---
title: "What Does an AI Project Cost in the Mittelstand?"
source: https://bitvaria.com/en/was-kostet-ki-projekt-mittelstand
---
24\. März 2026 · Daniel Schleipfer · [AI](https://bitvaria.com/category/ai/) · 5 min read
# What Does an AI Project Cost in the Mittelstand?
Custom AI applications for mid-sized companies cost between 15,000 and 60,000 euros. This article shows which factors determine the price and when the investment pays off.
A custom AI project for a mid-sized company costs between 15,000 and 60,000 euros. The running costs come to 500 to 2,000 euros per month. The return on investment typically shows up after 3 to 6 months.
That is the short answer. The longer one depends on exactly what you have in mind.
## Why the Price Range Is So Wide
There are worlds between an AI application that classifies incoming documents and one that generates complex quotes from historical data. Both are custom AI projects. Both deliver measurable value. But the effort involved differs significantly.
```mermaid
%%{init: {'theme':'dark'}}%%
graph LR
A[Simple application
15,000 - 25,000 €] --> B[Medium complexity
25,000 - 40,000 €]
B --> C[Complex application
40,000 - 60,000 €]
A1[One data source
Clear rules
Internal tool] -.-> A
B1[Multiple systems
Context-dependent
Broader user base] -.-> B
C1[ERP integration
Complex logic
Product UX] -.-> C
```
Three factors determine the price:
**1\. Complexity of the use case**
Classifying documents into clear categories is technically simpler than a system that has to interpret free text from various sources and respond depending on context. The more decision logic involved, the more development effort.
**2\. Integration with existing systems**
The AI application itself is rarely the biggest cost driver. Integrating it into your existing IT landscape is. When data has to be pulled together from an ERP system, a database, and an email inbox, the effort goes up. When a single data source is enough, it goes down.
**3\. Requirements for the user interface**
An internal tool used by 5 trained employees needs less UX effort than an application meant for 200 employees to use day to day. The broader the user base, the more the interface matters.
## Concrete Cost Examples
The following examples are based on real project sizes for companies with 200 to 2,000 employees.
### Document Classification and Routing
Incoming emails, invoices, or requests are automatically categorized and routed to the right department.
- **Investment:** 15,000 to 30,000 euros
- **Running costs:** 500 to 1,500 euros per month (API usage and hosting)
- **Time saved:** 85 to 95 percent per document
- **ROI period:** 3 to 5 months
- **Requirement:** At least 50 documents per day
### Automated Quote Generation
Sales receives AI-generated quote drafts based on historical quotes, price lists, and customer-specific terms.
- **Investment:** 20,000 to 40,000 euros
- **Running costs:** 800 to 2,000 euros per month
- **Time saved:** 60 to 75 percent per quote
- **ROI period:** 4 to 6 months
- **Requirement:** At least 20 quotes per week
### Internal Knowledge System
Employees ask questions to a system that assembles answers from internal documents, wikis, and emails. Often implemented as a RAG system.
- **Investment:** 25,000 to 50,000 euros
- **Running costs:** 1,000 to 2,000 euros per month
- **Time saved:** typically 2 to 4 hours per employee per week
- **ROI period:** 4 to 8 months
- **Requirement:** An existing, digital knowledge base (documents, wiki, intranet)
## The Typical Project Timeline
```mermaid
%%{init: {'theme':'dark'}}%%
gantt
title AI project: typical timeline
dateFormat YYYY-MM-DD
axisFormat %b
section Phases
Discovery (1-2 weeks) :a1, 2026-01-06, 14d
Development (4-8 weeks) :a2, after a1, 42d
Testing with real data :a3, after a2, 14d
Deployment and go-live :a4, after a3, 7d
section ROI
Investment recovered :milestone, 2026-06-01, 0d
```
## What Is Included in These Costs
A serious AI project involves more than programming. At bitvaria, every project includes:
1. **Discovery:** Analysis of the use case, assessment of the data situation, definition of scope. This step decides between success and failure.
2. **Development:** Concept, architecture, implementation, and integration with existing systems.
3. **Interface:** A user interface your employees will actually use. Not a demo, but a tool for everyday work.
4. **Testing:** Validation with real data and real scenarios.
5. **Deployment:** Installation in your infrastructure or in the cloud, depending on your requirements.
## What Is Not Included in These Costs
**Training your own AI models.** For most applications in the Mittelstand, this is not necessary. Modern language models (LLMs) can be used via APIs. That saves considerable cost compared to training your own.
**Management consulting.** We do not deliver strategy papers. We identify the use case during the discovery phase and then develop the solution.
**Maintenance after the project ends.** Running costs for API usage and hosting do apply, but the application is yours. No lock-in, no monthly license fees to us.
## Why Not Just Buy an Off-the-Shelf Tool?
For some tasks, off-the-shelf software is the better choice. If you need a chatbot for frequently asked questions, ready-made solutions exist.
Custom development pays off when:
- Your process is not covered by any off-the-shelf software.
- The AI has to work with internal data you do not want to hand over to external providers.
- The application has to fit into your existing systems.
- You gain a competitive advantage from a process that only your company runs the way it does.
The rule of thumb: the more specific the process, the more it makes sense to build a custom solution.
## How to Find Out Whether an AI Project Pays Off
Every project starts with three questions:
1. **Where do your employees repeatedly spend time on tasks that follow a pattern?** Those are the processes with the highest automation potential.
2. **Is there knowledge in your company that only a few employees carry in their heads?** That is a risk and an opportunity at the same time.
3. **How much does the current state cost per month?** Add up staff hours, error rates, and waiting times. That gives you the benchmark.
When the current state costs more per month than the running costs of the AI application, the investment pays for itself.
## The First Step
The first step is not a contract and not a specification document. It is a conversation.
Where is the biggest AI lever in your business? 30 minutes, no commitment. Together we identify the use case with the highest value.
[Get in touch](https://bitvaria.com/contact)
- [ki](https://bitvaria.com/tag/ki/)
- [mittelstand](https://bitvaria.com/tag/mittelstand/)
- [kosten](https://bitvaria.com/tag/kosten/)
- [ki-projekt](https://bitvaria.com/tag/ki-projekt/)
Share:
[Back to Blog](https://bitvaria.com/blog/)
## Terms explained on this page
- **ERP system**: Enterprise Resource Planning. Central company software for accounting, inventory management, production, and more. Typical providers: SAP, Microsoft Dynamics, Sage.
- **API usage**: Costs for using language models such as those from OpenAI or Anthropic via their programming interfaces. Prices depend on the amount of text processed.
- **RAG system**: Retrieval-Augmented Generation: The AI first searches your documents and then formulates an answer based on the information it finds. This way the system works with your company knowledge, not just general knowledge.
- **language models (LLMs)**: Large Language Models: Large language models such as those from OpenAI, Anthropic, or Meta. These models understand and generate text and can be used via programming interfaces without having to train them yourself.