Primetime for Big Data 2026 programme

A few years ago, Data Mesh was one of the most widely discussed concepts in the data world. Today, the spotlight has shifted to generative AI and AI agents. Yet many organizations are making the same mistakes in their AI adoption that once held back their data transformations.

What does a company need to get right in order to truly scale AI? Why isn’t deploying an LLM or an internal chatbot enough? And how do data quality, data products, governance, and domain ownership determine whether AI agents will actually work across the organization?

We will explore why the future of enterprise AI may not begin with models, but with the foundations we build them on.

How do you move AI from ideas and pilots into the actual day-to-day operations of a company? We will show two sides of the same story: the AI/Product perspective on use case selection, portfolio management, delivery, adoption, and value measurement, and the Business Owner perspective on what implementing AI really means for processes, people, and everyday operations.

Drawing on concrete experience from NN, we will share not only what worked, but also where we encountered challenges and what we would do differently today.

Over the past year, AI has gone from an experiment to an everyday working tool at our client companies. Answering a question like “Why did my margin drop yesterday?” now takes minutes instead of hours, and e-commerce business owners are vibe-coding their own applications on top of their data to manage inventory or marketing campaigns.

At the same time, we’ve learned where the real challenges lie. Without a single reliable source of truth, AI simply delivers wrong answers faster. And without a clear objective, it can get lost in the details.

Using real examples from retail and e-commerce, I’ll show what has genuinely changed in data analytics, why the focus is shifting from reporting to data quality and business context, and how analysis is turning directly into action, from product feeds and pricing to orders. I’ll also highlight the use cases that have proven most valuable.

Finally, we’ll explore what remains the biggest challenge, even after these advances, on the journey towards becoming a truly data-driven company.

Experienced manufacturing engineers are becoming increasingly scarce, and estimating the cost of a CNC component based on their expert judgment is time-consuming and can vary significantly from one person to another.

At MECASYS, we decided to capture this know-how in a data model – from cleaning years of accumulated manufacturing data, through feature selection and model choice (where, in the end, the deciding factor was not percentage accuracy but the deviation measured in minutes), to the most challenging part: earning the trust of the engineers who were initially expected to reject the project.

In this talk, I will walk through the entire process and show how the application now enables three manufacturing engineers with different levels of experience to produce nearly identical estimates – in seconds rather than hours.

Behind every car lies far more complexity than meets the eye. Millions of possible combinations of features, engines and configurations, changing customer preferences and constant pressure to improve manufacturing efficiency.

In-stock vehicles play an important role in this process. They help automakers maintain stable production while responding flexibly to current demand.

But how do you select the one vehicle, out of millions of possibilities, that best meets the needs of a particular customer?

Using the development of a recommendation engine for one of the world’s leading automakers as a case study, we’ll show how to bring together and analyze vast amounts of data. From production capacity and logistics to customer preferences, purchasing power and demographic trends, as well as pricing strategies, location and competing offers.

Most importantly, we’ll demonstrate how to turn this data complexity into a specific, relevant recommendation for an in-stock vehicle.

What can quantum computers already do with data today, and where do their capabilities still fall short? We will explain the basic principles of quantum computing in an accessible way, explore its current capabilities and limitations, and, most importantly, look at what it could mean for data analytics and machine learning.

Using two concrete examples of quantum machine learning, we will look at time-series forecasting in the energy sector and fraud detection. You will see where quantum computing may already have practical applications today and where we will still have to wait for it to deliver real value. No need to brush up on quantum physics beforehand. We will start with the basics and quickly get to what matters most: what quantum computing can bring to the way we work with data.

Recent advances in AI make it possible to close the research loop: hypothesis, experiment, analysis, next hypothesis. Academic and industrial labs are adopting this pattern, compressing the cycle time of discovery — first in mathematics, computer science, and other domains where validation is computational, and increasingly in biotech and chemistry, where a hypothesis must be tested in a physical lab.

I will walk through case studies of working systems, then ask the harder question: what happens to the training of young scientists when the discovery path can be skipped? We value the journey and reward only the result. If AI supplies results faster than humans can follow the reasoning, is human centrality in science something we can defend — or only something some of us prefer?