In addition to culture, the Jaarbeurs talked a lot about data quality, innovation and responsible data use. Organizations are searching for balance in a world full of possibilities. A report from the first day.
Less is more
"I've been here in previous years and back then it was still very much about collecting data," said Picnic's Daniel Gebler on the main stage of Data Expo. Nowadays, it's more about quality than quantity, according to the CTO of the supermarket on wheels. In fact, according to the keynote speaker, data quality is driving Picnic's AI-driven supply chain. Philips' Ger Jansen, who focuses on ethics and compliance, said that bias (distortion) will increase as AI continues to develop: the larger the data sets, the more synthetic results, the lower the quality. "I wonder if solutions like ChatGPT will become more accurate in the future," he said on the main stage. Earlier in the day, Roberto Flores of Magnum Ice Cream Company had told how the ice cream company deploys Small Language Models to perform tasks in a targeted way with smaller language models, which have less computational power but also lower costs than larger models. The message the three speakers delivered separately: less is more.
Towards a decentralized data organization
They are not yet as data mature as Picnic, Philips and Magnum at Ennatuurlijk . "Can we still get by with gut feeling?" was the question Eric Vanderfeesten asked himself aloud some time ago. Answer: no. The sustainable heating company started a project toward data-driven work, he told Data Expo, with the help of consultants from Digital Power. "If you do it alone you have a chance of tunnel vision," he pointed out of the added value of a partner. "Now we can draw on knowledge gained elsewhere in the market." Together they shaped a data strategy, based on different ambitions to produce optimally and match supply and demand. But that was just the beginning, according to the Manager, Digital, IT & Data. "A strategy alone does nothing." Much attention was therefore paid to "coverage" in the organization, which was increased with active communication and walk-in moments for people with questions. In addition, Ennatuurlijk invested in the internal level of knowledge about data: "It's a matter of speaking the same language and repeating it a lot," Vanderfeesten said. To implement the data strategy, ambitions were translated into roadmaps and a new scalable data platform was taken into use. The data base is now in order at Ennatuurlijk, which is preparing for the next step: "The change from one central data team to a decentralized data organization. That is a challenge, with new dimensions in terms of culture, working methods and competencies. Two pilots are now running in which teams are working with Digital Power on ownership and own initiative."
Internal visibility and engagement are essential to a successful data strategy, Vanderfeesten said. "Celebrating success helps to be visible, as does the 'Data Tuesday' we created to share findings and discuss employee questions."
Data contract-first
Rituals was facing another data issue, Lead Data Architect Melle Boersma told Data Expo. "How do we standardize data practices while respecting our character traits?" by which he was referring to the simple organizational structure, Rituals' can-do mentality and embrace of chaos. Rituals is growing fast, averaging 20 percent in recent years, stacking layers of features on top of each other. "Growth creates complexity, and complexity is the silent killer of growth," he cited a well-known consulting quote. In the search for standardization, the Dutch cosmetics company adopted an Enterprise Data Platform at the hand of IT consultant Xebia. Certainly as important as the technology are the people and the processes, Boersma explained. Rituals connects the three pillars with data contracts and metadata. "That's the glue for the whole transformation from project-based to product-based tech teams. The other day I heard someone say that Rituals has become a data contract-first company, I like that." Boersma also stresses the importance of visibility to bring the organization along. "Share data stories and celebrate successes everywhere."
Talent with passion
"Culture outweighs suboptimal decisions," said Daniel Gebler of Picnic at Data Expo. The company started 10 years ago "with zero data" and a new proposition in the grocery market: low prices and free delivery thanks to efficient logistics. Failing fast is part of Picnic's culture, said Gebler: "Shooting down ideas is the key to innovation." To that end, the company is looking for not just any talent, but passionate talent. The more concentrated that talent, the better. Success with automation, such as in Web supermarket fulfillment centers, depends, according to Gebler, mainly on the people behind the buttons. The days of collecting as much data as possible are over; Picnic is all about organizing the data architecture and data-driven learning. Gebler explained how Picnic is moving from MDM (Master Data Management) to a knowledge graph, which adds intelligence about the connection between products, recipes, nutritional values and customer preferences. This allows Picnic to work more efficiently, serve customers better and develop new services faster. The company is also working on self-optimization, according to the CTO: "How can we improve things without having to write a single line of code?"
