Modeling Service in Network Optimization
With so much effort focused on cost minimization, modeling service can often be relegated to the backburner. However, the rise of eCommerce and our seemingly endless desire for ‘just in time’ consumption has ratcheted the importance of service across retail with ripple effects in other industries, as well.
Whether or not service is considered as part of network design can have implications on facility locations, mode mix, and inventory from an operational perspective, which may translate to revenue, operating expenses, capital expenditures, and working capital from a financial standpoint. If you aren’t currently incorporating service into your models, now may be the time.
Defining service in network optimization
For a model to properly answer business questions about service, it is important to first define service. This can be distilled into two questions:
- What is the right measurement for service?
- Should service be used as an objective or a constraint?
In network optimization, there are two basic measurements for service – distance and time. While distance is commonly used as a proxy for service, time should supersede distance, especially if there are multiple transport modes which serve the same origin-destination pairs with different transport times, for instance LTL versus FTL or Parcel Ground versus Parcel Next Day. Of course, if you are using transport time, you’ll want to ensure your model data is accurate. Day-definite transport times are easy to blanket populate, but you’ll want more specific transit for other modes and service levels. These can generally be acquired from your service providers or LLamasoft’s Data Services.
Also note that, because of the nature of optimization using averages and aggregates in terms of flow and time periods, some traditional concepts of service such as order fill rate or on-time percentage, and inventory availability are limited and are best supported through simulation or safety stock optimization.
The second question to answer is whether to treat service as an objective or a constraint. When considering whether to model service as an objective or a constraint, you should attempt to mirror reality and the underlying motivation behind the optimization study. For instance, are the reasons for modeling service ‘external’ (e.g. driven by customers and/or competitive pressures), or is the motivation more intrinsic and exploratory (e.g. understand the cost of improving service)? In the former case, service can be viewed as a ‘requirement’ or constraint; in the latter, service is an objective.
The decision to model service as an objective or as a constraint is key to eliciting the desired answers to your business questions. Be wise, as the model results may differ depending on your approach; however, do not fear, as multiple approaches can easily be tested against the same model.
Click for larger image. Figure 1 Different service objectives may provide different answers.
Service as a constraint
There are several key business questions you can answer when representing service as a constraint.
- Facility count and sourcing assignments (“Do I need to add a site or change my customer sourcing assignments to satisfy service requirements?”)
- Mode mix / mode substitution (“What is my mix of Ground vs. Next Day parcel if I change my service requirement?”)
- Both of the above simultaneously (“Is it more favorable to incur the cost of an additional facility vs. the cost of expedited freight?” The answer may even differ by product!)
In essence, when a service level is used as a hard constraint, it ‘forces’ the model to respond with solutions which satisfy that service level. For instance, these could be expedited transportation modes at a premium cost or the startup and operation of a more proximal facility.
Here are a few methods for modeling service as a constraint in Supply Chain Guru:
- Service Level
- Supply Chain Guru allows you to specify a service level for each Customer-Product (and even order-level) combination. Demand will not be able to be satisfied via any lane or mode which exceeds the specified service level.
- In Supply Chain Guru v8, the service level can be set using the Demand table via the Due Date field. In Supply Chain Guru X, in either the Customer Demand or Customer Orders table via the Service Level field. Additionally, in both versions in Network Optimization Options, set ‘Classify Demand by Due Date.’
- Last Mile Service Constraints
- In Supply Chain Guru X v1.3.1, the Greenfield Service Constraints feature was expanded for use beyond greenfield analysis. The new Last Mile Service Constraints table can be used in both Greenfield and Network Optimization.
- Last Mile Service Constraints allows you to create graduated service requirements such as 80% of demand within 2 days, 90% within 3 days, etc. This provides the model with some flexibility on which customers are served within each concentric service level, which is favorable if you have outliers in more remote areas.
- Service constraints can be defined based on combinations of Customer and Product, as well as Period, Site, and Mode. You can specify the constraint basis as quantity, weight, or cubic.
Both Service Level and Last Mile Service Constraint methods pair well with converting demand to a soft constraint via an objective of profit maximization or with the minimum demand quantity and penalty cost features in Supply Chain Guru X. Also, remember that you can use scenarios to explore solution sensitivity to varying service levels.
Click for larger image. Figure 2 When using service as a constraint, demand can only be satisfied via any lane or mode which achieves the specified service level.
Service as an objective
Setting the right service objective can be a tricky problem for senior management, especially in a retail environment where customers are expecting free and fast deliveries. Should you chase Amazon’s two day delivery at all cost? Should you promise a more moderate service level, with a focus on store experience? There are multiple examples of retailers that successfully navigated the transition from exclusively brick-and-mortar to both traditional and online channels. What is the cost of improving service and when is this cost ‘too steep?’
As a supply chain designer, not only should you design the most cost effective network with service constraints, but also provide insights into how service changes could affect cost. Maybe you are not considering startups for additional facilities, instead exploring the impacts of expedited transportation within existing or proposed network structure. In such cases, the visualization of service against cost is a tremendous tool for senior management team to make the right decision. Rather than presenting a single solution, you can illuminate a tradeoff curve to guide stakeholder decision-making to ward a result they are comfortable owning.
In reviewing the chart below, you can observe the relationship between total cost and total demand percentage satisfied within 2 days. It clearly indicates a non-linear relationship, and generally the most costly customers are the ones farthest away from your shipping locations (they are also frequently the smallest customers in terms of demand). Combining this kind of analytical insight with company’s overall business strategy is critical in making a design decision.
Click here for larger image. Figure 3 Example of a tradeoff curve created via Sequential Optimization with both cost and service objectives.
The method described below uses Sequential Optimization for modeling service as an objective and creating a tradeoff curve.
- Generate a cost optimal network by running the model with Cost Minimization.
- Generate a service optimal network by running the model with Sequential Optimization, and ensure:
- Weighted transportation time is the first priority objective. Depending on the goal, it might need to be adjusted to outbound only by adding a customer set to Destination column.
- Total cost is the second priority objective.
- This generates the service optimal network with the best cost.
- With the two anchor points generated, you will need to build scenarios to identify data points in between these extremes. The general approach is:
- Flip total cost as the first priority, and weighted transportation time as the second objective.
- Give a tolerance to the total cost objective. Tolerance is to sacrifice the first priority objective so that the solution can be better in terms of the second priority objective.
- The tolerance should be between 0% and the maximum percent. The maximum percent should be calculated as the percent difference between the total costs from the cost optimal network and the service optimal network.
- Use scenarios to adjust the tolerance to generate enough data points for the curve.
- Note: If the startup or closure of sites is considered in your model, you may annotate the data points on the curve based on any noteworthy changes in network structure.
Answering your businesses explicit or implicit questions about service can be approached with the same intentionality and rigor as more traditional cost and profit objectives. Moreover, decisions on service can be embedded into otherwise cost-focused models to identify more robust ‘real world’ solutions that impact the top and bottom line alike. The modeling part is easy; challenging your organization on their definition of service and underlying motivations is key to discovering meaningful insights.
Like this article? There is more to explore! If you are a current customer, please visit our support site for more practitioner insights. Future customers can request a demo here.