The 20-Minute Decision That Now Takes Two

Every package delivered is a tiny math problem: which driver, which route, which cost. OneRail just made solving it 10 times faster.

Welcome Back,

OneRail has launched an AI-powered delivery platform that uses Nvidia technology to help retailers, wholesalers, and distributors decide how individual orders should be delivered, according to AI News. Called OmniSTAR, the system evaluates options including owned fleets, couriers, parcel carriers, and other delivery modes, then selects the lowest-cost option that meets the required service level.

The headline number is genuinely striking, OneRail says the system can reduce computation times by as much as 10 times. A calculation that previously took 20 minutes can be completed in under two minutes, while a calculation taking a week can be reduced to about two days. As OneRail's own executive put it to CNBC, "If you don't have the ability to make lightning-fast decisions, you're giving up margin. Last-mile fulfilment is expensive."

Today we look at how the platform actually combines Nvidia's optimization engine with OneRail's own delivery data, what happens when a driver breaks down or traffic changes mid-route, the real dollar results from customers already using it live, and why this specific speed jump matters more than it might first appear.

⚙️ Two Nvidia Tools, One Delivery Decision

The platform combines Nvidia's cuOpt decision optimisation engine and cuDF data processing software with OneRail's delivery pricing and performance data. Nvidia accelerated computing infrastructure is used to process the routing and delivery-mode calculations. Nvidia describes cuOpt as an open-source, GPU-accelerated optimisation library for vehicle routing and other mathematical optimisation problems, one that can account for vehicle costs, capacities, travel times, operating windows, starting locations, and other restrictions, with cost models based on distance, time, monetary cost, or a weighted combination of those measures.

Nvidia's cuDF handles the other half of the equation, GPU-accelerated tabular data processing, filtering, joining, and aggregating datasets, which is what lets OmniSTAR combine its own historical delivery data with live operational conditions fast enough to matter. It's worth being precise about what cuOpt actually does here too, Nvidia said cuOpt does not exhaustively test every possible route. Instead, the solver generates candidate solutions and iteratively improves them using GPU-accelerated heuristics to produce high-quality results within a set computation time.

That "high-quality, not exhaustive" distinction is a genuinely important engineering detail. The system isn't claiming to find the mathematically perfect answer every time, it's designed to find a very good answer fast enough to actually be usable inside a live operation, which is a meaningfully different, and more realistic, goal.

OmniSTAR evaluates owned fleets, couriers, and parcel carriers together before assigning each order.

🔄 A System With No Memory, By Design

Here's a genuinely important architectural detail buried in the technical description, because cuOpt is stateless, changes in operating conditions require the optimisation problem to be modelled and submitted again. Nvidia cites vehicle breakdowns, driver absences, road blockages, traffic, and new high-priority orders as examples of changes that can prompt this type of dynamic reoptimisation.

OneRail said OmniSTAR can rerun delivery scenarios as variables including fuel costs, weather, and shipping conditions change, and separately said its use of cuOpt allows it to evaluate more routing scenarios and recalculate routes faster than its previous approach. The 10x speed claim isn't really about doing something new, it's about doing the same fundamental optimization often enough to keep pace with a business that changes by the minute.

💰 From Prediction to a Real Decision

OmniSTAR is only one layer of a broader system. OneRail's machine-learning models separately estimate factors including service time, lateness risk, the probability of first-attempt delivery success, and expected price ranges, and those predictions feed into the optimisation systems that determine how an order should actually be executed.

That underlying dataset is genuinely the raw material making all of this possible, OneRail's models are built on millions of deliveries across a network the company says includes more than 12 million drivers and over 1,000 logistics partners, covering pricing and delivery performance across different transportation modes. OneRail said OmniSTAR can use that information to identify delivery rules that increase costs and assess how delivery choices affect item-level profitability, not just which route is fastest, but which delivery decision is actually the most profitable one.

Millions of historical deliveries feed OneRail's prediction models, which then inform each real-time delivery decision.

📊 The Real Dollar Results So Far

OmniSTAR isn't a lab prototype, it's already deployed with selected enterprise customers, and the reported results are genuinely substantial. At US Foods, OneRail said the system identified delivery configurations that were reducing margins, including low-margin products being transported long distances using higher-cost equipment. US Foods subsequently used those findings to adjust pricing and restructure some delivery patterns.

It's worth being fair and precise about that last figure, the customer was not identified, and the savings figure was provided by OneRail itself, not independently verified by CNBC or any third party. That's not a reason to dismiss it, but it is a reason to treat it as a company-reported claim rather than an audited result, the same standard worth applying to most vendor-reported ROI figures in enterprise technology coverage generally.

OneRail and Nvidia had worked on the project for three years before its launch, according to CNBC, including direct engagement with Nvidia's cuOpt engineering team on last-mile delivery and large-scale logistics optimisation, alongside OneRail's participation in the Nvidia Inception programme. That three-year development timeline is a useful reminder that a headline-grabbing 10x speed claim usually sits on top of a genuinely long, unglamorous engineering effort.

🚚 Part of a Bigger Push Into Same-Day Delivery

This launch doesn't exist in isolation, it builds on OneRail's existing infrastructure and partnerships. In March this year, FedEx launched FedEx SameDay Local in collaboration with OneRail, connecting customers to a national network of more than 1,000 delivery providers, a genuinely significant partnership that shows OneRail's technology already operating at real logistics scale before OmniSTAR's launch.

That last academic reference is worth noting, it independently corroborates that the underlying problem OmniSTAR is tackling, recalculating delivery decisions as real-world conditions shift, is a genuinely active, unsolved area of operations research, not a solved textbook problem that OneRail is simply packaging. Building a fast, workable system on top of a genuinely hard, still-evolving research area is a real technical achievement, whatever the exact scale of the dollar savings claims turns out to be.

OneRail's technology already underpins large-scale partnerships like FedEx SameDay Local, launched earlier this year.

🧠 AI Spotlight Analysis

This story is a genuinely good illustration of a broader pattern in enterprise AI that gets far less attention than flashy chatbot demos, unglamorous, backend optimization work that quietly compounds into real, measurable savings at scale. Nobody's going to be impressed watching a delivery routing calculation happen faster, but for a company shipping millions of packages, the difference between a 20-minute and a 2-minute decision genuinely determines whether AI-assisted optimization can run continuously throughout a live operating day, or only once at the start of it.

The stateless design of cuOpt is also a useful reminder that "AI" in a logistics context often means something quite different from a conversational chatbot, no memory, no ongoing context, just a very fast, very capable solver that gets re-asked the same fundamental question every time the world changes around it. That's a genuinely different, and in this case more appropriate, architecture than the large language models most people associate with the term AI today.

That line is really the whole business case in a sentence. Last-mile delivery is one of the most expensive parts of retail logistics precisely because it's fragmented across so many small, individually cheap-seeming decisions, which mode, which driver, which route. Making each of those decisions faster and slightly better, multiplied across millions of orders, is exactly where genuinely large aggregate savings tend to hide.

🔗 Sources and Further Reading

Thanks for reading AI Spotlight.

Our mission is simple: deliver clear, trustworthy, and actionable AI insights that help professionals stay ahead without the hype.