Bias risk management
"Culture is not an aspect of the game. It is the game," Ger Jansen of Philips quoted former IBM executive Lou Gerstner as saying. Without people no success in a world of data, he too made clear. And, cautioning and drawing on recent MIT research, "If you outsource your brain, you get dumber. So always use AI as an extension of the mind, not as a replacement." In his keynote presentation, the AI Ethics & Compliance Lead dwelled at length on the phenomenon of bias, more specifically on the misconceptions about it. For example, that bias only occurs with input data - there are more than 200 types of bias according to Jansen, it occurs everywhere. Bias-free solutions and humans do not exist, he further emphasized, no matter how loudly otherwise claimed. "AI just gives you a mirror," Jansen said. He explained how Philips has integrated bias risk management into existing processes and how the entire company has started taking AI awareness trainings. Furthermore, assessments on responsible AI and ethics have been rigged. According to Jansen, the knife cuts both ways: not only can responsible use of technology avoid fines from Brussels, but it also contributes to trust and thus customer loyalty.
Small Language Models
Roberto Flores, Global AI & Data Engineering Lead at Magnum Ice Cream Company, also dwelled on the ethical side of data, albeit briefly, in his presentation on Small Language Models (SLMs). Why run large data models when a request can also be fulfilled with smaller models that use dozens of times less energy to do so? Better for the environment, said Flores, and besides, "Money talks." Where the largest LLMs use trillions of parameters, SLMs have up to about 12 billion, Flores defined. "Compare it to a supercomputer versus a pocket calculator. Power versus focus." SLMs are the future, according to Flores, referring to research by chipmaker NVIDIA. "It's about selecting the smallest possible model to perform a task properly. To do that, you first have to answer an important question, which is very often forgotten. What does good look like?"
Agentic AI
Generative AI can do a lot, but it cannot merge four bank brands into one ASN Bank, Lead Data Scientist Steven Laan told Data Expo. Joked: "No, not that yet. But Agentic AI can assist." Laan, formerly working for the now-defunct Volksbank, described how the policy agent and the data lineage agent contributed to the migration. The former analyzed and compared different definitions and rules to help quickly create a unified policy framework for the four labels. The data lineage agent tracked and consolidated the origin of data to speed integration. Laan clarified when single agents can provide relief, when a hierarchy with a superagent provides relief, and when a "group chat" with several equivalent AI colleagues works best. "AI is not a band-aid, don't think of it as a Swiss army knife either," he concluded. "Think of AI agents as enthusiastic robot helpers, inclined to agree with you." Stay critical, is Laan's motto: "You should always test, check and validate."
From BigTech to FairTech
The day ended with a look at the bigger picture: board chairman Willem Jonker of the AI Coalition for the Netherlands (AIC4NL) talked about the state of AI in our country. "Very little, very late," he said of the efforts so far. But it's not too late, according to the program builder foreman. Jonker sees room for what he calls AI FairTech, the development of AI solutions that are ethical, reliable, privacy-conscious, transparent, sustainable and approachable. Things that European lawmakers are pushing for, but that Big Tech players are mostly reproachfully ignoring for now. Local companies have the opportunity to make a virtue of necessity and bring about a paradigm shift, Jonker said in the closing keynote presentation: BigTech becomes FairTech, with an eye for the human touch. Only then, the national level of ambition must rise. "AI is Chefsache," believes Jonker, who will soon hear whether European money will help the envisioned Dutch AI factory get off the ground.