Roland Bernáth is a digital business strategist, technology expert and investor, founder of Online Üzleti Akadémia and The Creators League. He works on Hungarian and international projects, advising company leaders to treat AI as part of communication, customer acquisition and operations rather than as a standalone tool. His method, called Cégagy, aims to collect dispersed corporate knowledge into a structured system usable by both people and AI.
What does building corporate context mean?
Bernáth observes that in many Hungarian SMEs significant knowledge is centralized in a single person or scattered across systems: the CEO's decades of experience, personal contacts, or historical decisions whose origins nobody fully remembers. These become bottlenecks — when a new hire arrives or the company tries to scale, much of that knowledge is not transferable.
For him, corporate context means placing the appropriate knowledge into a structured repository that an AI can use. Without that context, an AI agent attempts to assemble answers from chaos and lacks understanding of the company's history, decisions, processes and rules. When a continuously growing corporate "AI brain" exists, it becomes possible to build far better agents and processes on top of it.
Why will some companies use AI better than others?
Bernáth notes that almost every company today has access to the same AI tools (for example, ChatGPT or Claude), so the difference is what the company knows about itself. A generic large language model has limited conversational context; as dialogs grow longer, the model must summarize and select prior information. With a company knowledge system, corporate information can be stored separately and the model can retrieve only the relevant parts when needed.
He compares it to schooling: if someone has all their old textbooks, they can look up a detail from a fifth-grade chemistry lesson even if they don't remember it. He expects the same distinction among companies: the decisive factor will not primarily be whether a firm uses Claude or ChatGPT, but what proprietary knowledge it can put behind the model.
Where to start building a Cégagy?
Bernáth typically begins with processes. The first step is mapping the entire operation, from customer onboarding through quoting and sales to fulfillment. Even this exercise is valuable: telling someone to "make an offer" may not be enough, but specifying product groups, which ones allow discounts and pricing rules gives clear structure.
This applies equally to humans and AI: if processes are missing, people become frustrated and AI produces poor outputs. Often the problem is not the human or the AI itself, but the absence of a clearly documented system to work from.
Making tacit knowledge transferable
Much valuable knowledge exists in experienced employees' heads and is built over long periods, so Bernáth argues against expecting an AI to be "trained" in one day. Instead, he recommends a continuous learning process: summaries from meetings, explanations of decision backgrounds, and contributions from colleagues gradually flow into the system. The AI incrementally learns how the company operates, how decisions are made and what the culture is like.
AI's advantage is that many people can contribute in parallel, the system does not tire, and it can be enriched with external knowledge. With consistent, long-term work, the knowledge becomes cumulative.
Responsibilities and new roles
Currently, many companies lack a single person responsible for corporate knowledge, which creates demand for roles like Chief AI Officer. The meaning of such a role differs between a ten-person firm and a large multinational: in a small company it may define rules for using ChatGPT and what data may be uploaded, while in a large company it could own the whole AI strategy and organizational rollout.
This role need not come with a large team: a competent person can coordinate multiple agents. Today, Bernáth says, the position largely involves education and organizing corporate knowledge.
Quick wins and measuring ROI
The fastest results come where a lot of human work is devoted to digital, rule-based processes — for example accounting and administrative tasks. If a task can be described by rules, AI can perform it repeatedly for many customers.
Return on investment can be measured in several ways. The simplest is time saved: compare how many human hours a task previously required to how many it requires now. Another metric is reduction in errors, which varies by industry; in construction, for instance, a single wrong decision can cause tens of millions of forints in damage. Thus ROI should be aligned to where the company creates value.
Risks: automating past mistakes and security
Bernáth acknowledges the risk that AI trained on a company's past may automate old mistakes and ingrained habits. However, he points out a counterintuitive advantage: AI knowledge can, in principle, be examined and modified more easily than entrenched human habits.
That said, governance questions arise: who is authorized to change corporate knowledge? Proper authentication, traceability of changes and access controls are essential. The same applies to external attacks — if an AI learns from e-mails, weak cybersecurity could allow manipulation via crafted messages. The level of protection required depends on the company's risk appetite and culture.
What should leaders do tomorrow morning?
Bernáth would be satisfied if executives left his talk with the idea that AI can be used for much more than the common uses today. He aims to show how a practical corporate AI brain can look, how dispersed information can be made usable in one place, and what new opportunities that unlocks. If a leader starts the next day by asking where the company's knowledge is, who has access and how it could be organized so AI can use it, Bernáth considers that an important first step.
Conference note
Roland Bernáth's talk was part of the 2026 HVG AI Talks conference. Tickets are still available at https://aitalks.hu/.
The content was produced within the AI Talks event organized by HVG with the cooperation of HVG BrandLab. HVG hetilap and hvg.hu editorial teams did not participate in its production or editing.
