What process automation actually means (and what it isn't)
Process automation is software that executes a repeatable sequence of steps a person used to do by hand. You give it a trigger—a new email, a new spreadsheet row, a new ticket—it runs through a fixed set of rules, and it writes the result back to your system of record. It's not an AI assistant you chat with. It's not a chatbot. It runs silently on a schedule or event, not on conversation.
Here's the shape: a trigger arrives, a sequence of steps unfolds, AI or plain logic transforms the data, and the output lands in Holded, HubSpot, Google Sheets, or wherever you track that work. Most of the cost is plumbing—wiring APIs, parsing PDFs, connecting your tools—and maybe 30% is AI extraction or drafting. You'll likely use n8n, Make, or Zapier to glue it together, and GPT-4 class models to handle the judgment calls. The result saves someone an hour a day or more.
How to pick the first process: boring, repetitive, and owned by one person
Don't try to automate your entire operation. Use this filter: Does it happen more than 10 times a week? Are the rules written down, or can one person write them down in an afternoon? Does one person own the process today? Would a mistake cost under €500?
If you're nodding at three or four of those, you've found a candidate. If you're reaching for something "strategic" or "integral to our growth," stop. Strategic processes have exceptions, they touch six departments, and nobody agrees on the happy path. They will wreck your first automation. Save them for month four.
The hour-a-day test is your anchor: if automating the process saves one person one hour per day, that's roughly €6,000–€10,000 per year in loaded cost at a Spanish SMB (€25–€30/hour salary plus benefits). A €5,000 build pays for itself in two months. That's the math you need.
Four processes where SMBs actually see payback in month one
Quotes. Inbound RFQ email arrives → extract product, quantity, customer name with AI → cross-check your price list → draft a branded PDF quote → send it to your sales manager for a one-click approval. Typical save: 15–30 minutes per quote, and most SMBs handling 20–50 quotes per week. The process repeats. The rules don't change. One person (your sales ops lead or the sales director) owns the exceptions.
Invoices. Supplier sends a PDF invoice via email or portal → your AI extraction reads line items, dates, amounts → match it against your purchase order in the ERP → push it into Holded, Odoo, or Xero → flag mismatches for a human to review. The real win is removing the 2–4 hours per day your finance person spends copying numbers from PDF into a spreadsheet, then into your system. One person owns the exceptions (the finance manager). Our guide to what AI automation costs in Spain covers invoice automation in detail.
Reporting. Every Monday morning, pull sales data from your CRM, operational metrics from your project tracker, headcount from your HR system → merge them → generate a one-page summary with commentary (e.g., "sales +12% week-over-week, Q2 hiring plan is on track") → email to leadership. This replaces the Sunday-night Excel marathon. The ops manager owns the exceptions (missing data, weird numbers).
Tier-1 support. Incoming ticket arrives → classify it (FAQ, refund request, technical issue, feature request) → if it's an FAQ, auto-reply with hours, order status, or returns policy → if it's not, route it to the support team with context. This covers maybe 30–40% of your tickets automatically. Humans still own edge cases and review automation replies. The support lead owns the tuning.
What a first project costs in Spain (and what the money buys)
Assessment and scoping: €1,500 for one week of work where we sit with your team, map the current process, sketch the automation, and produce a build spec with a go-or-no-go recommendation. You walk away knowing exactly what you're building and whether it's worth it.
First build: €3,000–€8,000 for a single process with one or two integrations. Invoice automation or quote generation lives at the lower end. Add a third system, or if the PDFs are messy and need OCR tuning, you'll hit €8,000–€15,000. What drives the price up is the number of systems you're wiring, PDF quality (garbage in, garbage out), exception volume, and whether you need custom user interfaces or on-premises data handling.
Ongoing costs most people forget: LLM API spend (€50–€400 per month depending on volume), orchestration platform hosting (€20–€100 per month), and 2–4 hours per month of maintenance and tuning. A red flag in any quote is a vendor who doesn't name your internal owner, doesn't define your success metric in euros or hours, and doesn't include a rollback plan.
What breaks after go-live—and who should own it
Three things break in the first month. Upstream formats change: your supplier redesigns their invoice template, or your CRM changes a field name. API credentials or auth tokens expire. Edge cases the pilot didn't see appear in live traffic—a customer name with special characters, an invoice that's half in euros and half in dollars.
You need a named internal owner, usually the ops manager or the person who owned the manual process. They get the alert every time the automation flags an exception or fails silently. Set up a 15-minute weekly review in month one to triage what the automation sent to the exception queue. Budget for a "month 2 tune-up": small SMBs typically find 10–20% of cases need rule adjustments after the first month of real traffic.
Never let an automation fail silently. Every workflow needs a heartbeat alert—if no new jobs run for 24 hours, something is wrong—and a dead-letter queue where exceptions land for a human to handle.
A sensible 90-day plan to go from zero to one working automation
Weeks 1–2: List every repetitive process in your business. Score each one with the four-question filter. Pick one. Write a one-paragraph description of what the person does today.
Weeks 3–4: Scoping. Document every step the current owner takes, from the moment the trigger arrives to the moment the work is done. Agree on one success metric: "cut quote turnaround from 2 days to 2 hours" or "save 10 hours per week in invoice entry." Agree on who owns exceptions and how they'll be alerted.
Weeks 5–8: Build and test in parallel with the manual process. Your automation runs in shadow mode—it produces output, but a human reviews everything before it's sent or written to your system. You're checking for accuracy and edge cases.
Weeks 9–10: Cut over for a subset. One product line, one region, one customer tier. If it's quotes, automate them for one product category. If it's invoices, start with one supplier.
Weeks 11–13: Measure against your baseline metric. Decide whether to expand scope (more suppliers, more product lines, more ticket types) or pick the next process. Then repeat.
One process, one owner, one metric. That's the formula. When it works, the next one is cheaper and faster because you've already wired the integrations.
If you're ready to start, send us your top three candidate processes. We'll scope the best one in a week for €1,500 and tell you exactly what a build costs and how much time you'll save. Get in touch and let's scope it together.