Zalando has deployed AI-powered dual-arm robots from Sereact to process returned fashion items at two Ceva Logistics-operated sites — Greven, Germany and Świebodzin, Poland — with the systems already running in live production under a robotics-as-a-service model.

The deployment targets the hardest link in Zalando's reverse logistics chain. Returns arrive in unpredictable condition and configuration, unlike outbound fulfilment, where products come from known inventory locations in predictable forms. The Sereact systems use the company's Cortex physical AI platform to identify, grasp and sort items without individual training per product or a predetermined SKU list.

Zalando, Ceva and Sereact describe the two deployments as the first phase of a broader effort to automate returns across Europe.

What does the technology change?

"Returns are the hardest manipulation problem in fashion logistics. If a system only works on articles it has seen before, it does not work in a returns operation at all," said Ralf Gulde, CEO and co-founder of Sereact. "Running this in production with Ceva at Zalando volumes is the strongest proof point we have."

Rather than article-specific programming, Cortex uses AI to dynamically identify items, letting the robots manipulate objects they have not previously encountered. The stated goal is fewer manual touchpoints on repetitive physical work, with employees shifting toward station supervision, exception management and quality control.

How big is Zalando's returns problem?

The scale is substantial, according to Zalando's own figures:

• Around 50 percent of items ordered across its markets are returned.
• More than 96 percent of returned items are classified as in "ideal condition" at inspection.
• About 98 percent of returned fashion items can be re-offered through Zalando or its Lounge platform.

Zalando runs a European network of roughly 20 specialised return centres, where merchandise is inspected, sorted and, where needed, cleaned, ironed or repaired before re-entering inventory. Predictive analytics decide where returned items are stored, factoring in recent sales and inventory levels across sites.

Where does this fit in Zalando's automation build-out?

The Sereact project adds to an already accelerating robotics programme. In March, Zalando announced plans to deploy up to 50 AI-powered robots from Nomagic across its fulfilment centres for item-level picking, scanning and induction into automated sorting — including technology developed for the difficult task of picking shoeboxes. That rollout followed a pilot of 10 robots across three sites in Germany and Italy, with combined capacity of 100,000 picks per day and further systems planned for the Netherlands, Sweden, France and Germany.

As of early May, robots across the fulfilment network were handling about 2 million picks per month, according to Zalando.

Who owns what in the deal?

Zalando holds a financial stake in Sereact's success. In July, the retailer participated as a strategic investor in Sereact's $116 million Series B round.

The same month, Zalando extended its partnership with Ceva Logistics — a collaborator since 2009 — with logistics agreements covering fulfilment, returns and inbound operations in Germany, the Netherlands and Poland extended through 2030. Ceva operates returns infrastructure for Zalando, including facilities where workers inspect every returned item and decide whether it returns straight to inventory or requires refurbishment.

Ceva, owned by CMA CGM, brings high-volume returns experience beyond the Sereact deployment. The 3PL says it processes up to 450,000 returned items per week during normal operations for another European fashion player at a returns facility in Poland, rising to 700,000 in peak periods. That operation handles about 1.3 million returns per month with a 92 percent salvage recovery rate.

For sourcing and operations teams, the takeaway is direct: European fashion returns processing is moving from manual inspection lines toward AI-driven robotics capable of handling unpredictable, untrained items — and the 3PLs and retailers that have already automated at volume now have a measurable productivity benchmark.