← Back to blog
Network optimization5 min readNovember 10, 2026

Our optimizer found up to $5 million a year in a well-designed network during initial simulation testing

We built a full national parcel network in simulation and gave it to PackOS Enterprise Optimizer with one instruction: lower the cost and protect service. Here is what it found.

By PackOS Team

Network planning tools are usually shown off on a demo built to flatter them. We wanted a real test for PackOS Enterprise Optimizer: a full national parcel network the software had never seen, simulated end to end, then handed to the optimizer with one instruction. Lower the cost, and protect service.

In initial simulation testing it found savings of up to $5 million a year in an already well-designed network, and on-time delivery went up.

The Optimizer is part of PackOS Enterprise, the AI supply chain optimization suite we are building. It is in development and not yet on sale. Here is what it does, and how that result came about.

What PackOS Enterprise Optimizer does

It is network planning software for carriers and large shippers, and it does three things.

It models your network as it runs today. Load your buildings, lanes, linehaul schedule, sort plan, volumes and rates. The Optimizer simulates a full week of freight moving through that network hour by hour: pickups arriving, sorts opening and closing, trailers filling and departing, every package measured against its service promise.

It lets you test a change before you make it. Close a lane, add a departure, shift volume, open a building. See what it does to cost, on-time delivery, trailer fill and sort capacity before a single truck moves.

It finds better networks for you. Give it a goal, such as lowest cost with service held, and it returns specific, reviewable changes. Each recommendation is verified by a full simulation across several versions of the week's demand, so you know how confident the answer is. Your planners stay in control: nothing changes until they review and accept it.

Every figure on the screen comes from a stored simulation, so every number can be traced and explained.

The test: a national carrier, built from the ground up

To put the Optimizer through its paces we generated a complete synthetic parcel carrier, built from public geography and the operating rules these networks actually follow.

  • 171 buildings: 150 package centers across 90 metro areas, 20 ground hubs and an air hub
  • 626 lanes, with a full linehaul schedule and a declared sort plan
  • 2,420 shippers tendering freight for first-mile pickup, including large accounts injecting directly at hubs
  • About 870,000 packages every weekday, roughly 4.6 million a week
  • Two service levels: ground and next-day air
  • Package size types, sort capacity by wave, dock doors and operating costs at every building

Volume follows population and distance, as it does in a real network: a few dense corridors, and thousands of thin origin-destination pairs that only make sense when consolidated through hubs. That is what makes network design hard, and it is the problem the Optimizer is built for.

A strong network to start with

Simulated as built, the network delivered 98.7% of packages on time, with ground at 98.9% and next-day air at 94.1%, at about $7.38 a package.

That is a well-run operation. Finding improvement in a network like this takes real analysis, which is the point of the test.

What the Optimizer found

In about eight minutes, the Optimizer returned a short list of decisions any network planner would recognise.

  • Re-home package centers to a closer hub. One center was being served through a hub 315 road miles away while another sat just 139 miles away. The Optimizer found these across the network and showed the saving for each.
  • Open direct lanes on dense corridors, bypassing a hub wherever the volume fills trailers on its own, complete with the sort-plan change to make it work.
  • Protect premium service. Every routing change was applied only where it kept next-day air on schedule.

The result, verified by full simulation and better on every demand scenario tested:

  • Up to $5 million saved a year, between $78,000 and $102,000 every week
  • Ground on-time up from 98.9% to 99.2%
  • Next-day air on-time up from 94.1% to 95.0%

Lower cost and better service, at the same time, from a network that was already performing well.

We ran the optimization three times to confirm it. The savings held every time.

Why it works

Three things set the Optimizer apart.

It is specific. You do not get a score or a heat map. You get named changes: this center to that hub, this lane opened, this rule updated, each with its own saving.

It is verified. Recommendations are confirmed by detailed simulation before you ever see them, across multiple demand scenarios, with the confidence stated in plain language.

It is yours to decide. Each recommendation is presented as one clear decision your team can accept or decline, with the full detail one click away.

See it on your network

This test ran on a synthetic network we built to be realistic. It is a simulation result, not a customer result. The result that matters is the one from your own data, and that is what the Optimizer is designed to produce: load your network, see it simulated, and find out what it can save.

If you run a parcel, freight or fulfillment network, we would love to show you. If you are an investor or a partner following the program, we would like to hear from you too. Join the early list below, or get in touch to see your network modelled.

Shipping strategy5 min read

How to Stop Undercharging for Shipping on Shopify

Flat-rate shipping estimates are almost always wrong. Here is why Shopify merchants consistently undercharge for shipping and a step-by-step method to fix it.

January 22, 2025

See it on your network.

PackOS Enterprise is in development. Tell us about your operation and we will reach out as early access opens, or book a demo to talk it through.