Official program

Program overview

What you will be able to do

Judge what a model can actually do, and above all recognise a plausible but wrong answer before passing it on.
Decide what may be submitted to an AI tool, and under what conditions, given the GDPR and the sensitivity of the documents.
Produce professional writing: letters, notes, meeting minutes, case summaries.
Work across a document set and draw a sourced summary from it, verifiable item by item.
Automate a recurring task in your department, end to end, and measure the real gain.
Design a simple agent on a well-defined process, with explicit human checkpoints.
Run an AI project and support a team whose comfort levels are uneven.

Practical information

Audience: Anyone in an administrative role: secretarial and assistant work, administration, HR, accounting and finance, front desk, procurement, legal, enrolment and student records, back office. In companies, associations, professional practices, public bodies and local authorities alike. Groups of 4 to 12 trainees.
Prerequisites: Everyday use of an office workstation and email. Access to a generative AI tool in a professional environment. No programming skills are required at any point in the program.
Duration & format: 42 h across 3 levels (7 h + 21 h + 14 h), on site or remote. The levels run in order, may be spaced out over time, and each one can be taken on its own.
Tooling environment: The program adapts to what you already run: an office suite with a built-in assistant, an online enterprise AI service, or a self-hosted solution on your own infrastructure where data sensitivity requires it. Exercises are replayed in your environment, not in a demo sandbox.
Pricing: In-house: 13,500 € excl. VAT for the full 42-hour program (up to 12 trainees), i.e. 1,125 € excl. VAT per trainee for a group of 12. Single level: 2,250 € excl. VAT per day. Open enrollment: 900 € excl. VAT / trainee / day.

Course outline

3 levels · 14 modules

Level 1 7 h

Foundations and usage framework

Understand the tool, know what you are allowed to do with it.

1.1 - 2 h

Understanding generative AI

What a language model is, what it can do and what it cannot. Prompt, context, context window. Hallucinations, bias, dated knowledge. Guided demonstrations on real administrative situations.

Deliverable. A sheet listing five cases where the tool got it wrong in session, with the nature of the error and the signal that should have raised a flag.

1.2 - 2 h

Data, compliance and usage framework

Personal data and the GDPR, professional secrecy, sensitive documents. What may be submitted to a tool and what must never be. Information and transparency duties under the European AI Act. Building an "allowed, supervised, forbidden" matrix specific to your organisation.

Deliverable. The department's « allowed, supervised, forbidden » matrix, filled in against a dozen real documents.

1.3 - 2 h

Mapping your use cases

Inventory of the department's repetitive tasks. Effort and value matrix. Estimating the time that can genuinely be recovered. Selecting the cases to be worked on at level 2.

Deliverable. A map of the role: tasks inventoried, annual time estimated, and three cases carried forward to level 2.

1.4 - 1 h

The structured prompt

Role, context, task, expected format, constraints. Guided examples. First attempts on your own documents.

Deliverable. Three requests rewritten as role, context, task, expected format and constraints, tested and kept.

Level 2 21 h

Hands-on practice on your own case files

Produce, verify, automate.

2.1 - 3 h 30

Getting to grips with the assistant

Interface, history, workspaces, sharing with colleagues. Handling attachments and reference documents. Privacy settings and what happens to conversations.

Deliverable. The trainee's workspace, set up with their reference documents and naming conventions.

2.2 - 3 h 30

Advanced prompting and reliability

Breaking down a complex request, guiding by example, iterating. Getting sources and justifications out of the model. A two-stage verification protocol. A dedicated workshop on spotting answers that are wrong but credible.

Deliverable. A two-stage verification protocol, applied to a real piece of work and documented on one page.

2.3 - 3 h 30

Writing and document production

Letters, notes, meeting minutes, case summaries. Adapting register and tone to the recipient. Building reusable templates. Assisted proofreading and harmonising a mixed set of documents.

Deliverable. A letter and a set of minutes produced from your own material, plus a template the department can reuse.

2.4 - 3 h 30

Working across a document set

Querying a body of documents, extracting a precise piece of information, comparing two versions, producing a sourced summary. Limits and precautions with long documents.

Deliverable. A sourced summary drawn from a real document set, every claim traceable to its origin.

2.5 - 3 h 30

Spreadsheets and management data

Generating and explaining formulas. Cleaning and formatting lists. Tracking sheets and simple indicators. Sanity checks on the results produced.

Deliverable. A cleaned tracking sheet, with its formulas explained and its consistency checks in place.

2.6 - 3 h 30

Automating a task end to end

Each participant picks a recurring task from their own role. Building the full processing chain, from input file to deliverable. Testing, measuring the gain, and writing documentation the rest of the team can use.

Deliverable. A complete, documented automation, with the time taken measured before and after.

Level 3 14 h

Agents and project management

Moving from individual use to a departmental project.

3.1 - 3 h 30

Principles of agentic AI

What sets an agent apart from an assistant. Tools, memory and the decision loop. What agentic AI makes possible today, and what it does not yet. Documented case studies, successes and failures alike.

Deliverable. A qualification grid applied to three processes, each ending in a reasoned decision.

3.2 - 3 h 30

Designing an agent for an administrative process

Framing a candidate process. Breaking it into verifiable steps. Placing the human checkpoints. Building and testing a simple agent on a real case from the department.

Deliverable. A working agent on a real process, with verifiable steps and human checkpoints.

3.3 - 3 h 30

Running an AI project

From use case to proof of concept, then to deployment. Usage and quality indicators. Costs, vendor lock-in, reversibility. Defining the stop criteria for an experiment before you start it.

Deliverable. A project framing sheet: indicators, costs, reversibility and stop criteria.

3.4 - 3 h 30

Supporting the change

Bringing along a team with uneven comfort levels. Addressing concerns about jobs and about the quality of the work. Internal champions, community of practice, capturing what has been learned. Measuring adoption over time.

Deliverable. The department's adoption plan: sequencing, champions, answers to objections, adoption indicators.

Register for this program

For the full statutory information (terms, accessibility, assessment), see the catalog.

Accessibility: this programme is open to people with disabilities. Our disability officer, Angélique Bensalah (a.bensalah@way-up.io), reviews the necessary adjustments in a preliminary interview, and refers you to a competent partner if we cannot meet your needs. Complaints: same contact, acknowledgement within 5 working days and a reply within 15 working days (procedure).

Programme v1.0 · updated 20 August 2026

Generative AI for Administrative Roles