In short: in 2026 AI automation means software no longer just executes tasks by fixed rules but also reads, understands and decides: inbox triage, quotes, invoices, lead qualification, reporting. A first automation costs €2,950 to €7,500 and pays for itself in 2 to 6 months when it saves 1 day a week. Start with 1 process, measure the result and build from there.
What AI automation and AI agents mean in 2026
Classic automation follows rules: when a form comes in, add a row to the spreadsheet. That works as long as the input is predictable. The real work in a small or mid-sized business is not: emails in free text, PDFs in 40 layouts, a customer who wants 'something like last time'.
AI automation adds a language model that reads unstructured input and turns it into structured data and decisions. Three levels:
- An AI step in a workflow. An email arrives, the model determines category and urgency, the workflow does the rest. Cheap, reliable, the starting point for 80% of companies.
- An AI assistant. An employee asks in plain language 'draft a quote for customer X based on the previous one' and gets a proposal to approve. The human stays accountable.
- An AI agent. Software that is given a goal, uses tools itself (CRM, email, calendar, ERP) and executes multiple steps, within preset limits and with a human stepping in when in doubt.
According to Statistics Netherlands (CBS), almost a quarter of Dutch companies with 10 or more employees used AI in 2024, nearly double the year before. The difference between companies that benefit and companies that 'do something with AI' lies in the choice of process.
12 processes you can automate in 2026
For each process: what it does, an example and a realistic saving.
- Inbox triage. Read incoming email, categorize, prioritize, assign and prepare a draft reply. A transport company with 150 emails a day saves 2 hours a day on planning.
- Drafting quotes. Read the request, pull items and prices from the ERP, generate a draft quote in your house style. From 45 minutes to 5 minutes of checking per quote.
- Processing purchase invoices. Read PDFs, match them to purchase orders, post them in Exact or AFAS, flag discrepancies. 90% of invoices without manual entry.
- Lead qualification. Read the website form and LinkedIn profile, score for fit, enrich with Chamber of Commerce data, add to the CRM and start the right follow-up. Sales response time from 2 days to 10 minutes.
- Answering customer questions. An AI chatbot or email assistant that answers from your own documentation and order data, and hands over when unsure. 40 to 60% of questions without an employee.
- Reporting. Merge data from accounting, CRM and planning into a weekly dashboard with a written summary of the deviations. From 4 hours a week to 0.
- Onboarding customers or employees. Request documents, check them, create accounts, send welcome emails, assign tasks. At an accounting firm including the anti-money-laundering (Wwft) check and data processing agreement.
- Scheduling and appointments. Read requests, check availability, schedule, confirm, remind and reschedule on changes.
- Contract and document review. Summarize contracts, policies or tenders, flag deviating clauses, extract data into your system.
- Accounts receivable. Monitor open invoices, send reminders in the right tone per customer, record payment arrangements, escalate after 60 days.
- Content and marketing. Product descriptions from specifications, social posts from news items, a newsletter from your blog, always with human final editing.
- Quality and compliance checks. Check work orders, inspection reports or case files for completeness before they go to the customer or the regulator.
Tools: n8n, Make, Zapier or custom
| Tool | Cost 2026 | Strong at | Weak at | Data and GDPR |
|---|---|---|---|---|
| n8n | Free self-hosted (Sustainable Use License), cloud from about €20 per month | Complex flows, code steps, AI agents, your own server | Steeper learning curve, admin when self-hosting | Self-hosted in the EU: data stays in your own environment |
| Make | Free up to 1,000 operations, paid from about €9 per month | Visual flows, many connectors, quick start | Costs climb with volume, limited error handling | EU data centers available |
| Zapier | Paid from about €20 per month, rising quickly | Largest number of connectors, simplicity | Expensive at high task volumes, little control over logic | Processing in the US, standard contractual clauses needed |
| Power Automate | From about €15 per user per month | Microsoft 365 environments, approval flows | Limited outside Microsoft, licensing maze | Microsoft EU Data Boundary |
| Custom (Node.js, APIs, language model) | €7,500 – 25,000 build, no licenses | Everything the tools cannot do, scale, ownership | Higher upfront cost, developer needed | Fully under your control, EU hosting of your choice |
Our default choice for SMEs in 2026 is n8n, self-hosted on a European server. You pay no license per task, your data stays in-house and the tool scales with code steps and AI agents far beyond what Make and Zapier can handle. Zapier and Make are fine to start with if you want no admin and volumes are small. Custom comes into view when the logic is business-critical or you run more than 50,000 tasks a month.
