The Good Brigade/Getty Images
- AI can support last-mile delivery by optimizing truck routes and predicting errors before they occur.
- AI-enhanced predictive analytics can also help to prevent package theft.
- This article is part of "How AI Is Changing Everything: Supply Chain," a series on innovations in logistics.
In the last mile, the part of the supply chain that involves transporting goods from a warehouse to a consumer's home, many things could go wrong. A package could end up at the wrong address, shipments could be late due to traffic, or a thunderstorm could damage a parcel left out in the rain.
"You're dealing with humans and the real world and trucks and traffic," said Fred Cook, the cofounder and chief technology officer of last-mile delivery company Veho.
In an area long dominated by carriers like UPS, FedEx, and the US Postal Service, Veho and many other software providers are looking to solve the challenges that pervade this notoriously complex and expensive part of the supply chain. They're using AI to design more efficient delivery routes, improve accuracy and the customer experience, and predict errors before they might happen.
Erik Mattson, a partner in consulting firm AlixPartners' Manufacturing and Operations practice, sees "a big opportunity for AI to help this industry catch up to other industries."
E-commerce sales continue to grow, reaching new highs of $300 billion in the last two quarters. This makes the last mile busier than ever and ripe for a technology disruption. A McKinsey report found that in the last decade, about $80 billion in venture capital went to logistics startups, with on-demand last-mile delivery platforms getting the greatest share of those funds.
AI from the road to the front door
Last-mile routes typically involve multiple stops and individual small packages — rather than one truck delivering pallets to a single warehouse — making this supply chain segment difficult to manage efficiently and expensive for the businesses involved. Last-mile delivery makes up an estimated 41% of all logistics costs in the supply chain, according to the Capgemini Research Institute.
One of the earlier applications of routing technology in the last mile was a machine-learning application that UPS launched in 2013 called ORION, or On-Road Integrated Optimization and Navigation. Four years ago, the parcel company rolled out an upgrade to ORION, which shortened routes by an average of two to four miles per driver and rerouted drivers based on changing conditions.
"Historic technologies would be static and run the night before," Mattson said. If orders changed or construction started, the tech wouldn't account for those changes.
Today's AI models, on the other hand, adjust in real time.
"Compared to pre-AI methods that relied on static routing rules or dispatcher intuition, our platform now responds dynamically to real-world conditions at scale," said Andrew Leone, the CEO and cofounder of Dispatch, a last-mile delivery platform.
Dispatch uses AI to plan routes based on factors such as traffic, delivery windows, estimated time per stop, and driver capacity. More efficient routes can lower fuel costs, improve density, and enable more deliveries in a day, increasing revenue for providers.
Amazon has been at the forefront of bringing AI into its last mile, said Jett McCandless, the founder and CEO of project44, a supply chain software platform. Last month, Amazon announced an initiative called Wellspring, which uses generative AI to analyze satellite images, apartment building layouts, street imagery, consumer instructions, and photos from past deliveries. It can recommend which parking spot or apartment building entrance a driver should use to drop off a shipment. In a test this past fall, the tech identified parking spots at 4 million home addresses.
Veho uses AI for quality assurance on its deliveries. In an ideal world, Cook said, an employee dedicated to quality assurance tasks would look at the geocode of where a parcel was left, examine the delivery photo, gather feedback from the driver, and determine if anything should change for future deliveries.
"It's totally infeasible to do that on millions of deliveries. But those are the types of use cases that we see, in the very near term, that AI is ideal for," Cook said.
Delivery data also allows last-mile providers to keep consumers informed. Deliveright, a last-mile delivery service, saw customer service calls drop by 80% due to real-time tracking and more accurate ETAs, according to Doug Ladden, Deliveright's CEO.
Veho said that its large language model, which it created in-house, answers 60% of customer and driver questions and has cut average response times from 2.5 minutes to 15 seconds.
Predicting and preventing package mishaps
Veho uses AI to pinpoint commonalities among mishaps that occurred during the logistics process, like the same warehouse associate handling multiple packages that resulted in errors, or a trucking company in the middle mile that had damaged items.
The company forecasts the likelihood of issues for specific routes or deliveries. Then it makes decisions based on the patterns, like moving packages to different facilities or increasing rates on a certain route, so drivers will be incentivized to pick them up earlier in the day.
"We've taken that a step further now to where we're trying to predict defects," Cook said.
Swiped packages are a big issue in last-mile deliveries, with 58 million parcels stolen from doorsteps last year amounting to $16 billion in losses, according to a USPS watchdog report.
UPS created an AI-based software, DeliveryDefense, that analyzes historic factors such as loss frequency and delivery attempts. The AI then spots areas that could be targets for porch pirates in the future.
McCandless said AI can predict high-risk areas and times of day, allowing companies to plan delivery schedules and routes accordingly to minimize the chance a package might be stolen.
"AI could play a key role in identifying patterns, helping to prevent theft before it happens," McCandless said.
American Dollar Exchange Rate