AI generativa nelle aziende: come aumentare produttività e qualità del lavoro con una strategia centrata sulle persone.

Generative AI in companies: how to increase productivity and quality of work with a people-centered strategy.

Adopting new technology not only improves efficiency by saving time, but increases italso the quality of work, because it allows you to dedicate yourself to strategic and creative activities.But it is necessary to provide targeted training and adapt it to the different company functions.

As the evolution of Intelligence progressesArtificial Generative (AI Gen), the next challenge for leaders is clear: make itscalableand can provide ameasurable valueacross their organizations (Davenport and Bean, 2025).

As companies move from experimentation to enterprise-wide adoption, many struggle not with the tools themselves, but with the organizational transformation needed to meaningfully integrate them into people's daily work. Tools will continue to evolve: it isthe human side of the equationwhich determines the true success of AI Gen initiatives.

We studied one of the largest real-world AI Gen implementations to date, at the multinational pharmaceutical companyNovo Nordisk. His experience shows that success does not just depend on infrastructure, but also on the way people think, yesadaptethey collaboratewith AI. A key lesson: While the adoption of AI Gen and broader digital transformations have common roots, the former is particularly disruptive, redefining the very nature of work in unprecedented ways.

Like many organizations, Novo Nordisk started with a familiar expectation: that AI Gen would primarily increase productivity (Brynjolfsson, Li, & Raymond, 2025). Guided by the leadership principle “time is the ultimate currency” and a campaign called “Make your time count,” the company launched an enterprise-wide implementation of the toolCopilot AI Gen from Microsoftat the beginning of 2024, with the aim ofsave timeeimprove efficiency. And, in many ways, the company has achieved this goal.

Each employee saved an average of 2.17 hours per week after starting to use the tool. But something unexpected also happened: Those hours weren't what employees valued most. Employee satisfaction with Copilot was three times more related to perceived improvement inquality of workthat in the time saved. Employees reported improvements in the quality of content synthesis, content creation and ideation. Interestingly, many employees reinvested the time saved into interactions with people, strategic planning and creative work. As one of them said: “I can dedicate more time and energy to strategy and planning the launch of my project.”

This insight calls into question a fundamental assumption of many AI Gen implementations: that the core value of the technology lies in pure efficiency. In practice, the promise of AI Gen is broader and more human-centric.

Novo Nordisk's implementation experience, which went from a few hundred Copilot users in January 2024 to 20,000 in February 2025, offers important lessons to leaders who are tackling the challenge of scaling AI Gen. Through surveys conducted onover 3 thousand employees, internal analysis, and front-line interviews, we uncovered both employee dynamics and field-tested leadership approaches to drive meaningful adoption at scale.

Scaling AI Gen: It's not plug-and-play

Scaling AI Gen is not just a technical challenge, but a change management marathon. The real work issupportemployees as they experiment, struggle, and ultimately find their groove with these new tools.

At Novo Nordisk, the adoption of Copilot has not been linear, but has happenedthree phases. First there was apeak: after about a month, the23%of people were frequent users and the74%by moderate users. Then there was adeclineof interest, with the15%of the early adopter group that went inactive after three or four months and the average time saved dropped from2,29a2,14hours per week. One user said: "I still haven't quite figured out how to use it. I tried it a bit at first, with little success, and haven't used it since." A similar pattern occurred in app usage: Productivity and quality gains declined after users expanded their use of Copilot from one or two apps (such as Word and Excel) to four or five, but then rebounded with six or more apps.

This mid-cycle decline is typical of AI Gen adoption. People's initial enthusiasm gives way to frustration as the immediate benefits dry up and integration difficulties increase. If left unchecked, this decline can turn into abandonment. Our research has revealed that a few disappointments are enough to destroy people's enthusiasm for technology; in other words, a user might give up on using Copilot after a few failed attempts. However, employees who persist beyond this decline often report substantial performance improvements, likely due to the effects of accumulated learning. This is a crucial time for targeted training interventions.

