn't cover the cost of maintenance, it doesn't make sense to automate it yet. And if the process involves decisions with a sensitive legal or human-judgement component (for example, final decisions about customers), you should always keep a human review step before executing the final action.

Regarding personal data, any automation that handles customer or employee data must comply with the General Data Protection Regulation (GDPR). Among other things, this means limiting the automation's access to only the data it needs, documenting which processing activities take place and under what legal basis, and planning how access, rectification and erasure rights are exercised when part of the process is automatic. If the automation includes third-party AI tools, you should check where the data is processed and for how long it's retained.