Bringing Simulation to the Forefront of Supply Chain Design
Recently the Optilogic team has been running into projects that require simulation or a combination of simulations and advanced solving techniques to properly address customer questions about their supply chains. Here are our thoughts on why simulation is becoming more common as a modeling tool than in years past, and how it can make a major impact on any business.
So why are we seeing more and more projects requiring simulation? To better understand why we’re seeing this shift, let’s look at a recent project we encountered. Part of the project centered around understanding options for inbound sourcing with specific questions such as:
What happens if the company sources product-to-need versus to-ocean container sizes?
Does it make sense to centralize stocking of slow movers to one location?
How would the above changes impact mode mix delivery to customers?
The power of simulation lies in the fact that you can model detailed processes, like ocean container loading and truck delivery to customers, resulting in an enhanced understanding of the impact on sourcing, production, inventory, and transportation processes.
By using aggregated demand and average costs, optimization is helpful to understand the best flows for the above questions, but it would not be able to answer the first question in detail. However, by using buy-to-need versus buy-to-container size business rules, a simulation would be able to tell you how many containers you would use with each of those scenarios, how full the containers would be, and what the cost per unit would be.
In addition, simulation would model the specific truck types used, their utilization, and the cost to deliver details, helping organizations dig deeper into their delivery strategies before they implement those solutions.
When we ran those scenarios described above during this project, the results were a bit counterintuitive. You would expect that buying-to-need would provide a better inventory picture, while transport costs would be a little more due to the frequent shipments. However, our simulation revealed almost double the amount of containers required for buying-to-need, with transport costs being high as a result. Ultimately, these transportation costs outweighed the inventory benefit seen with buying-to-need due to the comparative low value of the products to keep in inventory for this company.
When we looked at the strategy of centralizing slow movers at a hub Distribution Center (DC), the simulation revealed that while inventory for the slow movers was reduced, the fast-moving items increased. This is due to the spoke DCs needing to buy container minimums and thus filling the containers with more fast moving items. Traditional safety stock optimization, if used for this problem, could not have provided this same insight. This is because the aggregate inbound flows are modeled versus an actual transaction-level process (like inbound shipping container loading rules). Only simulation can capture this level of detailed insight.
Similar to the detailed inbound processes, simulation was used to model the specific modes used to ship out to customers. Using orders, pack processes, load consolidation strategies and mode options, simulation models current and future options to see the impact of mode mix changes.
Understanding Where Supply Chains Break
Climate-related disruptions, pandemics, strikes, geopolitical turmoil, and natural disasters have all been impacting how companies deliver products to their customers. The capability to model detailed disruptions down to the order/SKU level that simulation brings helps organizations understand risk and resiliency metrics, in addition to the traditional cost and service impacts. Simulation can also reveal where things get backed up.
Let’s dive into one project that examined how well a DC would perform given increases in demand and using its current inbound receiving capacity. The simulation revealed that service levels were impacted due to the back-up of inbound shipments behind the DC. This was ultimately because the inbound receiving processes couldn't keep up with the new demand, creating a bottleneck.
What's interesting about this recent increase in the demand for simulation projects is that, while we are answering strategic questions, we are also modeling the transaction and tactical-level processes typically associated with planning and execution technology. While much of the data is similar, the real-operations scenario building and testing to compare current and future models is what makes this complementary to execution, visibility, and planning solutions.
Combining Other Advanced Solvers with Simulation
While simulation seems to be coming to the forefront on recent projects, we are also combining it with optimization and machine learning algorithms to exploit the power of these technologies where they are needed. Optimization certainly has its place as it helps make structural decisions on the best flows and assignments to map suppliers to plants and customers to DCs. That framework is then simulated to take the analysis to the next level of detail and validate the decisions made in optimization. Incorporating transaction level detail and adding the variability that orders and transport processes have is the closest organizations can come to implementing a new design without actually implementing it.
One really cool project that is just getting started for us is a combination of simulation with a genetic algorithm to search among the huge number of possibilities to test inventory target strategies. While safety stock optimization was the typical approach in the past, the problem in this case intertwines transportation costs and processes, final assembly processes and locations, and capacity limitations for supply. As it sits, this is too complex to model using the simplifications needed for safety stock optimization. The simulation can capture all of those complexities and properly represent how the business works and the genetic algorithm acts as a machine learning component to find the best configuration of the simulation.
The genetic algorithm will search to find better strategies and kick off additional simulation runs based on previous run results to further test the strategies, giving the user the best candidate plans while understanding the tradeoffs among several metrics such as costs, inventory, service, and margin.
The Power of the Azure Cloud, Kubernetes, and Python
The other reason we may be seeing more simulation projects is the advent of open source simulation technology and the ability to use our Python language design studio, Atlas, to quickly and easily tailor a company's processes to mimic current and future models. A company can model their supply chain quickly on the cloud, without local hardware/software, and get results in days/weeks versus months/years. Compared to out-of-the-box solutions, Optilogic has complete freedom to build the model to the exact needs of the customer, and do so affordably.
The ability to run hundreds or even thousands of replications was always a sticking point with traditional simulation models that were run on a desktop or with limited access to solver engines. Because Optilogic built a cloud-based solution on Microsoft's Azure Cloud and incorporated their Kubernetes technology, a user can spin up as many solvers as they need.
The options to leverage simulation and create multiple alternatives are now limitless. Take your optimizations and simulations further — start a free trial to see what Optilogic is all about.
About the Author:
John is the Vice President of Business Development at Optilogic. Prior to joining, he was part of the leadership team at LLamasoft, Inc. helping the company go from 12 people to over 500 across 10 years, with roles in pre-sales, professional services, alliances, and country manager. John also worked in pre-sales and business development roles for several small supply chain companies with technologies in inventory optimization, demand planning and causal forecasting, network optimization, S&OP, and finite capacity scheduling.