Sereact AI Robots Go Live on Zalando Returns in Germany and Poland

Robots that can handle a bag of potatoes

The most telling detail in this week's warehouse-automation announcement is not a spec sheet. It is a bag of potatoes. Sereact's chief executive, Dr. Ralf Gulde, recounted that the strangest item ever returned to Zalando was exactly that — a sack of spuds — and that his company's robots have had to learn to distinguish genuine merchandise from packaging material and outright garbage2. That anecdote captures why returns handling has long been the stubborn last frontier of warehouse robotics, and why the announcement that dual-arm robotic systems are now processing Zalando fashion returns in live production at CEVA Logistics facilities in Greven, Germany, and Świebodzin, Poland, matters beyond the press release16.

The three companies — Stuttgart-based Sereact, fashion platform Zalando, and CMA CGM-owned CEVA Logistics — confirmed the deployment on October 7, describing it as the first operational milestone of a strategic partnership intended to scale AI-driven returns automation across Europe1. The systems run on Cortex, Sereact's "physical AI" software platform, and are delivered under a Robotics-as-a-Service (RaaS) commercial model13.

Why returns broke robotics for a decade

Conventional warehouse automation excels at predictability. A robotic cell that picks a known SKU from a known tote, in a known orientation, can run for months. Returns break every one of those assumptions. Returned garments arrive crumpled, half-packed, missing tags, mixed with tissue paper or — occasionally — potatoes. No two items present the same way, which is precisely why pre-programmed robotic cells have never gained traction in returns operations12.

Cortex's answer is to handle objects it has never seen before. The platform groups unfamiliar items, identifying, grasping and sorting them dynamically without article-specific training or a pre-determined SKU catalog11. In Gulde's framing, that capability is not a nice-to-have but the entry requirement: a system that only works on articles it has seen before "does not work in a returns operation at all"12. Running the technology in production at Zalando's volumes, with CEVA operating the sites, is the strongest proof point the company can offer, he said12.

The operational stakes are well documented by the logistics side of the partnership. CEVA's SVP for contract logistics, Chris Walton, catalogued the pain: unpredictable volumes, manual quality checks, seasonality, value decay, and a disproportionately high cost per unit13. His pitch is that the RaaS model classifies items faster, cuts manual touchpoints, and delivers consistent quality at scale — converting returns, in his words, from a cost driver into a controlled, high-performance service13.

The numbers that made Zalando take the risk

The efficiency gap the robots are closing is stark. Gulde disclosed that Zalando's returns stream can hit 500 to 600 units per hour, while human workers manage roughly 20 to 30 units each — a throughput mismatch that made the process structurally expensive, since robots, unlike people, do not take breaks2. The system can also grade, fold and repack goods in a presentable condition, handling the reconditioning work needed to push items back into sellable inventory2.

The commercial courtship was notably fast by enterprise-standards. Zalando visited Sereact's demonstration, returned two weeks later, and — as Gulde tells it — was persuaded less by polish than by the pace of the system's learning2. In July 2026, that courtship turned into capital: Zalando joined Sereact's Series B round as a strategic investor, a $116 million raise for the Stuttgart start-up6. The October deployment therefore links an investment decision directly to a live operational outcome, an unusually tight loop between capital allocation and warehouse floor.

Sereact is not a newcomer experimenting on its first customer. Beyond Zalando, its roster includes IKEA, DHL, Rohlik, Arvato, alza.cz, BMW, Mercedes-Benz and Daimler Truck. The company's method has been to build on the shop floor rather than in the lab, accumulating data on roughly one billion picks — from groceries to metal parts — in single-arm pick-and-place tasks before graduating to harder problems like automotive kitting and returns handling2.

The people question

Every warehouse automation story now carries a labour dimension, and the companies are getting ahead of it. The official framing is that the robots absorb highly repetitive handling, reduce physical strain, and free employees for higher-value work — station supervision, exception management and quality control11. Zalando's SVP Logistics, Masood Choudhry, framed the collaboration around the belief that "true innovation happens when technology and human potential advance together"13.

The coverage is not uniformly rosy on this point. The Next Web noted, as context, Zalando's separate plan to close its logistics centre in Erfurt, Germany, by September — a closure the works council said affects more than 3,000 workers including temporary and service-provider staff, per dpa reporting from January. That report did not link the closure to automation, and it predates this deployment, but the juxtaposition of "robots go live" and "3,000 jobs end" in the same coverage cycle illustrates the reputational tightrope European retailers walk when automating6.

A fair reading of the evidence supports the augmentation claim at this specific deployment: the Greven and Świebodzin sites remain staffed, with humans repositioned into oversight and quality roles, and the robots handle the picking, folding, grading and repacking grind11. But the longer-term arithmetic — 500–600 units per hour versus 20–30 for a human — suggests that scaling this across a European network will eventually require far fewer pairs of hands per unit processed. That is the unresolved tension the press releases do not address.

Where the reporting agrees and diverges

Across the wire distribution and the trade coverage, the core facts are consistent: dual-arm systems, Cortex platform, Greven and Świebodzin sites, RaaS model, Zalando's July Series B investment, and an explicit plan to scale across a shared European footprint1313. No outlet disputes any element of the announcement, which is unsurprising given that most coverage traces back to a single sourced press release.

The divergence is in depth. The Retail Technology Innovation Hub and Post & Parcel offer straight recaps of the partnership statements3. Automated Warehouse goes further, adding the throughput figures, the potatoes anecdote, Sereact's billion-pick data history, and its customer list — the richest technical reporting of the batch2. The Next Web adds the Erfurt closure context and background on Zalando's broader AI work, including its customer-conversation AI partnership with Parloa6. A financial outlet noted Zalando shares ticked up 1.34 percent to EUR 22.70 on the news, a modest market verdict suggesting investors see the rollout as strategically meaningful rather than transformational on its own.

What this signals for logistics automation

The strategic significance sits at three levels. First, this is one of the clearest demonstrations yet that physical AI — perception-driven, generalisable manipulation — has moved from demo videos into paying production work on one of the hardest problems in fulfillment. As one robotics-industry analysis put it, the deployment shows physical AI graduating from structured picking into workflows demanding real perception and adaptability10.

Second, the RaaS model is doing real commercial work here. By selling outcomes rather than machines, Sereact and CEVA lower the adoption barrier for a technology whose reliability claims are, by definition, hard to verify in advance. For a third-party logistics provider like CEVA — which operates contract-logistics sites for customers such as Zalando across more than 170 countries and 1,700 facilities — RaaS turns automation into something it can promise clients without betting the balance sheet on unproven kit1113.

Third, Zalando's dual role as customer and strategic investor is a template worth watching. When a retailer buys equity in its automation supplier, the pilot-to-scale pipeline shortens dramatically: the supplier gets guaranteed deployment context and data, the customer gets early access and pricing influence. The partners say they are already working to extend intelligent automation across their shared European footprint1, and the choice of two countries, Germany and Poland, signals a pan-European ambition rather than a single-site showcase.

The bottom line

Returns are the cost centre every fashion e-commerce operator would most like to delete, and until now the economics have been locked in by manual labour. If Sereact's machines can sustain — at Zalando volumes, in live production — what they promise on throughput and adaptability without per-SKU training, the returns operation stops being a warehouse problem and becomes a software problem. That is the bet Zalando made with its Series B cheque in July, and the one CEVA is now running on the floor in Greven and Świebodzin. The bag of potatoes, in the end, may be the best benchmark of all: any system that can sort through what customers actually send back, sight unseen, has cleared the bar that kept robots out of the returns room for a decade.