From Warehouse Friction to Network-Wide Opportunity

Everyday Warehouse Friction Adds Up

In a distribution center, the work rarely slows down because of one big problem. More often, it is a steady stream of smaller moments: a wave that needs another check before it can run, a replenishment exception that needs attention, a cycle count that requires preparation, or an outbound supervisor trying to see what might fall behind.

Each moment is manageable on its own, but together, they consume the time from warehouse associates and supervisors who keep the operation moving.

For L’Oréal’s Łukasz Kierzkowski, WMS expert leading the agentic AI deployment, the goal was to put practical support directly into the work already happening on the warehouse floor, then determine whether the results could scale beyond the initial sites.

Seven Agents Across Four Sites, Focused on Everyday Workflows

Running a distribution center means managing a steady stream of checks, exceptions, and decisions. For L’Oréal, teams were spending too much time on the small moments to keep work moving:

• Wave execution
• Replenishment exceptions
• Cycle-count preparation
• Pre-wave checks
• Outbound supervision

But the company wanted teams to spend less time coordinating and investigating, and more time acting on what they found. So, they started with a six-month pilot of seven agents across four distribution centers in Mexico, Poland, Italy, and South Korea, testing how AI could help reduce the manual effort behind everyday warehouse work.

AI Agent Use More Than Doubled Across the Network in 13 Weeks

Technology pilots often stall quietly. They get adopted on paper but ignored in practice. This one didn't. Across the global network, distinct users grew from 100 in week one to 206 in week thirteen, a 106% increase. Within the four pilot DCs, distinct users increased from 44 to 52.

Kierzkowski’s goal was to gradually expand the scope of application without diluting commitment and making adoption difficult. At Manhattan’s EMEA Exchange in Milan, Manhattan CEO Eric Clark echoed this sentiment: “Complexity in the market is an opportunity, but complexity in your software is a tax.” Higher adoption proves the technology is finding its place in day-to-day operations.

The Work Itself Got Cleaner

The pilot showed how small operational gains can compound. At one pilot site, the deselection rate fell from 14.4% to 8.4%, a reduction of 6.0%. At the same site, cycle time dropped from 193 minutes to 76 minutes, or 60% lower. Overall, the team saw a 14% increase in productivity on order preparation, an 8% time savings between order and loading during replenishment, and a 13% decrease in delays between inventory and entry.

The value of AI agents was not only in answering questions but helping warehouse teams identify risks earlier (and faster). When they acted with more context, it reduced the manual effort between planning and wave execution.

The Real Outcome: A Roadmap, Not a Report

The clearest signal came before the pilot analysis was finished. Kierzkowski and team asked what this might look like across a broader network.

Time reclaimed across the four pilot distribution centers and five agents was used in an extrapolation study covering 39 distribution centers. The study projects 58,600 hours reclaimed annually for the entire network, or 1,400 hours/year per site.

This only represents the network opportunity modeled from the pilot evidence, a way to move from promising local results to a disciplined expansion case.

AI Support for the People Who Keep the Warehouse Moving

For L’Oréal, AI became operational when tested as a practical support for the people and processes already keeping the warehouse moving, not a separate layer of execution. It sets a baseline for evaluating where agent support might help across the broader network. As Clark said at the EMEA Exchange, “AI alone is not the answer; AI plus unification and a bit of human ingenuity make it all work.”

Manhattan ActiveAgents™ bring AI support into warehouse workflows, keeping teams in command of the operation, with agents assisting, not replacing, their judgment. Built into the Manhattan ActivePlatform™, they use real-time operational data and governed workflows to surface issues, guide resolution and take action, with autonomy configured by the customer and human review for higher-stakes decisions.

To learn more about ActiveAgents™, explore the solutions.