OneRail has introduced an AI-powered delivery optimisation platform designed to help retailers, wholesalers and distributors determine how individual orders should be delivered. The platform, called OmniSTAR, evaluates different fulfilment options and selects the lowest-cost approach that still meets the required service level.
Optimising every delivery decision
OmniSTAR can evaluate a range of delivery options, including company-owned fleets, couriers and parcel carriers. Rather than relying on a fixed delivery method, the system compares available choices based on cost and operational requirements.
The platform combines Nvidia’s cuOpt optimisation engine and cuDF data-processing software with OneRail’s own delivery pricing and performance data. Nvidia accelerated computing infrastructure is used to handle the routing and delivery calculations.
Faster calculations
One of the main advantages of the system is the speed at which it can perform optimisation calculations.
OneRail says OmniSTAR can reduce computation times by as much as 10 times. Calculations that previously required around 20 minutes can be completed in under two minutes, while calculations that once took approximately a week can be reduced to around two days.
The faster processing allows optimisation to take place closer to real time, giving delivery operators the ability to evaluate multiple fulfilment options before assigning an order.
For last-mile logistics, that speed can directly affect profitability because delivery is one of the most expensive parts of fulfilling an order.
Prediction meets optimisation
OneRail’s broader AI systems use machine learning to predict different aspects of a delivery before optimisation determines what to do with those predictions.
The company’s models estimate factors such as service time, the probability of delays, the likelihood of a successful first delivery attempt and expected pricing. Those predictions can then feed into optimisation systems that determine how an order should actually be fulfilled.
OmniSTAR takes this further by comparing different delivery modes and selecting the option that best balances cost and service requirements.
Nvidia’s role
At the centre of OmniSTAR’s optimisation capabilities is Nvidia cuOpt, an open-source, GPU-accelerated library designed for vehicle routing and other mathematical optimisation problems.
cuOpt can account for factors including vehicle capacity, operating costs, travel times, delivery windows and starting locations. Its optimisation models can also incorporate distance, time, monetary costs or combinations of these variables.
OmniSTAR uses cuOpt not only to optimise routes but also to help determine which fulfilment method should be used for an individual order.
Working with massive amounts of delivery data
OneRail combines Nvidia’s technology with its own delivery data and operational models.
The company’s dataset is based on millions of deliveries across a network that OneRail says includes more than 12 million drivers and over 1,000 logistics partners. The data includes pricing and performance information across different transportation modes.
This information allows OmniSTAR to identify delivery rules that could increase costs and assess how transportation decisions affect profitability at the individual-item level.
Recalculating when conditions change
Last-mile delivery is constantly changing. A driver can become unavailable, a vehicle can break down, traffic conditions can change or a high-priority order can suddenly enter the system.
Because cuOpt is stateless, changes to operating conditions require the optimisation problem to be modelled and submitted again. This allows the system to recalculate solutions when new information becomes available.
OneRail says OmniSTAR can rerun delivery scenarios as factors such as fuel costs, weather and shipping conditions change. The ability to rapidly recalculate these scenarios is central to the platform’s approach to dynamic logistics optimisation.
Moving into real-world operations
OmniSTAR is already being deployed with enterprise customers.
At US Foods, OneRail says the platform identified delivery configurations that were hurting margins. One example involved lower-margin products being transported long distances using more expensive equipment. US Foods subsequently adjusted pricing and changed some delivery patterns based on those findings.
OneRail also said that an unnamed large tyre distributor using the platform achieved $40 million in run-rate savings over three years. The customer was not identified, and the savings figure was provided by OneRail.
A broader logistics transformation
The OmniSTAR project reflects a broader shift toward using AI not simply to predict what might happen, but to make operational decisions based on constantly changing conditions.
OneRail and Nvidia reportedly worked on the project for three years before its launch, with the collaboration involving Nvidia’s cuOpt engineering team and a focus on last-mile delivery and large-scale logistics optimisation.
The company has also expanded its role in delivery infrastructure through its collaboration with FedEx, which launched FedEx SameDay Local using OneRail’s network of more than 1,000 delivery providers.
AI targets the cost of the last mile
For retailers and distributors, the challenge is not simply getting an order from a warehouse to a customer. They must determine which transportation method, route and operating configuration can deliver the order while protecting margins.
OmniSTAR is designed to automate that decision-making process by combining delivery data, predictive models and GPU-accelerated optimisation.
The result is a system capable of evaluating more possibilities in less time, potentially allowing logistics operators to respond to changing conditions while keeping delivery costs under control.
Source: https://www.artificialintelligence-news.com/news/ai-last-mile-delivery-optimisation/
