#43 Don't make an AI plan. Make a list of problems.

Four articles to reach a deliberately disappointing conclusion: there is no AI strategy. There is a strategy, and AI is one means among others of serving it. The organisations that will lose money over the next eighteen months are the ones that will have done the reverse, buying tools before identifying a problem.

The AI plan trap

The scenario is already visible in the field. An elected official or a director comes back from a trade show. They ask for an AI plan. A memo is produced, a budget is earmarked, a tool is bought, a licence is deployed across fifty workstations. Six months later, actual usage is concentrated among three people who were already using it, free, on a personal account.

The methodological flaw is simple: an answer was chosen before a question was framed. We saw it in other waves, CRM in the 2000s, the responsive website, the destination mobile app. The mechanism does not change, only the technology's name does.

The four questions that replace the AI plan

Ask them in this order, with your teams and not with a supplier.

• Where do we lose time re-entering data?

  Look for every task where information already written somewhere is retyped elsewhere. Count the minutes, over a week, with no software, using a notebook. It is the most profitable and the least glamorous seam.

• At what moment does the screen come between us and the person we are serving?

  That is the lesson from healthcare, transferable to any front desk, any reception, any membership relationship. A minute recovered for eye contact is worth more than a minute recovered in processing.

• What data is lying dormant here?

  Unread customer reviews, surveys never analysed, reports never reopened, visitor numbers never cross-referenced with national statistics. AI is only of interest when applied to material that already exists.

• Who owns the subject?

  Not a committee. One person, with dedicated time and a clear mandate. Without an answer to this question, the previous three are worthless.

What must be in place, with no room for debate

Two things are a matter of compliance, not ambition.

The first is transparency. Since 2 August 2026, most of the European AI Act applies, and a user must be told when they are addressing an agent rather than a human. If you have a chatbot on your site or a smart answering service, checking takes ten minutes.

The second is literacy. The obligation to brief your teams on the possibilities, the limits and the risks has been in force since February 2025 and fully enforceable since August 2026. This is not about training everyone in the technology, it is about being able to demonstrate that your staff know what they are handling. A well-built half-day covers the essentials, and it has the advantage of surfacing the unofficial usage already happening in your organisation.

Add a one-page internal policy: what may be put into an assistant, what never goes in, who reviews before publication, who signs. One page, not fifteen.

The order of priorities, as we recommend it

Three phases, over roughly six months.

First observe, over two to three weeks. Time the data-entry tasks, map the friction points, take stock of the usage already happening outside any framework. This is the only phase you cannot fully delegate.

Then frame. An internal policy, a half-day briefing, bringing whatever is already online into compliance. Low cost, high avoided risk.

Finally experiment, on a single use case, the one combining the most wasted time with the least risk. A report, a standard reply, a synthesis of customer reviews. With a measurement before and a measurement after, without which you will learn nothing and convince nobody.

So what now?

There is good news in this series, and it is slightly hidden. If the thesis holds, if we are returning to the relationship Licklider and Engelbart described sixty years ago, then the advantage does not go to the most technical organisations. It goes to those that are best at framing what they want, and that know their trade and their users best. A town of 8,000 people, an association, a tourist office start with less of a handicap than at any point in the last twenty years.

At Agence Alps, we approach this through diagnosis before tooling: digital maturity audit, task mapping, usage policy and training. Our capacity-building practice is certified for the training side, and our Strategic Diagnostic includes digital presence in its overall score. If you are preparing your 2027 roadmap or need to demonstrate compliance, write to us.