To combat the mid-cycle decline, Novo Nordisk has implemented a number of enablement strategies, including: targeted training interventions timed around key adoption stages; reallocation of licenses and waiting lists to reinvigorate interest; a network of AI Gen advocates to provide contextual guidance and support adoption momentum; targeted microcommunications, such as tip-focused newsletters, to address user-reported challenges; ongoing feedback, such as periodic surveys, usage dashboards and competitor benchmark monitoring, to evolve support as user needs change.

As the company's experience highlights,AI Gen requires ongoing training and support. The effectiveness of AI Gen depends on training people, not just AI models. Leaders should foster an ecosystem where employees feel supported, informed, and inspired to push the capabilities of the technology further (Puranam, 2025).

Tailor AI Gen enablement based on business function

However, different business functions require different types of support. Recognizing how people, with different roles and mindsets, interact with AI Gen is critical to driving meaningful adoption.

At Novo Nordisk, Copilot's impact varied significantly across features, resulting in a shift from uniform implementation to targeted enablement. The analysis comparing time savings and quality improvements in the corporate, commercial, production and research areas revealed strong disparities.

The corporate and commercial teamshave been among the most efficient in terms of improving productivity and quality of work with Copilot. On the other hand, departments such asResearch,DataeAIeClinical development, while continuing to benefit, saw smaller gains in both time savings and quality improvements. The employeesStem, accustomed to deterministic systems with constant outcomes, have often struggled with the probabilistic nature of AI Gen. Because it operates on models that produce variable responses, its results can conflict with precision-based workflows that require reliability. Thehallucinationsof AI(incorrect or meaningless results) have further complicated the integration of the tool into research-oriented activities, where accuracy and reliability are of paramount importance. These hallucinations have manifested themselves in a variety of ways, including fabrications, factual inaccuracies, errors in logic or reasoning, mathematical errors, and choices based on irrelevant models (Sunet al., 2024).

This variability has undermined people's trust in the system and created significant barriers to its adoption. As one researcher put it: “I don't see how to apply it to my line of work.” In contrast, one sales user described it as “a game-changing change for me in many aspects of my job.” Adapting to AI Gen would require a change in mindset among STEM employees: they would have to learn to manage the inherent unpredictability of such tools.

To facilitate the process, Novo Nordisk has moved from a uniform implementation to apersonalized. He launched a specific onboarding program for each function he createdplaybookof use cases and learning libraries aligned with job roles and worked with Microsoft to customize Copilot features for different teams. This flexibility ensured that employees in both creative and precision roles could find meaningful applications that aligned with their workflows.

Surprise champions: the most experienced employees

Contrary to some myths about digital natives, data has shown that isenior employeesof Novo Nordisk tended to use AI Gen more effectively than theirsyounger colleagues. The survey results indicated that experienced workers outperformed their younger colleagues in both increased productivity and improved work quality. Why? These older employees' deep understanding of workflows allowed them to quickly identify where tools like Copilot could add value. Furthermore, they were better equipped to evaluate the results generated by the AI ​​and integrate them into complex tasks with greaterprecision.

Younger employees, in contrast, often lacked the context to identify high-impact opportunities. One young employee said, “I don't know enough real-world use cases; what can I use it for?” Another noted: “I haven't had a specific use case where I could clearly see the benefits of using/exploring Copilot.”

The lesson: Rather than assume that younger workers will lead the AI ​​Gen charge, organizations shouldmake responsibleexperienced employees to act asamplifiers.

After this insight overturned Novo Nordisk executives' assumptions, a cross-functional network of primarily experienced staff members was created to conduct peer-to-peer demonstration sessions, provide role-specific training, and share practical examples tailored to specific work contexts. Meanwhile, internal corporate social media communities like Viva Engage (a social collaboration tool built into Microsoft Teams) have enabled knowledge sharing between senior and junior employees, driving adoption.

By investing in senior employees as adoption enablers, leaders can tap into these individuals' contextual expertise to drive meaningful use of AI Gen, tailoring training to workflows and seniority, and ensuring junior staff members receive clear, accessible use cases to build their confidence. By aligning enablement with experience, organizations can unlock the potential of AI Gen at every level.

AI Gen performance depends not on technological expertise, but on contextual fluidity, trust, and the human ability to integrate new tools into nuanced workflows.

