A single automated workflow built by an agency costs EUR 2,000 to EUR 6,000. A multi-system build with three or more integrations costs EUR 10,000 to EUR 25,000. Running it afterwards costs EUR 40 to EUR 600 a month in model and platform fees. Those three numbers are the whole answer, and the rest of this article shows exactly where each euro goes so you can read any quote you receive and spot what is missing.
What each project type costs to build
Website chatbot on your own content: EUR 2,000 - 4,000, 2-3 weeks. Content ingestion, prompt design, chat UI, handover to a human, one CRM connection.
One automated workflow, e.g. lead qualification: EUR 2,500 - 6,000, 2-4 weeks. Trigger, scoring logic, CRM write-back, notifications, monitoring dashboard.
Document and invoice processing: EUR 5,000 - 12,000, 4-7 weeks. Extraction schema, validation rules, human review queue, accounting-system export.
Internal knowledge assistant (RAG): EUR 6,000 - 14,000, 4-8 weeks. Document pipeline, vector store, access control per team, source citations.
Multi-system automation with 3 or more integrations: EUR 10,000 - 25,000, 6-14 weeks. Several agents, orchestration, error handling, audit log, staff training.
These are build prices for a system you own. They are not licence fees, and they do not include your ongoing model usage.
The four pricing models, and when each one is fair
Fixed project price. One number for a defined scope. Fair when the scope is genuinely known: a chatbot, one workflow, one integration. Insist on a written deliverables list, because the whole protection of a fixed price is that scope is written down.
Time and materials, EUR 70 - 150 an hour. Fair for discovery, for R&D where nobody knows the answer yet, and for small changes after launch. Not fair as the billing model for a whole first build, because you carry all the risk of their learning curve. If an agency will only work hourly on a first project, ask why they cannot price something they claim to have built before.
Monthly retainer, EUR 500 - 3,000. Fair after launch, covering monitoring, prompt tuning, model upgrades, and a set number of change hours. Ask exactly how many hours it includes and what happens to unused hours. A retainer with no hour count is a subscription to being someone's client.
Outcome-based, e.g. a share of hours saved or of new revenue. Rare and hard to measure honestly. It works only when the metric existed and was measured before the project started.
Running costs: the part that surprises people
The build is one-off. These recur, and in most quotes they are yours to pay directly, not the agency's:
- LLM API usage: EUR 20 - 400 a month. A chatbot handling 1,000 conversations a month on a mid-tier model typically lands at EUR 30 - 80. Document processing at volume, or long-context reasoning tasks, is what pushes this to the top of the range.
- Automation platform: EUR 0 - 100 a month. Make.com and n8n cloud plans sit at EUR 10 - 60 for normal SME volumes; self-hosted n8n is a server cost instead, roughly EUR 15 - 40.
- Vector database or search: EUR 0 - 70 a month. Free tiers cover most SME document sets.
- Hosting and monitoring: EUR 10 - 50 a month.
- Support retainer: EUR 0 - 3,000 a month, depending on whether you take one.
For a typical SME running two or three automations, total monthly running cost is EUR 60 - 250 excluding any retainer. Get this number in writing before you sign, separately from the agency's fee. An agency that cannot estimate your running cost has not modelled your volume.
The six line items that get hidden
- Discovery billed as a project. Some agencies charge EUR 1,500 - 5,000 for a strategy document, then quote the build separately. Ask whether discovery is credited against the build if you proceed. Ours is free, precisely because a paid audit that leads nowhere is worth nothing to you.
- API accounts in the agency's name. If the OpenAI or Anthropic account is theirs, your usage bill passes through them, often with a margin, and you cannot leave without a migration. Accounts should be yours from day one.
- Prompts and code not handed over. Ask directly: after final payment, do we receive the repository, the prompts, and the workflow exports? If the answer is qualified, you are renting.
- Integration work priced per connector after the fact. Get the connector list into the contract. "Connects to your CRM" is not a scope line; "writes lead score and source to HubSpot contact properties X, Y, Z" is.
- Rework after model changes. Models are deprecated on a 6-18 month cycle. Ask who pays to migrate when that happens.
- Training and documentation. Usually 4 - 10 hours of work. If it is not in the quote, it will be a change request.
How to compare two quotes properly
Put both on the same three-year basis:
Total cost = build price + (monthly running cost x 36) + (retainer x 36) + expected rework
A EUR 4,000 build with a EUR 1,500 monthly retainer costs EUR 58,000 over three years. A EUR 12,000 build with a EUR 200 retainer costs EUR 19,200. The cheaper build is often the more expensive relationship, and the retainer line is where the difference hides.
Then check the payback. If a workflow saves one person 8 hours a week at a EUR 25 fully loaded hourly cost, that is EUR 866 a month. A EUR 5,000 build pays back in under six months. If an agency cannot put a saved-hours number against their own proposal, they have not understood your process well enough to automate it.
What we quote and how
Discovery and the concept are free. You get an itemised fixed price per system, the connector list in writing, your own API accounts, full handover of code and prompts at final payment, and a monthly running-cost estimate based on your actual volume. First system live in 4 weeks.
Related reading
- What Is an AI Agency? A Guide to AI Automation for Business
- AI Automation Costs and ROI for Small Businesses
- How to Choose an AI Automation Agency
Ready to price your own project? See our AI automation agency services, delivery process and starting point.