Skip to content
NIKITA, home
EN
talk to a consultant Run an audit
Open the menu

Frequently asked questions

AI at work, in short answers.

What we are asked most often, before getting started and along the way. Each answer links to the page that covers the topic in detail.

Getting started with AI.

5 questions

Where should a company start an AI project?

With a task, not a tool. Choose a piece of work that comes back every week, takes time and has a result that is easy to check: sorting requests, preparing replies, copying information from one document to another. Measure the time it takes today, try an AI use, then compare.

The NIKISCORE assessment spots these tasks role by role and ranks them by priority.

Should we start with an assessment or with training?

It depends on what you already know. If you don't yet know where AI can save you time, start with the assessment: it starts from your real tasks. If your uses are identified but the team doesn't know how to go about them, start with training. To get an idea with no commitment, the first training module is free, no account needed.

How do we build an AI roadmap?

Keep it short and easy to revise: tools change every quarter. A good roadmap lists a few priority uses, who is responsible for them, how the result is measured and when you take stock. Add a use once the previous one is in place, not before. For a larger organisation, our consultants help you scope the steps and choose between doing it in-house, buying a tool or getting support.

How do we choose an AI consultancy or training provider?

Ask for verifiable evidence: Qualiopi certification for training funded in France, a detailed programme, the names and backgrounds of the trainers, exercises on your own tasks, what you get at the end and how the result will be assessed. Be wary of figures promised before anyone has looked at your work. NIKITA is a Qualiopi-certified organisation for training actions; you can meet the team on the About page.

What return on investment can we expect from AI?

There is no figure that holds for every company. The most reliable approach is to measure, on a specific task, the time spent before and after, and the quality of the result. The NIKISCORE report estimates the time that can be freed up per role; it is then checked on the ground by testing the chosen uses.

The NIKISCORE assessment.

3 questions

How does the NIKISCORE assessment work?

Each person answers about 80 questions about their tasks, in writing or by voice, without leaving their desk. No need to prepare data or block out a day. Everyone receives an individual report: the tasks reviewed, possible uses, the estimated time that could be freed up and their AI maturity level.

How much does the assessment cost?

The assessment is included in the NIKISCORE subscription: it is no longer sold separately. The terms are set out on the NIKISCORE page.

What should we do once we have the report?

Pick one or two uses, not ten. The team can learn to put them in place through training, or hand them to us for a custom implementation with our consultants. Either way, the gain is checked on your real tasks before moving on to the next.

Training.

6 questions

Which AI training should you choose for your employees?

Three formats, depending on the time available and the goal:

  • Augmented Employee: 2 days live with a trainer, on the team's tasks;
  • the online course: 7 modules and 22 interactive lessons, at your own pace;
  • AI Lead: a 20-day certified programme to champion AI use across the organisation.

For regulated professions, profession-specific programmes take into account the code of conduct and funding specific to each profession.

Do you need technical skills to take a course?

No. The courses start from each person's current level and their job. You learn to phrase a request, check an answer and reuse a method, not to code.

How long does it take, online or live?

Two days are enough for a team to leave with uses in place on its tasks; the online course is taken at your own pace. Live suits teams that want to work together on their files, online suits scattered schedules. You can start with a small group, then widen: the first online module is free so you can judge for yourself.

Do we work on real cases?

Yes. The exercises are based on the participants' tasks and the tools they already use. In several programmes, everyone builds their first agents during the course: accountants, for example, leave with five practice agents on day two.

Can AI training be funded?

In France, often yes. NIKITA is a Qualiopi-certified organisation for training actions, a requirement for funding by an OPCO or by the FIF PL for independent professions. The request is made to your funder before the session starts, with the quote and the programme; the amount covered depends on your situation and their approval. Each profession page lists the possible funding.

How do we measure the effect of a course?

Before the course, choose two or three reference tasks and note how long they take. Measure again a few weeks later, then every quarter: what counts is the use that stays in daily work, not satisfaction on the day.

AI agents.

5 questions

What is an AI agent, and how is it different from a chatbot?