Overcoming cultural resistance and AI shaming

Not everyone at Novo Nordisk welcomed Copilot. Thereresistenceculturaland the so-calledAIshamingposed significant obstacles, with some employees viewing AI Gen as unethical or scam-like. One user said: “I find Copilot ethically questionable: extremely high energy consumption, based on shameless privacy and rights violating practices.”

Others feared upsetting routines, making mistakes, or being scrutinized for AI-generated results. One user said: “I'm afraid that what I do with Copilot is wrong or that I will make mistakes in my work.” Such attitudes have led employees to be reluctant to integrate technology into their workflows. Concerns over ownership of results and changes in workflows have fueled further resistance to AI Gen.

This subtle but real cultural resistance can slow adoption even in high-tech organizations. Novo Nordisk approached the challenge with a multi-faceted strategy, focused on transparency and trust. The company introduced guidelines for ethical use, clarified expectations for ownership and disclosure of results, and launched the “Spend time to save time” campaign to redefine AI Gen as a strategic factor, not a shortcut.

Mark Navas, the Novo Nordisk executive responsible for implementing Copilot, reinforced this message: “Copilot is about enabling our employees to work better, not taking shortcuts.” Regular feedback, including surveys and usage analytics, has allowed the company to monitor resistance and adjust support. Champion-led demos normalized adoption by showing real-world examples of successful use of AI Gen. Safe spaces like the Viva Engage platform allowed employees to ask questions, share concerns, and gain trust without fear of being judged.

Leaders who are implementing AI Gen can adopt several strategies, as Novo Nordisk did, to mitigate cultural resistance,transform resistance into engagement and pave the way for sustainable adoption.

Clarify ethical use.Develop and communicate guidelines on the use, ownership and disclosure of AI. Make sure employees understand how to integrate AI Gen seamlessly.

Normalize the use of AI Gen across champions.Employ experienced employees to demonstrate practical and ethical applications, which make AI Gen more understandable and credible.

Promote safe spaces.Build internal communities for peer support, where employees can express their concerns and share successes without being judged.

Proactively address trust issues.Train employees on data privacy, environmental impact and responsible use to build trust and counter ethical objections.

Redefining the role of AI.Position AI Gen as a tool to improve the quality of work, not replace human effort, through consistent messaging from leadership.

People are the platform

Novo Nordisk plans to further expand its Copilot deployment from approx20milaa37 milesemployees in 2025. (The company has approximately75 milesemployees worldwide). According to Chief Digital and Information Officer (and co-author) Anders Romare, the company's future with AI Gen will depend on the ability to move beyond initial enthusiasm to build systems that support continuous learning, trust and integration into real-world workflows.

As the company continues its implementation, one lesson has become increasingly clear: AI Gen's success depends not so much on the tools themselves, but on the people who use them and how they adapt, collaborate and embrace change.

This success is not only driven by automation, but also by people willing to rethink the way they work, support each other and develop trust in new ways of working supported by AI. Champions become agents of change. Communities become accelerators. And experience, not youth, emerges as the hidden catalyst for adoption.

If you want to successfully scale AI Gen, start with people, not code.

Note before bibliography

The authors would like to thank Jingqi Liu, PhD student at ETH Zurich, for contributing to the writing of this article.

Bibliography

T.H. Davenport and R. Bean, “Five Trends in AI and Data Science for 2025,” MIT Sloan Management Review, April 15, 2025, https://sloanreview.mit.edu.E. Brynjolfsson, D. Li e L. Raymond, “Generative AI at Work”, The Quarterly Journal of Economics 140, n. 2 (maggio 2025): 889-942.P. Puranam, “Re-Humanize: How to Build Human-Centric Organizations in the Age of Algorithms” (Singapore: Penguin Random House, 2025).Y. Sun, D. Sheng, Z. Zhou, et al., “AI Hallucination: Towards a Comprehensive Classification of Distorted Information in Artificial Intelligence-Generated Content,” Humanities and Social Sciences Communications 11, no. 1 (September 2024): 1-14.S. Raisch e S. Krakowski, “Artificial Intelligence and Management: The Automation-Augmentation Paradox,” Academy of Management Review 46, n. 1 (gennaio 2021): 192-210.