A customer stops at a MOL service station for fuel. Inside, there is another reason to stay: a cup of coffee. Across Hungary, those small purchases have added up. In addition to its fuel business, MOL now is one of the biggest coffee-sellers in the country. 

For an energy company thinking about life beyond oil, that everyday transaction tells a bigger story. A service station can become a consumer destination. A waste management network can become part of a circular economy. An industrial process can use fewer resources to deliver the same output. 

Each shift asks the business to see familiar assets differently. Each also raises a question about how AI can help. 

In a conversation with Pranay Agrawal, Co-founder and CEO of Fractal, on the Not Just AI podcast, MOL Group CIO András Svébis describes the work behind that transition. Through MOL’s Shape Tomorrow strategy, one of Central Europe’s major integrated energy companies is expanding into alternative energy, waste management, the circular economy, and consumer services. 

The conversation moves from energy security to customer preferences, then to a challenge that runs through them all: getting people to change how they work. Svébis’s approach is practical. Begin with a business problem. Decide what data is good enough to solve it. And give people a reason to trust a different way forward. 

Beyond oil: transforming before the future arrives 

“What’s going to happen after the era of oil?” 

Svébis recalls that question at the origins of MOL’s strategy. It demands more than a new energy source. It asks what kind of company MOL needs to become. 

The answer stretches across the value chain. In exploration and production, MOL is looking at geothermal energy and carbon capture. In refining, it aims to direct more feedstock toward petrochemicals and higher-value materials. Waste management supports its circular economy ambitions. A network of nearly 2,400  service stations gives it a platform for expanding consumer services. 

In March 2024, MOL announced plans to spend more than $4 billion on green investments by 2030 and reiterated its ambition to reach carbon neutrality by 2050. These commitments put scale and a time horizon around its preparations for the future. 

AI’s role sits within that business direction. It can help MOL understand where resources go, how operations perform, and where customer relationships hold further value. Across factories, supply chains, recycling networks, and retail, better decisions can support the transition already underway. 

For Svébis, the strategic question comes first. The value of AI depends on how well it helps answer it. 

The energy trilemma: three objectives, one opportunity 

Preparing for the future also means keeping energy available and affordable today. Energy companies must balance supply security, affordability, and sustainability, the three objectives that can pull investment decisions in different directions. 

“The energy trilemma is a contradictory set of objectives,” Svébis says. Traditionally, he notes, “only two can be achieved at the same time.” 

Consider an industrial facility producing the same output with fewer resources. Its costs can fall. Its environmental impact can improve. Its exposure to constrained inputs can decline. That is where Svébis sees an opening for AI: finding efficiencies that help across all three objectives. 

Agrawal points to Google’s use of DeepMind to optimize data-center cooling. In 2016, DeepMind reported a reduction of up to 40% in the energy used for cooling a Google data center. That represented a 15% reduction in the facility’s overall power usage effectiveness overhead. It was not a 40% reduction in total data-center energy consumption. 

The system used predictive models to recommend more efficient cooling configurations within operating constraints. For industrial leaders, the example shows why AI matters when many variables interact, and improvements are difficult to spot through conventional approaches. 

AI cannot make the competing demands disappear. It can give operators more room to manage them. 

The circular economy is a data problem 

Follow a load of waste through a national network and the circular economy becomes a series of decisions. What has been collected? Where should it go? Which materials can be recovered? What requires further treatment, energy recovery, or landfill? 

In Hungary, MOL’s country-wide waste management operations span municipal collection, recycling, and landfilling. Coordinating those activities means understanding how materials move through a complex physical system. 

“The whole circular economy is about the data,” Svébis says. 

Every movement of material creates information. When that information is visible, an organization can identify bottlenecks, improve routing, and decide how waste should be processed. AI could help optimize those flows, but visibility has to come first. 

Computer vision and robotics offer another possibility: improving waste segregation and recycling. The economics, however, still matter. Svébis cautions that the technology remains relatively expensive. It must reach a cost level where “it would make sense from a ROI perspective.” 

For MOL, these AI applications remain an opportunity to pursue. They also offer a useful test for any business managing physical operations: can you see enough of the system to improve it, and will the improvement justify the investment? 

From fuel stations to consumer destinations 

Back at the service station, the challenge changes. Here, the opportunity begins with a relationship MOL already has. 

MOL’s nearly 2,400 stations across countries in Central and Eastern Europe create daily interactions with millions of customers. Svébis describes the starting advantage simply: “We have a hook, the fuel.” The next task is to give customers a reason to buy something more. 

Coffee shows what that shift can look like. By offering coffee at the pump, MOL’s position as Hungary’s largest coffee seller turns an established fuel stop into another kind of purchase occasion. 

Making those interactions more relevant requires knowing what customers prefer. Loyalty programs, point-of-sale systems, and fleet cards provide information about purchasing patterns. AI adds a personalization layer, helping match an offer to behavior. 

Svébis gives a straightforward example: “If that person is usually buying tea, we can offer a tea coupon.” 

The business opportunity becomes concrete at that level. Customer data can help shape an offer that fits the person receiving it, giving an existing interaction more value. The greater data workflows influence decision-making, the more holistically the business can understand the customer and deliver greater value.  

Start with the business problem 

The possibilities across MOL are varied. The way Svébis evaluates them is consistent. 

“We usually start with a business problem.” In contrast to leaders who start with the technical challenge, Svébis’s strategy to start with the business outcome prioritizes value-creation.  

