FrieslandCampina Distribution Center Capacity Study (2021)

Dairy manufacturer FrieslandCampina produces a wide range of dairy products, which are then stored in one of its distribution centers (DCs) for further distribution. This also applies to the recently built DC in Maasdam: a nearly fully automated warehouse where pallets of dairy products can be stored and picked.With the planned closure of the Rotterdam DC, the question arose whether the Maasdam DC has sufficient capacity to absorb the additional volume—or whether bottlenecks might occur.

FrieslandCampina’s question to SD&Co

“What is the current capacity of the DC, and in what ways can we increase this capacity?”

To answer this, we developed two simulation models together with FrieslandCampina’s project team, analyzed data from several IT systems, and conducted multiple practical tests on site.

During this project, we gained several insights that are valuable for any company looking to optimize the capacity of a distribution center or factory.

1. Determining the capacity (and bottlenecks) of a distribution center is not straightforward

Even in a highly automated and palletized DC, it proved challenging in practice to determine the actual capacity.For this project, three information sources were available:

1. The theoretical (design) capacity provided by the system supplier

2. The measured throughput over the past 10 days

3. The measured time between consecutive pallets over the past 10 days

However, these “three clocks” rarely show the same time. We observed that the theoretical capacity was structurally lower than the measured throughput (the maximum flow achieved in the last 10 days). Yet even the measured throughput often underestimated the true potential capacity.

This is because the speed of any process step depends not only on its own technical capacity but also on factors such as pallet inflow, downstream congestion, and operational coordination. As a result, both theoretical and measured capacities were often lower than the calculated capacity, which we derived based on the average time between consecutive pallets.

Conclusion:When you expect future bottlenecks in your distribution center or factory, it is essential to know the true capacity of your processes. Relying solely on theoretical or measured capacity risks underestimating actual performance.When discrepancies arise between theoretical capacity and realized throughput (per minute, quarter hour, hour, or day), a deeper investigation is recommended. This can involve calculating capacity directly or conducting targeted on-site tests — as discussed in point 3.

2. The higher the level of automation, the greater the human impact on system performance

It may sound paradoxical, but this phenomenon is often observed — and this project was no exception.At FrieslandCampina, the outbound loading process turned out to be the main bottleneck. In this process, outbound pallets are staged (pre-positioned) before being loaded onto trucks by the drivers.

Initially, this was not expected to be a limiting process: there were more than enough loading docks, loading speeds were high, and there was ample space for staging pallets.However, the data showed that peaks in staging and loading activity (periods of optimal outbound utilization) were consistently followed by troughs (periods of underutilization).

A dedicated simulation of the loading process revealed that this limited utilization stemmed from process coordination issues.On one hand, pallets were often staged truck by truck, meaning that only one of the two outbound lifts was active at a time—thus not using the full capacity.On the other hand, the team was cautious about scheduling too many truck arrivals per hour, to avoid driver waiting times. But when pallets are already staged and no truck is present, the process halts.This means that coordination with the transport planning can also become a limiting factor. Both aspects — truck arrivals (demand) and staging control (supply) — must be adjusted together to achieve optimal outbound performance.

Conclusion:When a process is largely automated, it’s worthwhile to also examine the human and operational control elements. The greater the automation, the more the system’s performance depends on human decision-making and coordination.In such cases, the biggest optimization potential may lie not in the technology, but in how people interact with it.

3. To extract maximum value from simulation, complementary testing and measurement are crucial

Finally, let’s reflect on the project process itself. This was a time-sensitive engagement — the team needed answers quickly. After defining and scoping the problem, we immediately began developing the first version of the simulation model.

This was an iterative process, where model development repeatedly led to new process questions (“How is this step controlled?”) and data analyses (“What is the capacity of this part?”).

After about a week, the first version of the model was ready. Together with the project team, we validated whether the model’s behavior matched reality. We also performed a sensitivity analysis to identify which variables had the strongest impact on key KPIs and to decide which ones needed more precise data.

That following weekend—during a quiet period—FrieslandCampina conducted targeted on-site tests (with GoPro recordings) to gain deeper insight into these variables and into process steps that were not yet accurately represented in the simulation.

These tests revealed that a potential bottleneck — the loading process — was not yet included in the model, prompting the development of a separate simulation for this process.In a follow-up workshop, the entire project group analyzed this new model and generated improvement ideas.Interestingly, most of these ideas came directly from FrieslandCampina employees themselves — after all, they know their processes best. The simulation helped test the potential impact of these ideas.

Finally, the ideas were tested both internally and with the system supplier to verify their feasibility and desirability for implementation.

Conclusion:The true value of simulation lies not in the model itself, but in working with the model.During development, we were forced to clearly define how each process actually functions:How are decisions made? Do we see that reflected in the data?While running scenario analyses, we jointly identified which variables really matter (and which don’t).The targeted tests greatly improved data accuracy and helped verify whether proposed improvements could realistically be implemented.

The close collaboration between FrieslandCampina and SD&Co was therefore essential to the project’s success.