A chatbot answers questions. An agent acts: it reads your emails, prepares a reply, updates a file or schedules a meeting, chaining several steps in your tools. A good agent lets you approve before any action that commits you. To go further: AI agent or automation? (in French)

What can you hand over to an AI agent?

Repetitive tasks that follow clear logic: sorting emails and preparing replies, chasing quotes or invoices, writing meeting notes, preparing a report, organising a calendar. NIKIBOT does this work in your tools; nothing goes out without your approval.

Can an agent work in our tools (email, CRM)?

Yes, as long as it is connected to them. NIKIBOT connects to Gmail, Calendar, Drive, Notion, Sheets and any MCP server. For a CRM, an ERP or business software, possible integrations depend on the tool: our consultants study them during scoping.

Do you need a technical team to put an agent into service?

No. NIKIBOT starts without a developer: you just connect your accounts. For a custom agent, we design and test the solution; your team approves it on real situations. The timeline depends on the scope, defined together before starting.

How do you control an agent and make it evolve?

Keep three habits: approve actions that commit you (a message sent, an order), keep the list of what the agent has done, and regularly measure the time saved and the errors corrected. An agent is adjusted like a process: its instructions are reworked when the work changes. See also: mistakes to avoid when deploying an agent (in French).

Data, GDPR and sovereignty.

3 questions

How do you protect data entrusted to an AI?

Before using a tool, check where the data is processed, whether it is used to train models and who has access to it. Don't put personal or confidential data into it without a contractual framework. Our guide your data in ChatGPT at work (in French) details the useful settings.

What do the GDPR and the AI Act require?

The GDPR applies as soon as an AI processes personal data: legal basis, informing the people concerned, a contract with the provider. Since 2 February 2025, Article 4 of the AI Act also requires companies that use AI to ensure a sufficient level of AI literacy among their staff, which training helps document. See the AI Act for companies and AI and GDPR (in French).

Can we work with a European AI?

Yes. European models such as Mistral's cover most office uses; our course for French civil servants is built entirely around them. NIKIBOT is hosted by Scaleway in Europe, and a dedicated installation on your servers is available on quote.

A few concepts.

3 questions

What is prompt engineering?

It is the art of phrasing a request to an AI: giving the context, the goal, the expected format and the limits to respect. A precise instruction gives a more useful answer that is easier to check; it is the first skill worked on in training.

How does RAG work?

RAG (retrieval-augmented generation) first has the AI search your documents, then has it write its answer from the passages it found. The answer can then cite its sources, and your documents don't need to be used to train a model.

What is the difference between ChatGPT and Perplexity?

Perplexity is first and foremost an answer engine: it searches the web and cites its sources. ChatGPT is a more general assistant, for writing, analysing or summarising, which can also search. Our comparison of assistants (in French) helps you choose by use.

Getting cited by AI.

4 questions

What is visibility in AI assistants (GEO, LLMO, AEO)?

These acronyms describe the same idea: getting an assistant such as ChatGPT, Perplexity, Gemini or Google's AI Overviews to cite your site when it answers. Classic SEO remains the foundation, because these assistants read the pages search engines find. What changes: some searches stop at the answer, with no click, hence the value of being the cited source.

How do you get cited by ChatGPT, Claude or Perplexity?

No one can guarantee it, but some practices help: answer your clients' questions clearly at the top of the page, date and sign your content, keep the same information on your site, your listings and directories, and let search and assistant crawlers access your pages. Useful, verifiable content remains the best argument.

Should we add structured data (JSON-LD)?

It helps: schema.org markup (organisation, service, FAQ, article) describes unambiguously who you are and what you offer, to search engines and assistants alike. It does not replace the visible content, which must say the same thing.

How do you measure visibility in AI assistants, and how long does it take?

Regularly ask several assistants the same questions, the ones your clients ask, and note whether and how your site is cited. The timing can't be predicted: it depends on when pages are re-read and on competition on the topic. Think in months, not days.

Answers updated on .

Your question isn't here?

Write to us or book a conversation with a consultant.

Contact us