Choosing between standard tooling and custom is the same trade-off as in custom software vs off-the-shelf.
Costs and payback time
A realistic cost picture for a first automation:
- Build: €2,950 to €7,500 for 1 process including analysis, build, tests and handover. An AI assistant or agent with 3 integrations costs €7,500 to €15,000.
- Tooling: €0 to €100 per month for n8n or Make.
- AI usage: €10 to €200 per month in API costs. Classifying an email costs a fraction of a cent; generating a 3-page quote a few cents.
- Hosting: €10 to €50 per month for a European VPS.
- Management: €450 to €950 per month for monitoring, improvements and new integrations, or 2 to 4 hours of your own time.
Worked example: an installation company automates purchase invoice processing. Build €4,500, running costs €120 per month. Saved: 12 hours a week of administration at a fully loaded €45 per hour, or €2,300 per month. Payback: 2 months.
Data and GDPR: where your data lives
Every automation that handles customer data falls under the GDPR. Four questions to answer before the build:
- Which data goes to the language model? Send only what is needed. To categorize an email the subject line and first paragraph are enough; the address and phone number are not.
- Where is it processed? Choose a provider with EU data processing or a European model (Mistral, for example) and sign a data processing agreement. Business APIs from the major providers do not use your data for training by default, but put it in writing.
- Who decides? For decisions with legal effects on people (rejecting a candidate, refusing credit) a human must decide. Let AI prepare, not decide.
- What does the AI Act say? The European AI regulation applies in phases. Since February 2025 the ban on unacceptable uses and the AI literacy duty for staff apply; from August 2, 2026 the transparency obligations apply, among others: people must know they are dealing with AI. For most SME automations that is manageable, provided you design for it up front.
Step-by-step plan: your first automation in 2 weeks
- Days 1–2: choose and measure. Pick 1 process. Measure how much time it takes now, how often it goes wrong and what the input is.
- Days 3–4: design the flow. Trigger, steps, AI decisions, exceptions, who checks what. On 1 page.
- Days 5–8: build and test on historical data. Run 50 real emails or invoices from last month through the flow and compare the outcomes with what a human did.
- Days 9–10: shadow run. The automation runs alongside, a human approves every outcome. Threshold: 95% correct.
- Days 11–14: live with a safety net. Exceptions go to a human, everything is logged, evaluation after 2 weeks with the measured figures.
When to bring in an agency
Building it yourself in Make or Zapier is feasible for simple flows without AI decisions. An AI integration specialist makes sense when:
- the flow makes decisions about customers, money or staff and you want to be sure of the safety nets;
- integrations are needed with ERP, accounting or industry software without a ready-made connector;
- you want to keep data under your own control (self-hosting, GDPR, data processing agreements);
- you want to automate more than 3 processes and prefer 1 architecture over 12 loose zaps;
- nobody in-house has time to manage it after the build.
RABZI estimate: an Automation Sprint from RABZI costs €2,950 excl. VAT (1 process, working and documented in 2 weeks), from €7,500 for an AI assistant or agent with 3 integrations and from €19,500 for a company-wide setup with your own knowledge base (RAG). Management and expansion through an AI Ops retainer from €950 per month. See AI integration and automation, integrations and APIs or the full pricing.
Frequently asked questions
How much does AI automation cost for a small business? A first automation of 1 process costs €2,950 to €7,500 excl. VAT, an AI assistant or agent with several integrations €7,500 to €15,000. On an ongoing basis you pay €50 to €300 per month for tooling, AI usage and hosting, plus management if you outsource it.
What is the difference between n8n, Make and Zapier? Zapier is the simplest and has the most connectors, but gets expensive at volume. Make is visual and affordable for mid-sized flows. n8n is the most powerful, can run free on your own server and keeps your data in the EU, but needs a bit more technical knowledge or a partner for management.
Is AI automation GDPR-compliant? Yes, provided you send only necessary data to the model, choose a provider with EU processing and a data processing agreement, have a human make decisions about people and preferably run the workflows self-hosted in the EU. Record it in your processing register.
Which process should I start with? A process that happens daily, is tedious and has digital input: inbox triage, quotes or purchase invoices. Measure how much time it takes now, so that after 2 weeks you can prove the saving and have buy-in for processes 2 and 3.
Conclusion
For SMEs in 2026 AI automation is no longer an experiment but a calculation: a process that eats time every day, a build of €2,950 to €7,500 and a payback of a few months. Pick 1 process, keep your data under your own control, let a human make the decisions that matter and expand from there. Want to know which process qualifies first in your business? Book an intro call and you will have a concrete proposal for a first sprint within 2 business days.