Orchestrating the Intelligent Warehouse: Software, Automation, and Human–Machine Collaboration
In the fourth epsiode of Warehouse Wizards episode, we sit down with Stefan Rusu, who leads Deloitte’s European Center of Excellence for Warehousing and Automation. We discuss what actually moves the needle in modern distribution centers. The constants remain: maximize customer experience and minimize cost. The variables are the tools and operating models we use to get there.
Discussion Overview
Efficiency is the invariant
Operational design starts from two constraints: service levels and cost. Any investment must raise one without harming the other. Metrics that enforce this discipline include order cycle time, dock-to-stock time, pick productivity, cost per order, and order accuracy.
"The focus on efficiency is always a constant. The focus on cost is always a constant."
Process before platforms before projects
Do not start with hardware. Map the current flow, remove waste, and tune the WMS first. Use Pareto analysis to target high-impact SKUs and routes. Use parameter changes and SOPs to remove friction. Automate only where a defined constraint remains, and tie the case to a measurable gap in throughput, space, or quality.
"Many companies do perceive automation as a bit of a silver bullet to all of the problems and all of the challenges on the floor."
Modular automation and the orchestration bottleneck
AGVs, AMRs, and task-specific robots improve scalability and redeployability. The new limitation is orchestration: deciding which resource executes which task when queues, locations, charging, and human work all interact. Practical steps:
- Unify task queues across fleets and zones.
- Normalize capabilities and safety envelopes per resource.
- Assign by marginal throughput gain, not by static rules.
- Recalculate frequently as inventory and congestion change.
AI already at work
Computer vision enables robot navigation, quality checks, and dimensioning. Machine learning supports dynamic slotting, travel minimization, and workload balancing. Retrieval-augmented tools condense SOPs and training into searchable guidance. Start with narrow AI that acts on trusted data and has a bounded decision scope.
Europe vs United States: constraints and talent
European networks optimize scarce space, which favors condensed storage and high-density systems. North American sites often solve for extreme throughput over distance. Both regions face a workforce shift. Digital-native employees expect guided tools, safer interfaces, and faster ramp-up. This is not a soft issue; it drives stability, quality, and overtime costs.
"I strongly I do believe that the challenge in logistics talent is a generational challenge."
UX “superchargers” that sit on top of WMS
Keep the WMS close to standard to reduce upgrade risk. Layer math-driven optimizers for slotting, routing, order grouping, and multi-scale packing. Treat them as apps that read operational data, run algorithms, and return decisions or instructions to mobile devices. The payoff is measurable in days, not months.
"we measured in our projects a move from five to six weeks to get an operator to the expected performance. And then once we deployed the supercharger, it got down to three to four me maximum 5 days."
The automation business case: three levers
Automation returns are volume driven and local to each site:
- Storage volume: compress the footprint to defer real estate or enable co-location.
- Throughput: raise hourly order lines and stabilize peak hour performance.
- Labor economics: substitute scarce or high-cost work with predictable machine time.
Model all three. In dense European locations, real estate often tips the case toward AS/RS even before labor savings.
Peak volatility: design realistically, flex intelligently
Size systems for credible peaks rather than outliers. For extreme surges, combine:
- Temporary labor guided by picking and pallet-building apps to prevent congestion.
- Robotics-as-a-Service to add short-term capacity where transport or buffer tasks are the constraint.
- Data checks on inbound mix and order profiles to pre-slot for the surge.
WMS “vanilla,” algorithms at the edge
Avoid deep WMS custom code. Use external services for optimization so you can upgrade core systems without rework. This aligns well with teams that own specific levers: a slotting product owner, a routing product owner, and a packaging optimization owner, each with KPIs and A/B tests.
Time-to-value and ROI targets
Run a subprocess pilot with a clear baseline. Lock the measurement window. If the tool is interactive guidance, expect floor impact within days once stable data feeds are in place. Target:
- Interfaced deployments: ROI within twelve months.
- Non-interfaced, batch-driven pilots: ROI within six months.
Digital twin vs digital shadow
A digital twin is a live, data-linked model that predicts outcomes and can push actions back to execution systems. Most current deployments operate as decision support with a human approving changes. Move toward closed loop gradually:
- Start with advisory recommendations.
- Autonomize low-risk micro-decisions.
- Expand autonomy as data quality and trust improve.
Outlook to 2030: toward a warehouse operating system
Large AS/RS will remain for density and rate. Modular robotics will scale. Competitive advantage will come from an orchestration layer that coordinates people, robots, and inventory in real time, fed by high-quality data and governed by clear safety and service rules. Human roles shift from supervising headcount to operating engineered systems, owning optimization products, and maintaining the model of the facility.