“From” is the floor of each engagement. We quote an exact figure after a short discovery, once we know how many processes, integrations and exceptions are in scope. Prices exclude VAT; for business clients in the EU the reverse charge applies.
| Engagement | One-off | What is included |
|---|---|---|
| Start — strategy sprint | from €500 | team interviews, process audit, opportunity map scored by impact and effort, 3–6 month roadmap, implementation brief |
| Pro — idea to system | from €1,250 | automation of 2–3 processes or one AI agent integrated with your CRM or knowledge base, team training, documentation, one month of post-launch support |
| Business — ready to scale | from €3,000 | systems for several teams, custom integrations, a local model inside your perimeter if needed, ongoing optimisation and metric reports |
| Support | from €170/month | maintenance, tuning, metric reports |
Typical timelines: Start 1–2 weeks, Pro 3–6 weeks from kick-off to launch.
Most agencies do not publish prices, so there is no reliable “average”. These are published figures from German AI consultancies and an EU directory of AI agencies, collected in September 2026. Use them to see what sits behind different quotes, not as a benchmark.
| What you buy | Published range |
|---|---|
| Template website chatbot, one-off (Germany) | €990–2,499 net |
| Discovery workshop | €4,500–9,500 |
| Proof of concept or pilot | €16,500–35,000 |
| One custom AI agent, fixed-price consulting (30 person-days) | ≈€33,000 |
| Production system | €45,000–120,000 |
| Senior AI engineering day rate (Germany) | €1,050–1,800 |
| Hourly rates across 129 EU agencies that publish them | €20–260, median ≈€65 |
| Model and token spend quoted by a German agency | €1,000–6,000/month |
Source: public prices of German AI consultancies and an EU agency directory, September 2026. An industry cost guide from June 2026 puts Eastern European rates (Ukraine, Poland) at $50–90 an hour against $100–170 in Western Europe.
Our Pro engagement (from €1,250) is close to the price of a template chatbot and costs about as much as one senior engineering day in Germany. The difference is scope: we start small, on one process or one agent, and grow from there. Large production systems are a different product with a different budget.
After launch an AI agent costs money every month, even without a support plan: each reply is a model call billed per token. A token is a small piece of text, roughly part of a word; the provider counts both what the model reads and what it writes. Below is our estimate as of 30 September 2026, not a figure from an invoice. Model APIs bill in US dollars.
Assumptions. One conversation is about 6 customer messages. Instructions, knowledge-base passages and chat history add up to ~20,000 input tokens per conversation, replies to ~1,500 output tokens. No caching or batch discounts, so the estimate is conservative. Prices are the providers’ official list prices per million tokens.
| Model | $ per conversation | 1,000 conversations/month | 10,000 conversations/month |
|---|---|---|---|
| gpt-6-luna | $0.0028 | ≈$3 | ≈$28 |
| Gemini 3.5 Flash-Lite | $0.0098 | ≈$10 | ≈$98 |
| Gemini 3.8 Flash (until 31 Dec 2026) | $0.021 | ≈$21 | ≈$206 |
| Gemini 3.8 Flash (from 1 Jan 2027) | $0.041 | ≈$41 | ≈$413 |
| Claude Haiku 4.5 | $0.028 | ≈$28 | ≈$275 |
| Claude Sonnet 5.5 or gpt-6.1-sol | $0.055 | ≈$55 | ≈$550 |
How to read it. Real spend can differ from this estimate by a factor of 5–10, depending on the size of the knowledge base, the length of conversations and the model. Caching the fixed part of the prompt (instructions and knowledge base) cuts the cost of that input by up to ~90%.
If data has to be processed in a specific region, expect to pay more: one provider lists regional processing at +10%.
Model prices change. Gemini 3.8 Flash, for example, doubles its price on 1 January 2027. We recalculate against current prices before launch.
| Platform | Plan | Price | Limit |
|---|---|---|---|
| n8n Cloud | Starter | €20/month (billed yearly) | 2,500 executions/month |
| n8n Cloud | Pro | €50/month (billed yearly) | 10,000 executions/month |
| n8n self-hosted | Community | free + server ≈$6.49–8.99/month | no execution cap; server price on a 2-year prepay, renews at $11.99–14.99 |
| Make | Core / Pro / Teams | $9 / $16 / $29 per month | 10,000 credits |
| Zapier | Professional | from $19.99/month (billed yearly) | 750 tasks |
If every customer message triggers one workflow run, 1,000 conversations are ~6,000 executions a month. An execution is one run of a workflow in n8n; Make counts credits instead, roughly one per action in a scenario. That needs n8n Cloud Pro (€50) or your own server. At 10,000 conversations, self-hosted n8n on a ~$9–30 server is the realistic option.
