Companies Bet on Trust to Integrate AI

Between stated ambitions and actual usage, the gap persists. Public figures and feedback highlight the stakes: trust is built on visible gains, well-defined use cases, and AI connected to the company's context. Management is called to measure utility rather than licenses.

Measurable Gains Validate AI in the Field

Several organizations report quantified benefits when they frame AI by use case. Some digital agencies that have adopted project management solutions like Wrike indicate saving between three and five hours per person each week, primarily due to a 95% reduction in time spent summarizing discussions and decisions during client calls. In the fields of legal consulting and business services, task consolidation, report automation, and improved transparency allow for a halving of the time allocated to follow-up meetings and the elimination of about 20% of time by removing manual document searches. While these savings may seem limited on an individual scale, their impact becomes significant when they involve several hundred employees and recur weekly, influencing budgets and establishing standards for future deployments. Useful indicators remain concrete: shorter meetings, accelerated validations, simplified searches, faster deliveries, and answers to key questions about usage, quality, time freed up, and reductions in delays or errors.

Connecting AI to Operational Data and Framing Access

Inadequate responses often stem from tools that ignore the organization they are meant to assist. A generic model can describe a project plan without necessarily distinguishing which projects are actually behind schedule in a given company. Without context, AI produces generic outputs. To be useful, it must rely on operational data, understand processes, and respect access rights. In France and Europe, these uses are inseparable from confidentiality, data governance, cybersecurity, and compliance. The goal is to build a structured and controlled knowledge environment where everyone only accesses authorized information, and where AI indicates the sources it mobilizes, in proportion to the stakes of the decision. The interplay between context, traceability, and permissions conditions the transition from experimental to a sustainable component of operations.

Trust is Earned Interaction by Interaction

Trust is not a slogan but a series of repeated positive experiences. An accurate response, twenty minutes saved on a task, or the anticipation of a problem gradually increases acceptance, while an incorrect, out-of-context, or unverifiable response pushes users back to old methods. Adoption depends as much on these daily interactions as on broader strategies. Teams must understand the tools, control the results, and report errors. A reliable system is not a black box: it assists while preserving the user's judgment. Training must be situated in the job, from formulating requests to verifying responses and identifying situations where human intervention remains necessary.

Progress in Usage and Reported Obstacles

Usage is progressing in France: by 2025, 18% of companies with at least ten employees reported using AI, up from 10% a year earlier, according to Insee. However, the lack of skills remains the most frequently mentioned difficulty, both by companies that do not use AI and by more than half of those that do but face challenges in expanding their usage. Internationally, Gartner predicts that more than 40% of agent-based AI initiatives are likely to be halted by the end of 2027 due to rising costs, an incomplete definition of business value, and perceived insufficient controls.

Moving from Slogans to Framed Use Cases

Despite repeated declarations of transformation through AI, actual usage struggles to take hold. Technical discussions about models and power interest teams less than concrete questions: detecting a risky project, avoiding repetitive reports, handling hundreds of requests without hiring, finding reliable information scattered around. Effective adoption begins with these pain points, using tools that facilitate, accelerate, or clarify work, understand the environment, and reduce non-value-added tasks. The lack of expertise also stems from the difficulty of aligning AI with processes, framing usage, and defining evaluation criteria. Prioritization should focus on use cases and their results, rather than on volumes of licenses, training, or content. The recommended approach: identify repetitive tasks and friction points, associate each problem with a specific AI capability, measure gains, verify actual usage, gather feedback, and then gradually expand the scope. For AI-first ambitions, it is suggested to first answer two questions: where is AI already working, and how does it concretely assist its users?