Generative AI and the question of real value for Vietnamese enterprises

Over the past few years, artificial intelligence – especially generative AI – has become one of the most frequently discussed topics in digital transformation strategies. From executive boardrooms to the day-to-day work of engineering teams, AI is everywhere, accompanied by high expectations around productivity, automation, and operational breakthroughs.

However, real-world implementation tells a more nuanced story. Not every enterprise that invests in AI ends up realizing proportional value. Many initiatives stall at the pilot stage: models may technically work, but fail to create tangible impact on business operations.

Drawing on real-world observations from large-scale digital transformation projects, Mr. Nguyen Trung Hieu (AI Computer Vision Team Lead, FPT) unpacks the maturity of Vietnam’s AI ecosystem. The article also outlines a more sustainable approach to AI – one that treats AI as a long-term organizational capability rather than a short-term technology initiative – enabling enterprises to turn data into a genuine competitive advantage.

Where is AI in the technology cycle?

On a global scale, AI is not a new story. Over the past two decades, the field has advanced rapidly, driven by major research breakthroughs from technology giants such as Google and Meta, supported by massive computing infrastructure and hardware ecosystems led by players like Nvidia.

Before the emergence of ChatGPT, most enterprise AI deployments fell under the category of traditional AI, focusing on relatively deterministic problems such as recognition, prediction, and automation. In Viet Nam, applications like AI-powered traffic surveillance cameras, eKYC, and document processing automation have been widely deployed and have proven their real-world value.

The arrival of ChatGPT and large language models marked the beginning of a new phase – often referred to as the era of generative AI. Its rapid development, broad applicability, and far-reaching influence have placed GenAI firmly in the spotlight. However, from Nguyen Trung Hieu’s perspective, this is still an early stage of the technology cycle, where potential and challenges coexist, and many assumptions have yet to be validated through large-scale, real-world deployment.

Where do Vietnamese enterprises stand?

Viet Nam has a strong foundation in software outsourcing and technology implementation. In the global AI wave, the domestic market is witnessing a surge in GenAI applications that are built by “standing on the shoulders of giants” – leveraging existing platforms and pretrained models to bring value closer to end users.

At the level of core research in generative AI, the playing field remains dominated by major global technology corporations. In contrast, the main driver in Viet Nam lies in applying AI to solve concrete operational and service-related problems.

In practice, Vietnamese enterprises currently exhibit two parallel approaches to AI adoption. A small group has sufficient resources to invest in developing in-house AI capabilities. The majority, however, focus on applying AI to generate direct value for customers and internal operations. Traditional AI solutions have already demonstrated effectiveness across sectors such as banking and finance, healthcare, transportation, and public services.

With generative AI, Viet Nam is generally assessed to be one to two years behind global markets in terms of technological maturity and operational readiness. While there have been notable examples of GenAI deployment in areas such as government, education, and healthcare, most enterprises remain in the experimental phase, searching for viable and scalable models.

Why is it so hard to move from pilots to real value?

One recurring pattern across AI projects is the gap between expectations and implementation reality. Generative AI unlocks many new possibilities, but the technology itself is still evolving and requires extensive experimentation and fine-tuning to become effective and reliable. When expectations are set too high from the outset, projects can easily end up in a state where they “run, but are not usable.”

Moreover, GenAI demands a high level of consistency in training data quality, problem-specific context, and domain documentation. Concepts such as prompt engineering or context engineering only deliver real results when enterprises have solid, well-governed, and standardized data foundations. The lack of data standardization and consistent context remains a major barrier to stable AI outcomes.

At scale, the challenge extends beyond technology. Governance, monitoring, and user training become critical. Helping users understand how to use AI correctly—and effectively—at scale is no small task, especially as enterprises expand their scope of application.

What determines the success of an AI project?

Based on real-world deployment experience, successful AI initiatives typically start by treating AI as a technology application—and every technology application should originate from end-user needs. Instead of leading with technology, enterprises need to clearly identify the problems that deliver the greatest value, offer measurable ROI, and directly address operational or customer experience pain points.

Data remains the decisive factor. The principle of “garbage in, garbage out” still fully applies to AI. Input data quality directly determines output quality, and in many cases, data itself becomes a long-term competitive advantage that enterprises uniquely possess.

For generative AI, a cautious approach – starting with small, measurable, and quickly verifiable use cases – often yields better results. This allows enterprises to learn incrementally while managing risks in a rapidly evolving technological landscape.

How does FPT accompany enterprises on their AI journey?

FPT positions itself not merely as an AI solution provider, but as a long-term partner supporting enterprises throughout the development and maturation of their AI capabilities. From consulting on problem selection and sharing implementation and data governance experience, to deeply integrating AI into business processes and measuring real usage effectiveness, FPT participates across the entire journey.

A representative example is Azinsu, an AI-powered insurance underwriting automation solution. Azinsu combines multiple core FPT AI technologies, including automated document processing, domain-specific information retrieval, and explainable payout recommendations. This enables underwriters and customers to engage in more transparent, evidence-based decision-making processes.

Viewing AI as a long-term capability

Rather than treating AI as a project with a clear start and end, a more sustainable approach is to view AI as a long-term organizational capability. There is no universal formula for all enterprises. Large corporations may strategically choose to invest across both research and application layers. For small and medium-sized enterprises, long-term capability is often built through the successful deployment of multiple specific AI use cases, grounded in internal experience that is difficult for external competitors to replicate.

Learning from both the successes and failures of early adopters is another effective strategy—helping reduce experimentation costs and shorten the time required to create value.

A perspective for Vietnamese enterprises in today’s AI landscape

From real-world observation, AI is no longer a distant future concept. Its capabilities are already embedded in everyday operations. However, the path to sustainable AI-driven value rarely begins with grand moves. Instead, it starts with choosing the right problems—areas where enterprises have the deepest expertise and the clearest understanding of user needs.

In this context, the enterprise data challenge is no longer about storing or querying billions of records. The more important question is how to transform that data into a knowledge network—one where humans and AI can interact, collaborate, and co-create value. This is the foundation that enables AI not just to “work,” but to truly become part of an enterprise’s long-term operational capability.