Your questions about AI, our straight answers.
Forty questions business leaders ask us. A short answer to go fast, a full page to go further.
01 · 5 questions
Where to start?
- We already have ChatGPT or Copilot licences. Is that an AI strategy?No, it is equipment. A strategy starts from what the company has to achieve and from the places where AI really changes how it is done. Licences can be part of it; they cannot take its place.
- Should we appoint a head of AI?Above all, a member of the leadership team needs to own it. An isolated head of AI, with no hold over the business, soon becomes just another department. Decisions about AI must be taken where decisions about the company are taken.
- Where do you start when you do not know where to start?With a map, not a tool. Understand how value flows in your company, which departments depend on which, where the bottlenecks are. The right first uses almost always appear once that map is drawn.
- Are we behind?Less than the announcements suggest. Many companies have multiplied trials without achieving any real transformation. Arriving later with a clear vision is often better than having been first with twenty orphaned pilots.
- Should everyone be trained before launching anything?Training everyone on a generic tool rarely achieves much. Training pays off when it starts from a concrete use in each person’s job. People learn best by doing.
02 · 5 questions
Where is the value?
- How do we choose our first use cases?By combining three criteria: the impact on a flow that really matters, feasibility with the data you have today, and the effects on neighbouring departments. The most visible case is not always the one that pays the most.
- How do you measure the return on investment of an AI project?By measuring the whole flow, not just the automated task. A department that saves time may cost another one time. Indicators are set before launch, on the business side, and tracked after going live.
- Is AI mainly about cutting costs?It is the easiest argument to defend internally, rarely the most interesting. Lasting gains often come from what the company becomes able to do: a new service, a faster decision, more consistent quality.
- Is one big project better than several small ones?Neither, as a matter of principle. Small projects with no common thread add up badly, and a big monolithic project takes too long to prove anything. We prefer successive steps that serve a single direction.
- Can you transform a line of work, not just equip it?Yes, and that is often where the real value lies. Equipping means doing the same thing faster. Transforming means asking whether the step still needs to exist, or whether the service could be delivered differently.
03 · 5 questions
Teams and the business
- How do we stop our teams rejecting AI?By involving them from the design stage, not at roll-out. Rejection rarely comes from the technology. It comes from a tool designed without the people who will use it, or one that unbalances their work without anyone seeing it coming.
- Will AI cut jobs in our company?It will shift tasks, that much is certain. What it does to jobs depends on your choices, not on the technology. Telling teams clearly, and early, avoids a great deal of resistance.
- Who should lead the project: IT or the business?Both, but the business is in the driving seat. IT makes sure the whole thing holds together and stays secure. A project led by technology alone often answers a question nobody was asking.
- How do we keep human expertise when AI does the work?By consciously deciding what people continue to do, check and understand. An expert who no longer practises ends up unable to check. AI’s level of autonomy is also set to preserve that know-how.
- Are our managers ready?Their role changes more than their teams’. They arbitrate between what AI suggests and what their experience tells them, and have to explain those decisions. Supporting them is often the most neglected part of a project.
04 · 5 questions
Data and technology
- Is our data ready for AI?Rarely all of it, and that is not an absolute prerequisite. What matters is the quality of the data the intended use depends on. Start there, rather than with an endless clean-up.
- Should we host our own models or use a major provider?It depends on how sensitive your data is, on volumes and on the dependence you are willing to accept. There is no general right answer. The mistake would be to choose before asking those three questions.
- Should we train our own model?Almost never at the start. Existing models, used well and connected to your own knowledge, cover most needs. Training a model is justified for very specific cases, and is decided with full knowledge of the facts.
- Is our data safe with generative AI?Yes, if the architecture and the contracts were chosen for that. The real risk often comes from unsupervised use, such as a confidential document pasted into a consumer tool. Supervising works better than banning.
- Do we need to replace our existing systems?No. AI should build on what works. Replacing a system that does its job in order to ‘do AI’ is one of the surest ways to waste time and lose the teams’ trust.
05 · 5 questions
Risks and dependence
- How do we avoid becoming dependent on an AI supplier?By keeping control of what makes your value: your data, your processes, your decision criteria. An architecture that lets you change models without rebuilding everything costs a little more at the start, and much less later.
- What do we do when AI gets it wrong?Plan for it to get things wrong. Each use must have a check in proportion to what an error costs: a suggestion reviewed, an action approved, an alert when the AI is unsure. It is a design decision, not an option.
- Can automating one department unbalance another?Yes, and it is the most common trap. One department speeds up, the next receives more than it can absorb, or loses information it used to receive without knowing it. It is to see these effects in advance that we designed the AI Flow Map.
- How far should AI be allowed to decide on its own?Action by action, according to what an error costs and whether it can be put right. Between ‘AI suggests’ and ‘AI acts alone’ there are several settings. Choosing the right one is often the real strategic decision.
- How do you know whether an AI use is drifting over time?By monitoring it like a process: indicators, samples reviewed, alerts when behaviour changes. An AI use is never finished on the day it goes live.
06 · 5 questions
The legal framework
- What does the EU AI Act change for our company?It classifies uses according to their level of risk and imposes proportionate obligations: transparency, documentation, human oversight for sensitive uses. The first step is to list your uses, including those nobody has declared.
- Who is responsible when an AI makes a mistake?In most cases, it is first and foremost the company using it that has to answer for it. All the more reason to know who decides what, and to be able to show it.
- Is generative AI compatible with the GDPR?Yes, provided you know which personal data is flowing, where it is processed and why. The difficulties mostly come from improvised uses. A clear framework, set from the start, settles most of them.
- Should we tell our customers they are talking to an AI?Yes, and the regulations point that way. Beyond the obligation, it is a matter of trust: a customer who finds out afterwards that they were talking to a machine rarely takes it well.
- Do we own the content produced by AI?The question is still evolving across countries and uses. Check your suppliers’ terms and keep a record of the human share of the work. For sensitive cases, a lawyer’s opinion remains essential.
07 · 5 questions
Making it last
- Why do our pilots never scale up?Because they were often designed to succeed in isolation. A pilot proves that a task can be automated. It says nothing about what happens when it is automated across the whole company.
- Who should govern AI in the company?A light body that brings together leadership, business, technology and legal, and that makes decisions. Not one more committee that approves everything, but the place where priorities, autonomy levels and stops are decided.
- Should you know how to stop an AI project?Yes, and it is a rare skill. A project that does not deliver must be able to stop without drama. Setting the criteria for that decision in advance makes it much easier to take.
- How do you keep up with a technology that changes every month?By separating what changes fast, the models and tools, from what changes slowly: your processes, your data, your criteria. A well-designed architecture absorbs new developments without rebuilding everything.
- Once in place, does AI still need work?Yes. Data changes, uses change, models change. You have to measure, adjust, sometimes start again. It is the last step of our method, and it leads back to the first.
08 · 5 questions
True or false?
- ‘We need to clean all our data first.’False. You first need to know which data the intended use depends on. A big clean-up beforehand delays everything and rarely serves the use that will matter.
- ‘AI is an IT project.’False. It is a company project that needs IT. When technology alone owns it, the transformation often stops at the tool.
- ‘The bigger the model, the better the result.’Not always. A smaller model, well chosen and well connected to your data, often does better, faster and at lower cost. The choice depends on the use, not on this month’s rankings.
- ‘A successful pilot proves it will work everywhere.’False. It proves it works within its scope. At scale, other departments, other data and other constraints come into play.
- ‘A human in the loop is enough to stay in control.’Not necessarily. A person approving five hundred proposals a day is no longer controlling much. Control has to be designed: who checks what, at what point, with what information.