Tell us about your project
One assistant configured for each task.
The needs first, the tool second.
AIxH, a digital marketing and AI agency in Luxembourg, deploys ChatGPT or Claude in organisations that subscribe to them or are about to: needs gathered department by department, projects and skills created for each task, usage rules, team training. For Fondation Cancer Luxembourg, the assignment followed the AI training of its teams: needs collected from each department, then assistants configured in ChatGPT between December 2025 and January 2026.
Most organisations that pay for a ChatGPT or Claude subscription use it as a blank page: everyone writes their own prompts, results vary from one person to the next and the tool mainly serves those who taught themselves. Yet both publishers provide what is needed to install the work: a project brings together the instructions and reference documents for a task, a skill describes a reusable procedure, and the team finds the same assistant, set up once, at every use.
This configuration takes business time before it takes tool time. You need to know which tasks come back every week, which documents are the reference, what output format the department expects, and what must stay out of the tool. We run this assessment with each team, then build, test and present each assistant to the people who will use it.
This service is a specialisation of our AI audit & integration service. It often follows a custom AI training course, and it differs from our sovereign European chatbot: here, the tool belongs to an American publisher, and the scoping sets what may go into it.
Starting point
Two situations come to us, and they do not call for the same work at the start.
ChatGPT or Claude is in place, with uneven usage: a few people use it a lot, others have given up, and nothing is shared.
We start from what exists: the prompts each person has written, the recurring tasks, the results judged good or bad. The needs assessment turns these individual habits into shared assistants.
The teams know what the tool can do and have identified their use cases. They still need to be installed so that the usage lasts.
This is the case of Fondation Cancer Luxembourg: training in August 2025, needs gathered from the teams, assistants set up between December 2025 and January 2026.
In both cases, the first conversation is free and covers your teams, your recurring tasks and the subscription in place or under consideration.
Scope
An interview with each team: recurring tasks, the time they take, reference documents, languages, expected formats. Each need retained becomes an assistant to build, with the task it covers and the person who will validate it.
For each assistant: the information to gather from the team, the documents to add to the project (guidelines, validated examples, glossaries, templates), then the written instructions: role, steps, output format, what the assistant must ask and what it must not do.
The project or skill is created in ChatGPT or Claude, tested on the team's real cases, corrected, then presented to the people who will use it, with hands-on time. An assistant that is not presented stays unused.
What goes into the tool and what stays out (personal data, sensitive information), the human review expected before publishing or sending, then a review after a few weeks of use to adjust the instructions and add the assistants that are missing.
Four steps per assistant, from gathering the information to presenting it to the team. The time for each is estimated before the start, assistant by assistant.
Assistants
The examples below come from the assignment carried out for Fondation Cancer Luxembourg; every organisation has its own.
Writing and translation
Article writing by topic, translation into the public's languages, spelling and grammar proofreading, social media copy, accessible text and image descriptions, with the guidelines and validated examples in the project.
Replies
Replies to the emails that keep coming back to a department, prepared from validated answers and up-to-date information, then reviewed before sending. For Fondation Cancer, enquiries related to an event and to fundraising.
Analysis and monitoring
An annual report made searchable, help with finding ideas and structuring a piece of content, monitoring a topic with a research tool such as Perplexity, preparing pictograms and visual elements for validation.
Data
ChatGPT and Claude are published by OpenAI and Anthropic, two American companies: what your teams type there is processed on their servers. The business plans of both publishers exclude customer data from the training of their models by default, under their current terms, which we read through with you at the time of subscription.
The scoping therefore sets what goes into the tool: content meant for publication, internal reference documents without personal data, templates, glossaries. And what stays out: personal data of beneficiaries, donors or customers, named medical, financial or legal information. These rules are written into each assistant's instructions and repeated to the team.
When an organisation needs to query documents that cannot leave a European perimeter, the tool changes: that is the role of our sovereign European chatbot, built and hosted in Europe. The two set-ups can coexist, each with its own data.
Case study
Health · Luxembourg · 2025-2026
In-company AI training for the teams on 25 and 26 August 2025, then needs gathered from each department.
Assistants configured in the Foundation's ChatGPT subscription between December 2025 and January 2026: writing and translation, proofreading, social media, replies to enquiries, annual report, monitoring.
A twofold assignment: train, then install the uses in the tool.
Training
Our custom training courses are built on the team's own cases: the needs identified during the training become the first assistants to install.
Training is delivered in person only, in Luxembourg, France, Belgium and Switzerland.
Scoping
A first conversation is enough to scope the needs assessment. It goes faster if you come with:
Before the start, we settle together who administers the accounts, who validates each assistant and how the fees are calculated: subscriptions are still paid to the publisher, our work is quoted after the needs assessment, assistant by assistant.
FAQ
Both are configured in the same way: a project brings together instructions and reference documents for one task, a skill describes a reusable procedure. If your organisation has already subscribed to one of the two, we work in that tool. Otherwise the choice is made after the needs assessment, according to the tasks to cover, the languages, the document formats and each publisher's data processing terms.
It is processed by the publisher, OpenAI or Anthropic, on its servers. The business plans of both publishers exclude customer data from the training of their models by default, under their current terms, to be checked at the time of subscription. The scoping work sets what goes into the tool and what stays out: personal data, sensitive information, confidential documents. For content that must remain within a European perimeter, our sovereign European chatbot meets a different need.
The sovereign chatbot is a solution built and hosted by AIxH in Europe, on your documents, to answer your customers or your teams. ChatGPT or Claude integration configures a tool you already rent from an American publisher, for your teams' daily tasks: writing, translation, proofreading, replies to enquiries, analysis. The two complement each other and do not process the same data.
It depends on the number of teams and tasks. Each assistant goes through four steps: gathering the information, building the document base, writing the instructions, testing and presenting it to the team. For Fondation Cancer Luxembourg, the training took place on 25 and 26 August 2025 and the assistants were set up between December 2025 and January 2026.
No, but it changes the outcome. A trained team knows what an assistant can do and states precise needs; the needs assessment goes faster and the assistants get used. Our custom training courses are built on the team's own cases, and the integration follows on from them. An organisation already comfortable with these tools can start directly with the needs assessment.
Tool subscriptions are still paid to the publisher. Our fees are quoted after the needs assessment, assistant by assistant, according to the information to gather, the document base to build and the instructions to write. AIxH does not publish a price list. We work from Differdange, in Luxembourg, across the Greater Region and in France. Write to hello@aixh.com for a scoping adapted to your organisation.
Client reviews
★★★★★ 5.0 / 5 on Google (45 reviews) · See our Google profile →
“I had the opportunity to work with AIxH on a project to integrate AI into our marketing acquisition team. The team is competent, available and results-oriented. Exchanges were always clear and the recommendations concrete. A collaboration I recommend without hesitation.”
“A team that truly specialises in AI and in how to use it in your business, with people at the centre of their teaching and support. Recommended in this AI jungle.”
Contact
A first conversation is enough to scope the needs assessment, on your teams and your day-to-day tasks.
Talk about your ChatGPT or Claude deployment →