Google and Stanford: 5 Keys to Effectively Integrate AI at Work

A Revealing Study on AI Adoption

While 95% of generative AI projects fail in businesses, according to MIT, Google sought to understand why some of its employees succeed in leveraging AI while others end up abandoning it. In collaboration with researchers from Stanford University, the company observed its teams' practices regarding AI tools for 18 months. The results of this study, published in the Harvard Business Review, show that mastering prompts is not enough. The most effective users adopt a product management mindset to integrate AI into their daily routines.

Five Strategies to Go Beyond Simple Substitution

Stanford researchers observed that the majority of users remained stuck in what they call "simple substitution," meaning replacing an existing task with AI without rethinking the method. In this mindset, the effort required to achieve a satisfactory result often outweighs the benefits gained, and the tool is gradually abandoned. In contrast, users who derive real value from AI often, albeit unconsciously, apply five distinct strategies.

Start with the Problem, Not the Technology

The first instinct of effective adopters is: "Do not start with technology, but with the work," writes Google. This involves, in practice, identifying bottlenecks in one's professional daily life, time-consuming tasks, or moments when there is not enough time to delve into an analysis or be creative. It is this mapping of irritants that reveals where AI can truly make a difference, rather than trying to impose a tool on a process that is already functioning.

Choose the Right Tool, Beyond the Chatbot

Once the need is identified, the study recommends exploring the diversity of available AI tools. Google compares generative AI to "a Swiss Army knife: a versatile technology with dozens of functions." Limiting oneself to a conversational chatbot is akin to using only one blade. The product management logic helps precisely evaluate which tool sustainably meets the identified need, even if it means changing work habits.

Start Small, Iterate Quickly

Researchers emphasize the importance of not trying to overhaul an entire workflow at once. The most effective approach is to prototype on a limited use case, test quickly, adjust, and then gradually expand. This iterative process allows for validating the real interest of the tool before investing in a broader deployment and avoids the pitfalls of overly ambitious projects from the outset.

Think Systemically

The most useful AI is not the one that automates an isolated task, but the one that fits into a complete process. According to the study, "the most significant gains often come from cross-referencing datasets, chaining multiple manual tasks in an AI workflow, or the ability to feed strategic thinking by aggregating multiple areas of expertise." This cross-sectional vision distinguishes advanced users from those who remain confined to sporadic uses.

Document and Share Practices

The last strategy identified by researchers is to transform successes into reproducible models. This involves documenting workflows that work, formatting them into reusable templates, and sharing them within the team. The goal is to prevent each employee from starting from scratch and to create a collective capitalization effect on productivity gains.

Recommendations That Echo Observed Difficulties in France

The findings of the Google-Stanford study resonate with available data on AI adoption in France. The Ipsos report conducted for Google reveals a persistent gap between executives and employees. Indeed, 70% of executives observe productivity gains thanks to AI, while employees struggle to identify relevant uses for their own work. Only 21% of them have also benefited from dedicated professional training.

The phenomenon of Shadow AI illustrates the trap of "simple substitution" described by Stanford researchers. 42% of employees use AI at work with their personal accounts, without a structured framework or method. The CRÉDOC barometer further confirms that 64% of generative AI users adopted these tools on their own initiative, without support from their employer. The five strategies proposed by Google can therefore provide a methodological framework to transform these spontaneous uses into practices that are genuinely integrated into work processes.