Better Models

Optimization models must run closer to execution and provide value to operations.

By Chris Horacek

A key benefit of e-commerce is that it allows sellers to more tightly integrate with their customers. Whether providing products to consumers or businesses, shipments have become smaller and more frequent and are now delivered to more locations at specific times based on customer demand.

While this shift in commercial exchange has tightened the seller-buyer relationship, it has created tremendous uncertainty for the logistics providers charged with operationalizing the physical connection between them through the transfer of goods.

Prior to the widespread adoption of e-commerce, logistics providers had a consistent view of demand across their networks over a wider time frame. In most cases, planning for that demand was a matter of taking the previous year’s volume, adding a growth factor, and accounting for localized changes such as population growth or the opening of a new demand node.

As we have seen during holiday peaks the last several years, the demand for logistics capacity is less predictable and can change dramatically across a network quickly. The impact of these changes in demand result in a gap reduction between planning and execution – compressed from months and weeks to days, hours, and, in some cases, minutes.

Recently, optimization models have been successfully used in the logistics planning process to support major operational areas like local pick-up and delivery, line haul and sortation/distribution. Many technically savvy operators have learned to leverage these models to improve performance both financially and operationally.

However, with the compression of the planning cycle, the game has changed and optimization models must run closer to execution to be relevant and provide value to the operation.

Compressing the Planning Cycle – Making It Work

Providing real-time, actionable insight to the operation is accomplished through the development of optimization-enabled analytics applications. Deployment of these applications into a working operation relies on streaming data to continuously learn and improve predictive and prescriptive capabilities.

Historically, “fresh” data availability was the major impediment to developing real-time analytics capabilities. However, the broad deployment of sensors and Industrial Internet of things (IIoT) services to ensure essential connectivity has largely reduced this barrier.

Applications capable of feeding large amounts of streaming data into in-process advanced analytics (such as stochastic optimization) are now a reality. They provide actionable information to a human or a machine that can be leveraged to make logistics operations more efficient and effective. The following are four use cases where in-process optimization has added value by reducing operating costs and improving service.

Dynamic Service Management

With more exacting customer requirements, real-time tracking and environmental monitoring becomes more intense. Variables such as in-transit location, potential service constraints, temperature and motion are monitored to ensure shipments are received on-time and in the expected condition.

In most cases, service providers can identify and remedy potential service problems while a shipment is in transit, eliminating potential customer service issues. This is especially important for shipments involving mission critical items.

Transshipment Hub and Distribution Center Operations

As shipping demand becomes more variable with more pronounced seasonal peaks, operating transshipment hubs and distribution centers at full capacity becomes increasingly important. A machine learning variant called reinforcement learning has been used to optimize capacity utilization through the better allocation of resources.

Variables and constraints such as service level requirements, available staffing, and physical capacity can be monitored and managed dynamically over time, enabling the operational decisions needed to maintain an optimum balance of assets, people and shipments as the work day progresses.

Transshipment hubs and distribution centers are complex systems consisting of the unpredictable arrival of shipments, finite queuing and storage, and throughput capacities subject to service level constraints for multiple product shipping types. Human staffing issues, including employee skills, task assignment and constraints on worker availability, work type, worker utilization and safety further complicate the system.

Organizations need to understand how to operate efficiently, without losing flexibility to handle the inevitable degrees of random disruption – a balance between lower cost efficiency-driven systems and higher service level quality-driven regimes. The most ideal approach is to leverage technologies, such as machine learning and real-time data analytics, to automatically find a balance between maintaining service level quality while still accounting for randomness.

A hybrid model-based optimization analytics platform that incorporates state-of-the-art reinforcement learning enables the system to be as close as possible to this ideal operating point. Such a system can break down the silo between legacy planning systems with longer-term horizons and the needs of real-time operational decision-making.

Line Haul Operations

Variability in moving shipments from handling node to handling node also has a major impact on cost and service. While this impacts all transportation modes, it is most acute for air shippers due to compressed delivery times, high operating costs and constraints on assets and infrastructure.

The ability to closely align capacity with demand across time and space is critical, as over capacity can be the difference between profit and loss on a move, while under capacity can cause transit delays, lost revenue and exposure to guaranteed service claims.

Equipment Health Monitoring

Maintenance is the unsung hero of the logistics cycle. After all, if assets are not moving shipments, they become liabilities. Downed equipment leads to unmet customer expectations and unplanned costs that negatively impact the bottom line.

Real-time monitoring is commonly used to ensure assets are available and operating optimally. Deployment of sensors and embedded analytics has played a key role in evolving maintenance from a reactive process to a more predictive and prescriptive one.

These are just a few examples of how optimizing key components of the logistics cycle drives business value. While these examples focused on service and cost factors, optimizing analytics can impact other areas of business such as crew scheduling, employee satisfaction and workplace safety.

Optimization capabilities will continue to evolve and become even more closely intertwined with in-process operations. Build out of the IIoT and deployment of sensors with two-way communication capabilities will further facilitate bringing advanced analytics into the real-time operational theater.

As pressure from the public for more dynamic service capabilities intensifies, logistics service providers need to be more creative in how they select, deploy and manage operational assets. The conventional thinking of building in capacity buffers is a diminishing option as demands on capacity become less predictable and much more pronounced across time and space. The importance of bringing analytics and planning closer to operations will only increase in an ever-changing and highly dynamic logistics market.

Horacek headshotChris Horacek is a strategic account executive at SpaceTime Insight, with over 30 years of experience applying technology and near-technology solutions in the transportation, retail and process manufacturing industries. At SpaceTime Insight, he drives customer engagement operations and provides expertise to key customers. Prior to working for SpaceTime Insight, he worked for various companies, including BravoSolution, Elemica and FedEx.