Train your teams in AI

M1.1

What a model can do, and what it cannot

You tell reliable tasks apart from those where the model makes things up.

M1.2

Write a prompt that produces a usable result

Role, material, constraint, expected format: a structure instead of endless rewording.

M1.3

Provide the context: documents, examples, expected format

You attach what makes an answer usable in your line of work.

M1.4

Check an answer: the source, never memory

You trace every statement back to the original document.

M1.5

What never goes into an online tool

You identify the data that must not leave the company.

M1.6

Choose the right tool for the task

A general-purpose model, a tool connected to your documents, or no AI at all.

M2.1

Draft, rewrite and translate a business document

You correct a clean first draft instead of writing it.

M2.2

Summarise and compare long documents

You pull the essentials out of a long document and compare two versions.

M2.3

Extract structured data from a PDF or a spreadsheet

You turn a scanned document into usable rows, then check the extraction.

M2.4

Prepare minutes, an email, a follow-up

You start from your raw notes and deliver a document ready to send.

M2.5

Write in the house tone and vocabulary

You impose your professional vocabulary and register on the tool.

M2.6

Build a library of reusable prompts

You document the prompts that work and share them with the team.

M2.7

Spot the tasks where AI wastes time

You leave out the tasks that are not worth the proofreading.

M3.1

Build a searchable internal document base

You gather, name and version your reference documents.

M3.2

Query your own documents rather than a general model

You get an answer tied to your documents, not to general knowledge.

M3.3

Make it cite the source, make it fail cleanly

An exact citation for every answer, and "I cannot find it" rather than an invention.

M3.4

Check the version of the reference document

You verify that the answer relies on the version in force.

M3.5

Spreadsheets and data: cleaning, cross-checking, consistency

You reconcile two dissimilar files and catch the discrepancies before delivery.

M3.6

Build a set of test cases

Around ten cases with known answers, replayed after every change of prompt.

M3.7

Write a procedure the team can follow without you

Operating steps, checkpoints, stop conditions: a colleague runs it without you.

M4.1

Map a process and spot what can be automated

You break a process into steps and isolate those that can run on their own.

M4.2

Connect the tools: email, storage, spreadsheet, CRM

You remove a re-keying step by passing data from one tool to the next.

M4.3

Build an end-to-end automation scenario

You assemble a complete chain, from trigger to deliverable, tested on real cases.

M4.4

Agents and delegated tasks: scope and safeguards

You write down what an agent may do and what must stop it.

M4.5

Put the human checkpoints in the right place

You place human sign-off where an error becomes irreversible.

M4.6

Log, monitor and recover in production

You record every run and restart a chain where it failed.

M4.7

Measure the real gain and decide what not to automate

You weigh the time saved against the time spent supervising, and decide.