There is a pronounced divergence in AI adoption among Hungarian companies: the largest banks and telecoms have moved ahead, while many insurers and mid-sized financial firms lag behind. This assessment comes from Csorba Gyula, CEO of Clarity Consulting, and Kormos Benjámin, head of BI & Data Solutions at Clarity Consulting, in an interview with Financial IT.
Telecom companies often benefit from greenfield investments and microservice-based development, which make it easier to design solutions around agents rather than monolithic systems. By contrast, legacy infrastructure and differing regulatory constraints present major bottlenecks for many financial institutions when trying to integrate AI architectures.
Which use cases already deliver value?
Practical AI use cases have emerged that yield measurable resource savings: code generation and replacement of simpler workflows already create visible value. Kormos described an example where, at a client, developers ran out of token allowances and reported that working without AI was no longer efficient — tasks that would take developers four hours can be coded with AI in ten minutes and then validated with AI in another thirty minutes.
The real breakthrough, however, will occur when organisations stop thinking about AI as merely copying or substituting human work and start redesigning processes, tasks and software development around agentic AI — with precise decision logic, suitable knowledge contexts and built-in control points.
Organisational barriers and internal divides
According to Clarity Consulting, technology itself is rarely the main obstacle. More decisive factors are management’s willingness to adopt an organisation-wide AI strategy, the expectation of quickly measurable results rather than multi-year adaptation, and the need to build adoption capabilities in a fast-changing technological ecosystem. A critical risk factor is poor employee involvement: in many companies there is an AI-evangelist group that pushes forward, but in many cases 90 percent of the workforce is not properly included in the change process, creating internal rifts.
As a result, AI adoption is increasingly treated not as an IT project but as a culture-shaping HR project.
Efficiency gains and the fate of jobs
Efficiency gains from AI raise questions about the fate of roles that historically handled administrative or mechanical tasks. Csorba does not observe mass layoffs; instead, he expects job profiles to evolve and administrative duties to shrink. Freed capacity is typically used to reduce backlogs and focus on higher-value activities.
Data, cybersecurity and regulation
Hungarian and EU financial institutions operate under strict regulation — GDPR, NIS2, DORA, AI Act and other directives. That makes the question of which corporate data an AI system can access especially sensitive. Kormos identifies two main challenges:
- integrating systems with data-security in mind, which requires a Policy Gateway layer to govern which employees can use which models (open or closed) and agents, along with guardrail features to control communications between users and AI and prevent sensitive data leakage;
- managing AI-related costs, where FinOps dashboards and license governance will become central.
Many firms currently rely on major cloud providers — Microsoft Azure, Amazon AWS — accepting the providers’ assurances that prompts won’t be used to retrain models. The alternative is private cloud or on-premise deployment: keeping data stored locally in an isolated tenant or physical data centre and running open-weight language models installed on the corporate infrastructure so that the data is not used to train third-party models.
Costs and financial management
Licence-based access can be expensive: the article notes that subscribing to multiple services can lead to several dozens of dollars per user per month (for example, three different services at roughly 3×20–25 dollars monthly), which becomes costly at organisational scale. Clarity recommends an API-driven model invocation framework to centralise and better control costs. In the long run, managing AI costs will be a strategic task shared by CFOs and CIOs.
Platform competition and emerging models
Competition between platforms and models is complex, influenced by technological, political and macroeconomic factors. Recent months have seen attention on Claude and Anthropic, as well as Google’s developments; Kormos believes there is no clear winner yet but sees Google and its Gemini as well-positioned given integration with daily-use products such as Gmail, YouTube and Chrome, and partnerships like the one with Apple for Siri.
New tools that uncover software vulnerabilities — for example, Anthropic’s Mythos model, which reportedly found long-standing bugs — raise questions about trust and software security if such capabilities become widely available.
Conclusion and event
Clarity Consulting’s experts conclude that the Hungarian market is split: companies that invested in innovation and digitalisation over the past decade are now well placed in AI, while many others are still exploring how to start. The topic will be discussed in depth at the Portfolio Financial IT conference on May 28, which focuses on the AI boom, fintech innovation and digital transformation.
The article was supported by Clarity Consulting. Photos by Mudra László.