That starting point links the work to a business KPI before a technology is chosen. The desired outcome might be an improvement in EBITDA or a reduction in production cost. With the outcome clear, the team can assess what kind of AI implementation is appropriate and how success should be measured. 

“Our energy consumption is too high. Can we reduce it?” is one such question. At MOL’s scale, even a 1% reduction would be substantial. 

The percentage is modest. The underlying business problem is not. That connection gives an AI initiative a purpose that can survive beyond the initial experiment and a measure against which its performance can be judged. 

Good enough data beats perfect data 

Once the business problem is clear, another question follows: is the data ready? 

For many organizations, that question can turn into a long pursuit of perfection. Svébis draws a distinction between perfect data and data that is good enough for the task. Agrawal agrees that pursuing perfection can become an impossible and prohibitively expensive exercise. 

Waiting carries a cost of its own. It delays the opportunity to learn from real deployment and discover which gaps actually affect performance. 

The standard still needs to be rigorous. A model recommending a low-risk marketing offer has different requirements from one informing a critical industrial operation. The consequences of an error, the degree of uncertainty, and the availability of human oversight all influence what is acceptable. 

The practical work is to identify the gaps that matter. Some may be managed through validation or human review. Others must be resolved before deployment. Making that distinction allows teams to invest where better data will make a meaningful difference. 

Seen this way, data quality becomes a business investment decision, with a purpose and a proportionate cost. 

Moving beyond personal productivity 

AI often enters an organization through an individual’s working day. A task becomes faster. A tool reduces effort. Those gains are useful, but Svébis believes the larger opportunity sits in how the company works as a whole. 

“Currently, AI is mainly focused on personal productivity,” he says. “If we elevate it to where the company does the processes better, that will be the huge breakthrough.” 

Agrawal shares that view: “If AI is used only for personal productivity, we are going to miss out on a very large opportunity.” 

The shift is from improving a task to reconsidering the process around it. How are decisions made? How does work move between teams? How should operating structures change when AI can take on parts of the workflow? The key is moving beyond the individual and understanding AI implementation on a organizational workflow level.  

These questions bring the conversation back to MOL’s transition. Better individual tools can support employees. Redesigning enterprise processes offers a way to change how the business performs. 

The real bottleneck: people 

That ambition brings a different kind of constraint into view. 

“Technology is not a challenge anymore for me,” Svébis says. Resistance to change is harder to overcome. 

People have built expertise around existing processes. Familiar ways of working carry the reassurance of experience. Asking someone to adopt a new workflow can change their responsibilities, their confidence in a decision, and their understanding of their own role. 

The stakes rise when AI moves beyond providing information. Would a leader trust it to help develop an operating expenditure plan? Would a team accept a recommendation that challenges a process it has used for years? Who remains accountable when that recommendation influences the outcome? 

Agrawal’s response centers on retaining meaningful human agency as work is redesigned. He describes three complementary foundations that are fundamental to successful AI implementation. 

Agrawal’s AI Implementation framework consists of:  

1. AI-led business transformation: Redesign processes so machine learning, LLMs, and agentic AI can improve enterprise performance while retaining human judgment and agency. 
   
   

2. AI foundations: Establish the data, security, governance, and traceability required to scale AI responsibly. 
   
   

3. AI workforce transformation: Prepare employees to use new tools confidently and adapt to changing responsibilities. 

Together, these address the question Agrawal poses: “How do you redesign the new process so that the work that people are doing is reshaped, yet you are retaining the agency in the person who’s doing the work?” 

“Technology is not a challenge anymore for me.”

András Svébis

On the difficulty of changing established ways of working

“Technology is not a challenge anymore for me.”

András Svébis

On the difficulty of changing established ways of working

In an industrial environment, operational knowledge and accountability remain essential. A stronger process needs to give people better information and reduce unnecessary effort while preserving their ability to exercise judgment. 

The most important skill may be unlearning

Even with the right process, trust takes time. Svébis points to a habit familiar to many organizations: people have well-established reasons for doing things a certain way, and those reasons can outlast the conditions that created them. 

The familiar approach continues to feel safer. AI asks people to revisit that assumption. 

Agrawal illustrates the shift through his experience with autonomous driving. Learning to trust a system with more of the driving means questioning the belief that personal control is always the safest or most effective approach. 

In an enterprise, a similar adjustment happens when people try a new workflow or consider a recommendation they would once have dismissed. Expertise still matters. It provides the judgment needed to test a different approach and understand where it works. 

Unlearning means being willing to examine an established practice rather than carrying it forward by habit. For leaders, that makes adaptability something to build into the organization, alongside the technology itself. 

Shaping tomorrow requires more than technology

The conversation ends with people, the thread connecting every part of the transformation. 

Drawing on his experience at P&G, Svébis emphasizes that “people are the most important assets.” Leadership determines how new capabilities become meaningful change: whether employees understand the direction, feel able to question established practices, and take an active role in what comes next. 

That is the work beneath MOL’s Shape Tomorrow strategy. Investments in new businesses and technologies prepare the company for a future beyond oil. Its people have to make those changes work in daily operations. 

For other enterprise leaders, the conversation offers a practical place to begin. Define the business problem. Connect it to a measurable outcome. Build the data and controls the decision requires. Bring employees into the redesign of their work. 

The coffee counter is a small example of what seeing an established business differently can produce. Across the enterprise, AI opens further possibilities. Turning them into performance will depend on people willing and able to shape what comes next. 

Written by Vanessa Thompson