Technological advancements have established the foundation for completing last-mile deliveries in a challenging environment in a timely, safe, and cost-effective manner. Selections are more diverse and deliveries are faster, yet costs are on the rise. As time frames tighten—from days to hours and hours to minutes—customer sensitivity to late deliveries increases, especially for emerging categories like grocery and pharmacy.
Fortunately, advances in machine learning (ML), deep learning, and generative artificial intelligence (AI) capabilities now allow the execution of last-mile operations with utmost efficiency for all types of deliveries, balancing various constraints such as transporter safety, customer promise, and cost. No longer limited to simple sequencing, algorithms now account for parking, building access, driver experience, and other variables that may impact delivery quality and timeliness. This ensures long-term sustainability and unlocks a path to meet carbon-neutral targets for the delivery fleet.
What is a last-mile delivery?
Last-mile delivery describes the final step in the delivery process in which customers receive their purchased items. Despite the implication that it is only one step in the process, this final stage disproportionately impacts the overall success of the entire delivery process and, subsequently, on customer experience and satisfaction. The “last mile” accounts for more than 53 percent of the overall shipping cost for an average delivery, and inefficiencies within the process are costly to end-to-end order fulfillment.
The delivery’s final step emphasizes numerous areas of technological advancement. Significant research goes into each of the following:
- Supply chain and fulfillment operations: Companies are emphasizing the storage of high-demand products as close as possible to the customer for faster deliveries while predicting customer demand for each stock keeping unit (SKU) and ensuring sufficient inventory. This balances fixed costs and incremental growth in customer orders with differentiated fast deliveries.
- Last-mile routing: The delivery sequence is only one consideration. With last-mile, drivers complete numerous deliveries within a tight radius on a single route. Parking, traffic, and other neighborhood-specific conditions are essential to estimating delivery times and optimizing the ideal route. In densely populated areas, gaining access to a building or community is also part of the equation.
- Driver and customer specificity. ML allows delivery route optimization based on a driver’s profile, which includes familiarity with a given area, experience with different job challenges, and physical abilities. The more data supplied in the algorithm’s creation—both driver-specific and external factors such as traffic patterns on certain streets, time of day, weather, and city regulations—the more optimal the routing process is.
Machine learning for customer satisfaction and transporter safety
Amongst countless changes and advances in the delivery industry, vast selections and fast, reliable deliveries remain critical. The most important consideration is that the delivery arrives within the promised time frame. The challenge is the ever-tightening windows, from several days to the next day, and with food and other perishables, timeframes are as tight as hours or even minutes.
As delivery windows shorten, customer sensitivity to lateness increases. A delivery promised within an hour that takes 90 minutes is more likely to agitate a customer than a promised two-day delivery that, in fact, takes three days to complete. Some types of deliveries are especially time-sensitive. For example, more than half (52 percent) of customers will request or demand a refund for a late food delivery.
Late deliveries are the easiest way to lose business, but ML provides companies an advantage by using data to foresee traffic jams or other obstacles in a specific area. Learning models can predict future outcomes using information from previous deliveries and recommend more optimal routes if necessary. This same technology allows for more precise delivery time predictions, which is a crucial consideration when minutes matter.
Safety is the biggest concern for the transporters completing the deliveries. ML and optimization techniques can help in various ways, starting with the ability to complete quality deliveries, such as proper packaging for ease of transport, optimal drop-off points to prevent package theft, and route sequencing decisions that allow transporters to deliver to rural areas (no lights in the dark) and high-crime neighborhoods in each city when it is safe to do so.
Regarding personal safety and convenience, the focus changes to sequencing deliveries in the direction of travel. In highly congested areas, transporters hope to avoid crossing the street repeatedly. Cross-street deliveries are planned only in areas or streets that are easy to traverse, such as residential developments or sparsely populated areas.
Additional parameters include the specific transporter’s schedule. Some transporters work full-time, while others deliver packages in their spare time or on a more sporadic schedule. Routing algorithms can consider these aspects to lessen overall exertion on longer routes and potentially maximize the intensity of a shorter route by having the transporter park in a centralized location and deliver multiple parcels by walking. Route creation includes dynamic factors, including temporary road closures, sporting events that draw large crowds to an area, weather intensity, and even specific transporter preferences.
Measuring the success of routing algorithms
The key to constructing appropriate metrics is working backward from the customer’s perspective to capture specific pain points. For example, for last-mile routes, one can consider the predicted time for delivery versus the reality in transit and during servicing at individual buildings. Other factors that can be evaluated include route sequence adherence by drivers, mileage, or other variables.
Operations-focused metrics determine the utilization of the fleet for optimal deliveries while maximizing efficiency. For example, if a full-time transporter goes out on a route with 20 packages in the back of a van, what percentage of the van’s capacity is utilized? Were the packages themselves fully optimized? If a transporter goes on a two-hour shift, does that delivery route require a van?
Metrics such as defects per million orders (DPMO) measure the quality of all deliveries. This is simple to determine: How many complaints do customers lodge for every million packages delivered?
The future of routing algorithms
The broadest trends in the future of deliveries are due to an industry-wide move toward carbon-neutral goals. While transporters typically make deliveries via van, local regulations and parking difficulties in major European cities and select metro areas in the United States make such deliveries inefficient. Companies are moving towards a multi-channel delivery approach—transporters using e-bikes or walking to complete deliveries with limited capacity relative to vans, but with a more interconnected hub-and-spoke model. Obviously, this method is only possible for smaller packages, but it is faster, more efficient, and, above all, greener.
Similarly, companies such as Walmart and UPS are increasing the use of drone deliveries. Drones are even used to deliver fast food and medical supplies. From an algorithmic standpoint, challenges include determining the appropriate sizes and weights of packages for such deliveries, optimizing the routing of deliveries, and determining the safest places to leave packages.
Large companies emphasize increased fleet electrification, with charging infrastructure presenting the greatest challenge to realizing the goal in a timely manner. Organizations are often at the mercy of government regulations, but preparations for the future can include diversification of shipping options and flexibility in delivery options.
Combining innovation and optimization in last-mile deliveries
In the past five years, a wide range of technological developments have improved the operations and efficiency of last-mile deliveries. These innovations have enhanced customer expectations for speedier, more comprehensive, and higher-quality service.
Despite the changes, one principle remains: timely, thorough deliveries are crucial to customer retention. Fortunately, the capabilities of ML and other emerging technologies continue to expand, enabling organizations to adjust their algorithms for safety, fuel conservation, driver preferences, and more. The available data creates a favorable environment for efficient deliveries today, promising further advances moving forward.
About the Author
Hrishikesh Paranjape is a senior product manager with broad-based experience across the e-commerce, real estate, customer service, and financial industries. His work has transformed user experiences and optimized delivery routes for more than 100,000 drivers delivering packages. Hrishikesh graduated with a bachelor’s degree from the Indian Institute of Technology. He earned a Master of Science degree from the University of Oxford and an MBA from the MIT Sloan School of Management. For more information, contact hrishi@alum.mit.edu.
Disclaimer: The author is completely responsible for the content of this article. The opinions expressed are their own and do not represent IEEE’s position nor that of the Computer Society nor its Leadership.
