Artificial intelligence is spreading among Hungarian small and medium-sized enterprises, yet SERCO Informatika Zrt. warns that the lack of structured education and secure data handling presents significant business risks. Varga Botond, AI and Innovation Director at SERCO Informatika Zrt., said the period of theoretical debate is over: the question is no longer whether to use AI but which organizations can embed it effectively into their operational processes.
How widely is AI used and for what tasks?
A 2025 Microsoft study ranks Hungary among the world’s top 20 countries for AI adoption, reflecting growing use in the domestic corporate sector. Companies mainly apply AI to automate repetitive, low value-added tasks such as invoice processing, email handling and accelerating content production.
A major concern is Shadow AI: employees often use personal subscriptions—or accounts shared with family members, such as ChatGPT—for work tasks, which can lead to sensitive corporate data being uploaded to public cloud services without proper control.
New legal framework arrives in summer 2027
Data security and the handling of information uploaded to the cloud will face stricter legal rules: the European Union’s AI Act takes effect in summer 2027. Together with GDPR and NIS2, it will define companies’ data-handling obligations.
Organizations that already work with especially sensitive data—such as accounting firms or public administration bodies—can opt today for closed, on-premise solutions. SERCO Informatika Zrt., a Hungarian IT and system integrator with 43 years of history, offers so-called sovereign AI solutions that run on the customer’s own infrastructure, reducing the risk that business secrets reach unauthorized third parties.
Lack of skills is the main hurdle
Besides choosing the right technology, Varga says the greatest current obstacle is the knowledge gap. Although companies are receptive to tools, many lack the basic skills required to use systems effectively and securely.
"You can achieve a very different quality of outcome from AI if a question—the prompt—is professionally formulated… There is a technique, a four-step rule about what context, wording, task and method to use when prompting AI. If someone masters this, they realize they receive radically better answers from AI," Varga said.
To address this training gap, SERCO developed the AI Lab educational platform.
What the AI Lab offers
AI Lab is free after registration and follows a microlearning approach: short video modules of 10–15 minutes designed for busy SME managers and staff. The foundational module guides participants through the basics of AI, generative AI and large language models in about 4–5 hours. The course is mobile-friendly, includes informational units and test questions, and awards a certificate upon completion.
As a special feature, the developers used AI to produce the videos, narration and subtitles, and deliberately left small digital errors in place to give a realistic impression of the technology’s current limitations.
Data governance and the true barriers to implementation
Successful enterprise AI integration is not merely an IT issue: experience shows a project depends roughly 30 percent on the technical infrastructure and 70 percent on management decisions, HR processes and data preparation. Reliable systems need clean, up-to-date data; poor input leads to erroneous, so-called hallucinated outputs.
For example, launching an internal onboarding chatbot requires prior cleanup of corporate documents, connections to CRM and sales systems, and appropriate training of the model on organizational data. If humans remain the ultimate decision-makers, automating routine tasks can significantly boost efficiency—for instance, preparing a personalized, data-driven offer can take minutes instead of days.
Hardware options and sovereign infrastructure
Hardware vendors have begun to respond to Shadow AI and cloud risks: AI-accelerating chips now make it possible to run models on company-owned servers or endpoint devices. Varga considers on-premise deployment the best option from a data-security perspective because data need not leave the corporate infrastructure.
A sovereign setup allows an organization’s information—from organizational charts to employee details—to be uploaded and tied to internal systems, so the AI can learn from trusted, internal sources and provide more comprehensive, context-aware answers.
The cost of delay: market transformation in three to five years
Varga argues that postponing the technological transition can be fatal for market survival. Development has become a compelling force in global competition; he compares its significance to the industrial introduction of electric power in the 1880s. Unlike that earlier shift, he believes the AI-driven transformation will not take decades but rather three to five years, and only companies that rebuild their workflows around the new technology are likely to remain competitive.
Note: this article was supported by SERCO Informatika Zrt.
