The modern milkman: how Picnic used data to reinvent grocery shopping

At NCAS26, Frederik Nieuwenhuys demonstrated how the online supermarket became a game-changer by fundamentally rethinking its distribution model.

During the National Congress on Autonomous Systems (NCAS’26), Frederik Nieuwenhuys, CTO of Picnic, demonstrated how the online supermarket became a true game-changer by fundamentally rethinking its distribution model. He described the company as the “modern milkman” and pointed out the striking similarities between their iconic electric delivery vans and the small milk carts that used to drive around the United Kingdom in the 1970s.

When Picnic launched, the company successfully tackled three major barriers to online grocery shopping: high delivery fees (through price parity and free delivery), unpredictable wait times (with reliable 20-minute delivery windows), and cumbersome ordering processes (via an intuitive, app-only “three-minute grocery run”). The company now generates more than €2 billion in revenue, processes hundreds of thousands of orders per week, and is rapidly scaling up in the Netherlands, Germany, and France.

To manage this enormous logistical puzzle, Picnic relies heavily on autonomous systems, data, and machine learning—which is precisely why Nieuwenhuys was invited to NCAS26. A key component of their success is an advanced routing algorithm that solves the complex“traveling salesman problem.”Remarkably, this algorithm was originally developed by a 21-year-old mathematics intern at TU Delft, and it remains among the world’s best on public datasets.

Machine learning predicts everything: from the exact number of bananas needed tomorrow to the precise duration of a delivery window, with continuous adjustments based on the individual delivery person’s experience. In addition, Picnic uses hardware telemetry to ensure the traffic safety of its young delivery drivers. By monitoring accelerometer data for speed and cornering, the company developed an automatic “driving coach” that proactively alerts drivers at roundabouts, for example—thereby fostering a culture of safety.

Behind the scenes, Picnic expects to pick more than a billion items this year. Many products are still picked by hand—especially heavy items like water, which require ergonomic solutions—but the company is rapidly scaling up its use of physical robotics.

Picnic uses high-speed conveyor belts equipped with cameras and Large Language Models (LLMs) to automatically count inventory, detect damaged products, and determine the orientation of items. Because robotic arms struggle with neatly arranged products but handle chaotic stacks well, this vision system helps determine immediately whether a crate should go to a robot or a human picker.

However, implementing robotic arms also presents unique physical challenges. Nieuwenhuys shared a striking example involving sugar: a few loose sugar granules were sucked into a robot’s vacuum pump, heated by the airflow, melted, and damaged the pump. These kinds of practical details illustrate exactly what obstacles Picnic must overcome as it scales up its robotic picking cells to operate 20 hours a day.

In addition to physical logistics, Picnic is fully committed to generative AI to accelerate its digital development. Nieuwenhuys explained that between 300 and 500 employees now start their workday using AI tools for code generation. This integration goes a long way: designers and product owners can now build functional software prototypes themselves without the need for developers, and data analysts use AI to write complex SQL queries directly.

AI is even being used in HR: voice AI agents conduct the initial screening interviews for a portion of the 400,000 job applications Picnic receives each year.

From smart route algorithms to robots that move warehouse crates: Picnic’s presentation made it clear that true autonomy is only possible with a deeply curious, entrepreneurial culture—one in which engineers dare to launch new ideas, learn from mistakes, and only then scale up.

Source: ioplus.nl