Generative AI: Not a Sprint but a Marathon – Guiding Principles for AI Success

Generative AI (Gen AI) has become a transformative force in business, with the potential to fundamentally reshape how we work. It offers tools to accelerate or automate tasks, improve delivery quality, and foster innovation and creativity.

Unlike traditional AI techniques, Gen AI, powered by large foundational models, is a versatile technology capable of addressing a wide range of use cases within an organization.

This broader scope of transformation impacts every employee. However, Gen AI also comes with inherent risks, including inaccuracies, hallucinations, unethical usage, security breaches, data leakage, and legal challenges, among others.

Despite these challenges, there is no doubt that Gen AI is here to stay, thanks to the significant benefits it offers. Companies that embrace this reality will prepare for the long haul: the marathon, not the sprint. Relying solely on the opportunistic hope of achieving quick wins is no more an AI strategy than winning the lottery is a business plan. While you might get lucky, you’re far better off creating your own fortune by simultaneously strengthening your foundations—data management, governance processes, and people acculturation.

Early successes are essential to building momentum, but don’t let shortsightedness constrain your transformation. In this article, we explore key principles for adopting a marathon approach to Generative AI transformation, ensuring your organization remains competitive and thrives in this rapidly evolving space.

Data quality first

As an organization, you now have access to incredibly powerful AI models such as GPT, Claude, Mistral, and others. But keep in mind: so does everyone else. For most companies, the differentiator between success and failure is no longer the ability to develop the most advanced algorithms, but rather the quality of the data you provide them with. In other words, as foundational models become commodities, the battleground is shifting from algorithmic supremacy to data excellence.

If AI transformation is a marathon, data is the fuel—the runner’s nutrition plan. And like any good nutrition plan, it needs to be healthy. Your data must be up-to-date, accurate, complete, and as unbiased as possible—and it must remain that way over time. This is the number one condition for reliable AI. A good decision based on flawed information is ultimately a bad decision, and no organization can afford to run its operations under such risks. 

Historically, AI, much like traditional BI, primarily worked with structured data—tables of rows and columns. Ensuring the quality of this data was already a challenge. But the game has changed: today’s AI can process unstructured data, such as text files, images, audio, and video. This vastly expands the scope of what AI can do but also significantly increases the complexity of ensuring data quality.

Consider this: the number of products a customer has purchased, stored in a CRM table, is either correct or incorrect—a relatively straightforward validation. But what about a page of meeting minutes summarizing the sales team’s last interactions with that customer? The quality of such unstructured information is far harder to assess. It might be partially accurate, biased, or contain sensitive information. With unstructured data, quality metrics exist on a continuous spectrum, making governance a far more intricate challenge for organizations. This growing complexity demands stronger data management processes. 

Curious to learn more? Check out our white paper "Generative AI - Visa for a more interactive digital future".

Information management remains a human responsibility

As of now, no magic solution exists where AI alone can reliably assess the quality of data being fed into another AI system. It would certainly make things easier, but we’re simply not there yet. Information management remains, for now, very much a human responsibility.

That said, AI can certainly assist with data management tasks to some extent. It can flag sensitive documents, evaluate writing quality, and even detect potential misinformation. However, these capabilities act merely as accelerators. Without human oversight, poor data quality can quickly propagate at scale, compounding inaccuracies and risks.

To prevent this, organizations must implement robust processes to ensure data integrity, focusing on eliminating inconsistencies and addressing gaps in data coverage. Existing data governance frameworks should be strengthened and adapted for the unique challenges posed by Generative AI. Moreover, experts must be incentivized to produce high-quality information, as this human input is the foundation of any successful AI strategy. Improving data quality is not a one-time effort or an overnight transformation. Progress will come incrementally, focusing on one use case at a time. 

Travel light

Creating a flashy Gen AI demo is easy, but deploying it into production is a far greater challenge. Attempting to tackle everything at once can strain resources and lead to setbacks or risks—reputational, financial, and operational.

Instead, focus on small, manageable use cases. Even better, prioritize the core feature of your MVP and put that into production first. This approach allows you to gather feedback, validate governance processes, and build a strong foundation step by step. To cross the “production wall,” travel light—small successes always outweigh big failures.

It’s all about people

In the long term, AI could redefine the very nature of work, profoundly impacting the daily lives of all employees. Already, we see AI narrowing the performance gap between top employees and others, which may eventually prompt a reevaluation of how performance is measured.

As with any transformative technology, what starts as optional soon becomes expected—and ultimately mandatory. Today, using AI in the workplace remains optional for most. However, as its productivity benefits become undeniable, it is likely to evolve into an expectation and, eventually, a requirement.

Leaders must prioritize AI training, equipping all employees with the skills needed to adapt. This proactive approach ensures equal opportunities for everyone to thrive, preventing a divide between early adopters and those slower to embrace the technology.

people marathon

A long-term vision for success

There’s much more to explore when it comes to AI transformation. Topics like accountability, AI governance, compliance with evolving regulations, and the sustainable and ethical use of AI are just a few key dimensions of a successful transition. 

Ultimately, AI is a powerful tool capable of driving industry-wide transformation when approached thoughtfully. Success depends on adopting a marathon mindset: prioritizing information quality, fostering a culture of AI adoption, pacing progress carefully, and maintaining a clear and strategic vision.

Want go further? Watch the replay of the webinar "Generative AI is not a sprint... but a Marathon".

About the author

Jérémy El Aissaoui - AI Expert
 

Jérémy is a former theoretical physicist who traded black holes and string theory for the equally fascinating field of AI, spending the past ten years supporting organizations across diverse industries to solve complex business challenges with cutting-edge technology. With a passion for innovation, he helps businesses achieve their full AI potential with strategic roadmaps, tailored solutions, targeted coaching, awareness sessions and much more. 

Jérémy El Aissaoui

Suivez-nous