01
A team under support pressure
Repeated customer questions or internal requests are consuming time that does not scale.
Practical AI for real business work.
I help organisations decide where AI is genuinely useful, validate the idea with real data and build the integration when the case is sound. The advice stays independent of any single model provider, and it starts with the questions that decide whether a project survives production: privacy, running cost and measurable value.
Who this is for.
02
A business processing documents
Invoices, contracts, emails or forms still require repetitive manual review.
03
A company choosing AI tools
You need an independent comparison of Copilot, ChatGPT Enterprise or a custom solution.
04
An entrepreneur with an AI idea
You want an honest feasibility check before committing to a larger build.
What is included.
Honest evaluation
We separate what current AI can reliably do from what only works in a polished demonstration.
GDPR-aware design
Data flows, storage, processors and European hosting options are considered from the start.
Working prototypes
The concept is tested on representative business data before a full implementation is approved.
Provider independence
Code, prompts and data flows remain portable where the use case allows it.
Scope and pricing.
Strategy session
from €125
A focused conversation to assess an idea, identify constraints and choose a sensible next step.
Idea validation, priorities or a second opinion on another proposal.
Audit and roadmap
from €1,250
A review of processes, realistic opportunities, providers, privacy and implementation options.
Decision-ready roadmap with budgets, risks and timing.
Implementation
from €5,000
A production AI integration connected to your documents, workflow or existing software.
Document processing, internal knowledge search, assistants or agent-supported workflows.
Prototype before making a large bet.
AI output is probabilistic, so a convincing demo is not enough. A useful test uses representative data, defines what counts as correct and measures where human review remains necessary.
A short prototype usually reveals more than weeks of strategy slides. If the result is not reliable or economical, you still receive a clear answer before a larger investment.
Privacy and provider choice are architecture decisions.
Depending on the data, a solution can use OpenAI, Anthropic, Google, Mistral or a model hosted in a controlled environment. Contracts, storage location, retention and access all matter.
Where possible, the application is designed so changing a model does not require rebuilding the complete workflow. The appropriate level of portability is agreed before development.
How we work.
01
Understand
We examine the process, data, risk and result the business actually needs.
02
Validate
A roadmap or prototype tests the most important assumptions on real examples.
03
Implement
The approved workflow is integrated in stages with measurable checks.
04
Operate
The solution is documented, monitored and reviewed as models and costs change.
Frequently asked questions.
Is your question not listed? Email me directly.
How much does an AI integration cost? +
A strategy session starts at €125, an audit and roadmap at €1,250, and a focused production integration at approximately €5,000. Larger systems can reach €25,000 or more. Scope, usage costs and assumptions are made explicit upfront.
Which AI providers do you work with? +
I work with OpenAI, Anthropic, Google and Mistral, as well as locally hosted models where appropriate. The choice follows privacy, quality, latency and cost rather than a fixed vendor preference.
What happens to our data? +
We map what data is sent, where it is processed, how long it is retained and which agreements are required. Sensitive use cases may call for European cloud regions or models hosted in a controlled environment.
Do you train new AI models? +
Usually not. Existing models combined with clear instructions and retrieval from your own documents are faster and more economical. Fine-tuning is considered only when testing shows a specific, repeated gap.
How do we know it works before investing too much? +
Implementation starts with a limited prototype or evaluation using representative data and explicit success criteria. You see the quality, limitations and expected operating cost before approving the full build.
Will we be locked into one model? +
Not necessarily. Provider boundaries can be isolated so a model can be replaced with limited work. Some advanced features are inherently provider-specific; those trade-offs are documented before development.
Relevant work