16 Julio 2026
How much does it cost to implement AI in a company? Subscription vs. owned systems
Every company looking to get started with automation asks this question: “What is the estimated budget for an AI-driven process automation project?”
And almost all the answers you hear sidestep the important issues.
So I’m going to answer it just as we do for our clients: by setting out the actual cost structure and explaining the decision that really determines how much you’ll end up paying.
1. WHAT DETERMINES THE PRICE (IT'S NOT THE TECHNOLOGY)
An AI-powered process automation project has three cost components:
Component 1 — Process Design. The part that almost no one budgets for: mapping how it works today, deciding what needs human input and what doesn't, redesigning before automating. It's 80% of the project's success and a fraction of the cost. Skipping it is the number one cause of failed AI projects, and these are the ones that are expensive.
Component 2 — System Construction. The agent or flow itself: solution architecture, system integrations, decision logic, human oversight, testing. The range depends on the complexity: classifying customer inquiries doesn't cost the same as reconciling invoices per file with disbursements and currencies.
Component 3 — Recurring Costs. And here's the important decision…
💡 What you can do:
- Map the process as it is TODAY (not as it should be)
- Identify who does what, with what information, and using what tools
- Separate which decisions require human judgment and which do not
- This is the least glamorous part of the project, but it determines whether automation will bring benefits or not.
2. WHAT TO AUTOMATE FIRST? THE 30% RULE
There are two ways to pay for automation. And their cost curves are opposite.
Option A — Rent (one tool or program per process). You hire one AI tool for customer service. Another for prospecting. Another one for invoices.
→ Each one, their monthly subscription — which grows with your usage, your contacts, or your team.
→ After two years: a drawer full of subscriptions that add up to more than any project cost.
→ Your data, scattered in black boxes that no one internally maintains.
→ And if you stop paying, the system disappears.
Option B — Build your own system. The system is built on open-source tools — in our case, n8n self-hosted on your infrastructure — documented and auditable.
→ You pay for the project once.
→ The recurring cost is reduced to infrastructure (tens of euros per month, not hundreds per user) and AI consumption at cost price, without intermediaries.
→ When the project ends, the system is yours. Like your ERP. Like your ship.
The major Catalan industrial companies leading the adoption of AI understand this clearly: they didn't buy tools, they built an AI system—governance, their own agent repositories, and trained teams.
💡 The good news for businesses:
You don't need your budget to apply the same principle. You need to start with a process.
3. HOW TO PRIORITIZE SO THAT THE PROJECT PAYS FOR ITSELF
Once the use cases have been identified, it's important to audit and score each proposal to ensure that all initiatives are aligned with the company's fundamental objectives. For example, each initiative could be scored against these principles:
→ Alignment with strategy
→ Technical feasibility
→ Impact on the profit and loss statement
→ Improvement of competitiveness
In a small business context: start with the process that has the greatest impact on your profit and loss statement, not the most visually appealing one. The questions we ask in a diagnostic assessment are:
→ How many hours per month does this process consume, and whose time is it? → What is the cost of each error? (A typo in the order, an unmatched invoice, an unanswered inquiry)
→ What would you stop losing if the answer came back in minutes instead of days?
When the process is well-chosen, the return doesn't depend on optimistic projections. It depends on multiplying hours by hourly cost.
4. HONEST RANKS
Every project is different, and be wary of anyone who gives you a price without seeing your process. What I can tell you honestly is this:
→ A first fully automated end-to-end process—the "minimum that works"—is a project of weeks, not months. And thousands of euros, not tens of thousands.
→ Process design is always included: we don't build on unmapped processes.
→ The recurring cost of owning a system is a fraction of the equivalent in subscriptions. And it doesn't grow with your team.
→ There are also public funding options for the digitalization of SMEs that can cover part of the project. I'll let you know in the assessment if your case qualifies.
If you want to transform your business with AI
Book a free taster session