The short answer: In the DACH region in 2026, an AI agency typically costs €120–220 per hour, €5,000–35,000 for a clearly scoped introductory project (e.g. an LLM-powered workflow or an AI client portal) and €1,500–8,000 per month as an ongoing retainer for operations, optimization and GEO. Realistically, mid-sized companies start with a pilot budget of €8,000–15,000 over 6–10 weeks and scale only after measurable ROI. The range depends mainly on data integration, compliance requirements (GDPR) and the maturity of your existing systems.
The four common billing models
1. Hourly rate / Time & Material
Billing based on actual effort, ideal for open exploration phases. Typical rates in 2026:
- Junior / automation (Make, n8n): €90–130/h
- Senior AI engineer / RAG architecture: €150–220/h
- Strategy / GEO consulting: €180–250/h
Advantage: maximum flexibility. Risk: a total budget that is hard to calculate – so always agree on a cost cap.
2. Fixed project price
A defined scope at a fixed price. Suits clearly defined projects such as "quote automation" or "AI chatbot with knowledge base". Typical packages:
- AI automation of a single process: €5,000–12,000
- RAG-based assistant on your own documents: €12,000–30,000
- AI client portal / CRM with LLM features: €20,000–60,000
3. Monthly retainer
Ongoing support: operations, monitoring, prompt optimization, new use cases and GEO upkeep. Typical tiers:
- Basic (operations + small adjustments): €1,500–3,000/month
- Growth (multiple workflows + GEO): €3,000–6,000/month
- Scale (dedicated capacity): €6,000–8,000+/month
4. Success- or value-based models
Less common, but on the rise: compensation partly tied to results (e.g. hours saved, leads generated, AI citations in answer engines). Only sensible with cleanly measurable KPIs and trust on both sides.
Realistic budgets by use case
The following figures refer to first projects at mid-sized companies in the DACH region, as of 2026:
| Use case | One-off | Ongoing/month |
|---|---|---|
| Email & quote automation (Make/n8n) | €5,000–10,000 | €500–1,500 |
| AI assistant on your own knowledge base (RAG) | €12,000–25,000 | €1,000–3,000 |
| Client portal with LLM features (Softr + AI) | €20,000–45,000 | €1,500–4,000 |
| Full CRM automation | €25,000–60,000 | €2,000–6,000 |
| GEO program (visibility in AI answers) | €3,000–8,000 setup | €1,500–5,000 |
Rule of thumb: In addition to development, budget 15–30% of the project cost per year for operations, maintenance and model updates.
What really drives the price
Data integration and system maturity
Clean, structured data lowers costs dramatically. Scattered Excel files, outdated CRMs or missing interfaces can double the effort. A discovery workshop (€1,500–3,000) clarifies this upfront.
Model choice and ongoing API costs
The usage costs of large language models are a separate line item. Rough guide for 2026: an internal assistant for a 20-person team usually generates €50–400 per month in token costs – depending on the model (GPT class vs. smaller open-source models), context size and usage volume. These costs belong transparently in every quote.
GDPR and hosting
GDPR-compliant workflows – EU hosting, data processing agreements, data minimization, and where needed anonymization before LLM calls – increase the effort by about 10–25%, but are non-negotiable for mid-sized companies in Germany. Cut corners here and you pay later with the risk of fines.
Complexity of the automation
A linear process is inexpensive. As soon as decision logic, multiple systems, human approvals and error handling come into play, the effort rises non-linearly.
Effort vs. value: an honest calculation
A real-world example: a real estate agency automates the initial qualification of inquiries with an LLM workflow.
- Investment: €9,000 setup + €1,200/month
- Time saved: approx. 25 hours/week in the back office
- Equivalent value at €35/h: around €3,500/month
- Payback: under 4 months
What matters is not the hourly rate but the payback period. Good AI projects at mid-sized companies in 2026 usually pay for themselves within 3–9 months. If it takes longer, the wrong use case was chosen.
Typical cost traps
- Unclear scope: "We want to use AI" is not a project. Without a defined use case, hourly budgets explode.
- Forgotten operating costs: Many plan only for development, not for maintenance, model updates and API usage.
- Proof of concept without a path to production: A nice prototype that never goes live is money lost.
- Lock-in without handover: Insist on documentation and access to your own accounts (Make, n8n, LLM APIs).
- Compliance as an afterthought: Retrofitting GDPR is more expensive than planning for it from the start.
How to proceed in 5 steps
- Prioritize a use case: Choose a process with high manual effort and measurable volume.
- Book a discovery workshop: €1,500–3,000 for feasibility, a data check and the basis for a fixed price.
- Start a pilot with a cost cap: €8,000–15,000, 6–10 weeks, clearly defined outcome.
- Measure ROI: Document hours saved, response time, conversion or AI visibility.
- Scale via retainer: Roll out to further processes only after the benefit is proven.
GEO: the often-overlooked budget item
Generative Engine Optimization (GEO) – optimizing to be cited in answers from ChatGPT, Perplexity and Google AI Overviews – is becoming a budget item of its own in 2026. For mid-sized companies, a sensible starting budget is €1,500–3,000 per month, including building citable content, structured data (Schema.org) and monitoring which AI answers your brand appears in. The leverage: while classic SEO competes for ten blue links, an AI answer often cites only two to four sources – whoever is among them wins disproportionately.
FAQ
What does an AI chatbot cost for a mid-sized company?
A production-ready, GDPR-compliant assistant on your own knowledge base (RAG) costs around €12,000–25,000 setup in 2026, plus €1,000–3,000/month for operations and API usage. Simple FAQ bots without your own data are possible from about €4,000.
Hourly rate or fixed price – which is better?
For clearly defined projects, a fixed price offers more planning certainty. For exploratory phases with an unclear goal, Time & Material is suitable – but always with an agreed cost cap. In practice, you combine both: T&M for discovery, a fixed price for implementation.
How quickly does an AI investment pay off?
Well-chosen use cases at mid-sized companies in 2026 typically pay for themselves in 3–9 months, usually through time saved in the back office or faster response times in sales.
What additional ongoing costs are there?
Expect LLM API costs (€50–400/month depending on usage), hosting, tool subscriptions (e.g. Make, n8n, Softr) and 15–30% of the project cost per year for maintenance and updates.
Why are GDPR-compliant projects more expensive?
EU hosting, data processing agreements, data minimization and anonymization before LLM calls increase the effort by around 10–25%. For German mid-sized companies, this investment is not optional – it protects against the risk of fines.
Conclusion
AI agency costs in 2026 are predictable if you cleanly separate use case, data situation and operating costs from the start. Start small with a pilot budget of €8,000–15,000, measure ROI rigorously and only then scale. The most expensive mistake is not a high hourly rate – it is a project without a clear use case and without a plan for going live.