What AI Agents Are Actually Delivering in Warehouses

Will Drewes

Founder, Fern Strategy · 3 min read

The numbers are in

Warehousing and logistics is one of the few industries where AI outcomes are measurable almost immediately. Cycle times, pick rates, truck miles, inventory accuracy: these aren't soft metrics. You either hit them or you don't. So when large operators start posting results, it's worth reading carefully.

Amazon: $1.6 billion and 25% faster facilities

Amazon used machine learning to optimize logistics workflows across its transportation network and pulled $1.6 billion in costs out of the operation. That's not a projected figure; it's a reported reduction in transportation and logistics spend.

Separately, Amazon's AI-coordinated robotics fleet drove a 25% increase in overall facility efficiency, cut delivery times by 25%, and created 30% more value-added roles alongside the automation. That last number matters. The floor-level concern about headcount displacement is real, and Amazon's data suggests the outcome was role evolution rather than elimination. Whether that holds at smaller operators is an open question, but it's a data point worth having.

DHL: 30% operational efficiency gain from autonomous mobile robots

DHL deployed autonomous mobile robots across warehousing and distribution operations and reported a 30% jump in operational efficiency. The underlying mechanism is straightforward: robots handle repetitive horizontal movement, humans handle exceptions and judgment calls, and the coordination layer (AI) keeps the two from colliding. The efficiency gain comes from removing the dead time between those handoffs.

For a network the size of DHL's, 30% is a significant number. It also sets a benchmark that procurement teams at mid-market 3PLs are now being asked to explain.

Walmart: 90% inventory accuracy and 30 million fewer truck miles

Walmart's AI-driven demand forecasting reached 90% inventory accuracy and eliminated 30 million unnecessary truck miles. Those two outcomes are connected: better demand signals mean less safety stock, fewer emergency replenishment runs, and tighter routing. The truck-mile reduction has a cost dimension and an emissions dimension, which matters increasingly for customers with Scope 3 reporting obligations.

Inventory accuracy at 90% is also a compliance-adjacent number. In food, pharma, and regulated retail, inaccurate inventory creates traceability gaps. Getting that number right is a prerequisite for audit readiness, not just an operational nicety.

What this means for operations and supply chain leaders

These are large, well-resourced operators with significant engineering capacity. The outcomes are real, but the path to them involved years of instrumentation, data infrastructure, and integration work that most mid-market logistics operators haven't done yet.

The build-vs-buy question in this space is sharper than it looks. Amazon and Walmart built proprietary systems because they had to; the tooling didn't exist at scale when they started. That's less true now. The more pressing question for most operators is whether their existing data is clean enough and their workflows documented well enough to support an agent layer at all.

That's usually where a structured audit earns its keep: not deciding whether to automate, but finding out what's actually in the operation before committing to a build. The numbers above are the ceiling. The floor is a deployment that automates a broken process and locks in the dysfunction.