Total, infrastructure only (estimate): 1,000 conversations a month ≈ $10–110; 10,000 conversations ≈ $40–580 with self-hosted n8n.
These are not quotes for your process. They show which parts a budget is made of. One-off figures are our price list; monthly figures are the estimates above.
| Scenario | One-off | Monthly (estimate) |
|---|---|---|
| Support agent on your knowledge base, one CRM, ~1,000 conversations/month | from €1,250 (Pro) | model ≈$3–55 + n8n Cloud Pro €50 or a server ≈$6–15 + optional support from €170 from month two |
| Sales agent: qualifies leads, writes to the CRM, hands off to a sales rep | from €1,250 (Pro) | model scales with the number of conversations, same logic; platform; optional support from €170 |
| 2–3 automations without chat: lead → CRM → task → report | from €1,250 (Pro) | platform from €20 (n8n Starter) or $9 (Make Core); model only where AI reads text; optional support |
| Several teams, custom integrations, local model | from €3,000 (Business) | hardware or private cloud quoted separately; support from €170 |
Monthly saving = manual hours per month × hourly cost − support − model − platform. Payback period = one-off cost ÷ monthly saving.
Example with made-up numbers (an estimate, not a client result). A team spends 60 hours a month on the task at €25 an hour: €1,500. Minus €170 support, ≈€25 for the model (Claude Haiku 4.5, 1,000 conversations) and €50 for n8n Cloud Pro leaves ≈€1,255 a month. A €1,250 implementation pays back in about one month.
This is an estimate, not a measured result: level V4 in our methodology. It ignores training time and the transition period while people still review the agent’s work. After launch we replace it with measured data.
To compare two quotes, convert both to first-year cost: one-off + 12 × monthly fees. For Pro under these assumptions: €1,250 + 11 × €170 support (month one is included) + ≈€75 × 12 for model and platform ≈ €4,020 in year one.
Know the risks up front. Gartner forecasts (June 2025) that over 40% of agentic AI projects will be cancelled by the end of 2027 due to escalating costs, unclear business value or inadequate risk controls. In McKinsey’s 2026 survey, one in five respondents say their organisation limits AI use because of operating costs. That is why we calculate the monthly bill before we start, not after.
The main barrier is usually not money. Among EU enterprises that considered AI but did not use it, 70.9% named lack of relevant expertise as a reason (Eurostat, 2025).
Our prices are from this site as of October 2026. Market ranges are public prices collected on 30 September 2026. Model and platform prices are from the providers’ official pricing pages on the same date. Cost and payback figures are calculations, not measurements (level V4 in our methodology).
Last reviewed: 1 October 2026.
With us, from €1,250 one-off for one agent integrated with your CRM or knowledge base, one month of support included. Published EU prices range from about €1,000 for a template chatbot to €16,500–35,000 for a consultancy pilot. Model usage and platform fees come on top every month.
Three parts: the model (our estimate: ≈$3–55 per 1,000 conversations depending on the model), the platform (€20–50 on n8n Cloud or ≈$6–15 for your own server) and optional support from €170. We calculate the exact figure for your volume.
“AI agent” covers very different things, from a template bot with no integrations to a system with a CRM, a knowledge base and human hand-off. What is included also varies: tokens, support, training. Ask for a breakdown and compare first-year cost.
Yes. The strategy sprint (from €500) gives you an opportunity map, a roadmap and a brief you can take to us or to anyone else. Agents also go live on part of the traffic first.
For one or two flows, an agency or freelancer is usually cheaper than a hire: in Germany one senior AI engineering day costs €1,050–1,800. In-house makes sense once AI work is continuous. Whoever builds it, make sure the workflows and accounts are in your name.
If the flow is linear and someone on your team will maintain it, yes, that is cheaper. A build pays off when there are many exceptions, several systems without ready connectors, or an AI step where a mistake costs money.
Start with a short brief — we reply with an estimate.