AI integration · ChatGPT · Claude

AI that works within your processes.

I am thoroughly familiar with ChatGPT, Claude, and other AI models. I bring AI to where it noticeably reduces your workload: into your apps, flows, and documents, with clear data protection rules and with a person who makes the final decision.

  • ChatGPT and Claude - tools I know thoroughly
  • Flows AI as a building block in the process
  • GDPR data protection clarified in advance

01Use cases

Where AI helps you

Orientation and use cases

Where does AI really help, and where does it not? I examine your processes for tasks that language models do well - understanding, summarizing, and classifying text, as well as extracting data - and assess benefits, costs, and risks.

AI in Power Automate

Integrate language models such as ChatGPT or Claude into flows through an API: with fixed instructions, a structured response, and a check of the result before it is processed further.

Analyzing documents and emails

Extract data from inquiries, invoices, delivery notes, and forms, summarize the content, and assign it to the right case - with AI Builder or a language model, whichever fits better.

AI in Power Apps

Suggest text, summarize long entries, or populate structured fields from free text - directly in the app, as a suggestion with a preview instead of silent automation.

Maintaining instructions properly

Instructions, examples, and the expected response format are stored centrally and versioned. This makes it possible to trace how the AI works and make targeted improvements.

Putting Microsoft Copilot in perspective

My focus is on ChatGPT and Claude. I give you an honest assessment of Microsoft 365 Copilot and Copilot Studio: What do they deliver for your case, what do they cost, and when is a custom flow the better choice?

02What I focus on

Quality you notice in everyday work.

Responsibility

The AI suggests. A human decides.

Language models are strong at understanding and organizing text - but they are occasionally wrong and still sound convincing. That is why I build solutions so that decisions with consequences are confirmed by a person: approvals, payments, and external replies.

Instructions and responses are logged, uncertain results go on a review list, and the process keeps working even if the AI service is temporarily unavailable.

  • Approval before sending
  • Log of instruction and response
  • Fallback without AI
  • Spot checks in operation

Privacy

Your data stays under control.

Which data goes to which model, in which region, and under which contract? I clarify this with you in advance. I rely on business accounts instead of private accounts and leave out personal data where it is not needed for the task.

The legal assessment remains with those responsible for data protection in your organization - I provide the technical facts and implement the requirements.

  • Business accounts instead of private accounts
  • Data minimization
  • Data processing agreement
  • Costs with a cap

Test first, then roll out

A pilot with real examples.

Before AI is used in a production workflow, I test it with real, anonymized examples from your everyday work. This way, you see in advance how often the results meet expectations, where rework is needed, and what a transaction costs.

The solution goes into production only when quality, effort, and data protection requirements are met - and it continues to be monitored there.

  • Real, anonymized examples
  • Measuring hit rate and rework
  • Cost per transaction
  • Decision on a solid basis

03Approach

From the first conversation to a solution in production.

Consulting and implementation from a single source: whoever writes the concept also builds it - with no friction between consultant and developer.

  1. Initial consultation

    Free and without obligation via Microsoft Teams: What is the challenge, what is the goal, and which licenses do you already have? Afterward, you will know whether and how the Power Platform fits your needs.

  2. Process analysis

    Together with the people who run the process every day: How does it work today, and where do manual handoffs between systems or media, delays, and errors occur?

  3. Solution concept

    Data model, screens, flows, permissions, and licensing - in writing, easy to understand, and with a clear effort estimate.

  4. Implementation in stages

    Short iterations, each with a working interim result. You see early on what is being built and can adjust course.

  5. Rollout and handover

    Acceptance, deployment to production, documentation, and handover to your team - so the solution does not depend on me.

  6. Further development

    If you wish, I will continue to support the solution afterward: enhancements, new requirements, and updates from Microsoft.

04Toolbox

Technologies in this area

  • ChatGPT
  • Claude
  • OpenAI API
  • Anthropic API
  • Azure OpenAI
  • AI Builder
  • Power Automate
  • Power Apps
  • HTTP and REST
  • JSON schema
  • Copilot Studio
  • Microsoft 365 Copilot

05Real-world projects

Projects in this area. No names, real substance.

As a rule, I do not name clients - their industry, size, and project requirements say enough.

11/2021 - 12/2021Implementation

Shading systems

more than 300 employees · international sales

Rollout of Microsoft Power Automate as the central automation platform and automation of invoice processing with Microsoft Power Automate (cloud and desktop flows), Microsoft AI Builder, and Microsoft Dynamics 365 Business Central. In addition, real-time visualization of presence status via the Microsoft Graph API in SharePoint and Microsoft Power BI, as well as training on Dataverse and canvas apps.

  • Power Automate (cloud flows, desktop flows)
  • AI Builder
  • Dynamics 365 Business Central
  • Graph API
  • Power BI (streaming dataset, REST API)
  • SharePoint Online
  • Dataverse

View all case studies

06Questions and answers

Frequently asked questions

Which AI is the right one: ChatGPT, Claude, or Copilot?
That depends on the task, the data, and your licensing situation. ChatGPT and Claude are very strong at texts, analyses, and structured answers, whereas Microsoft 365 Copilot is tightly integrated into Microsoft 365. I test the candidates with real examples from your process before we decide.
May we pass customer data to an AI?
Not automatically. The contract, storage location, and purpose matter. Together, we determine what data is necessary, which service and contract are suitable, and how to avoid using personal data. The legal assessment is handled by whoever is responsible for data protection in your organization.
What does AI cost in operation?
Language models are usually billed by the amount of text processed. For typical tasks such as summarizing an email, this is often only a fraction of a euro per transaction. I calculate costs in advance based on your volumes and build in caps so that costs remain predictable.
Can AI take over our processes completely?
Rarely, and that is a good thing. AI takes a lot of reading, sorting, and writing work off your hands. Decisions with consequences - approvals, payments, and replies to customers - should be confirmed by a person. That is how I build the solutions.

Your process. My solution.

Tell me where things get stuck today - in a free, no-obligation initial consultation via Microsoft Teams, we will clarify what the Power Platform can do for your process and what a sensible first step looks like.