Here's a problem that keeps transportation leaders up at night: You've got hundreds of thousands of shipments to plan. Thousands of pickup and delivery points. Multiple transportation modes. Delivery windows that can't slip. Driver regulations that can't bend. And capacity constraints that change daily.

Oh, and you need to optimize all of this by Tuesday because rates are changing and your executive team wants to know if switching from air to intermodal will actually save money without killing service levels.

Traditional transportation planning software chokes under this strain. Either it oversimplifies the problem to get you an answer, or it takes so long to compute that the answer is irrelevant by the time you get it.

"By the time traditional tools give you an answer, the problem has already changed."

We recently worked with a company facing exactly this scenario. Here's how we approached it, and why it required rethinking transportation optimization from the ground up.

The Problem

The scope was massive:

• Hundreds of thousands of shipments across a multi-month horizon

• Thousands of origins and destinations

• Multiple transportation modes (truck, rail, air, intermodal)

• Multi-pick and multi-drop routing

• Driver duty rules and delivery time windows

• Millions of possible lane combinations

This isn't just "find the shortest route." It's consolidate shipments intelligently, identify cost-effective mode shifts, generate executable multi-leg routes, and do it all while respecting inventory availability, lead times, delivery windows, truck capacities, and driver constraints.

Most legacy transportation planning platforms crumble under this pressure. Why?

They rely on monolithic models that don't scale and generic solvers that weren't built for the conflicting priorities of modern transportation.

Translation, one giant optimization problem that takes forever to solve and can't adapt when constraints change.

They either oversimplify the problem (giving you a plan that looks good on paper but can't actually be executed) or take hours, sometimes days, to produce results. By the time you get an answer, the problem has already changed.

How We Cracked It

Here's the key insight - you don't need to solve the entire problem at once. You need to break it into intelligently grouped subproblems that reflect how your network actually works.

Simply put, shipments moving through the Northeast don't really interact with shipments moving through the Southwest until you get to consolidation hubs. So why force them into the same optimization run?

Spatial and Temporal Decomposition

We mirror the natural topology of the network, separating problems geographically and by time horizon, but retaining aggregate constraints so the local solutions still add up to a coherent global plan.

In other words, break the problem into geographic zones and time windows that make sense for your actual network, then solve them in parallel.

Lyric's Proprietary Parallel Solving

Instead of one solver grinding away, we run multiple optimization algorithms simultaneously across thousands of cloud nodes (some exact, some heuristic) and let them compete. The best answer wins.

Dynamic Load Balancing

As more data flows in, the system scales without choking.

This architecture lets us process millions of route possibilities quickly, which means planners can explore real trade-offs: Should I delay this shipment to fill a truck? Can I combine this transfer with another mode? Can I reduce partial loads by pooling shipments across DCs?

The result: We didn't just deliver an answer. We delivered a decision product. Not an episodic analysis, but an always-on system that reduced transportation costs by double digits, improved vehicle utilization, and cut planning cycle time compared to the incumbent solution.

What This Actually Looked Like

Here's one example that shows the power of this approach:

The company had been planning stock transfers between distribution centers independently. DC A would send a partial truckload to DC B on Monday, and DC C would send another partial load to DC B on Tuesday, using different carriers and routes.

Lyric's optimization spotted that if you delayed the shipment from DC A by 12 hours and consolidated it with the shipment from DC C, you could:

• Fill an entire truck (instead of two partial loads)

• Use a lower-cost carrier with better backhaul opportunities

• Still hit the delivery window at DC B

• Save 18% on that lane

Multiply that across thousands of shipments, and suddenly you're talking about millions in annual savings just from better consolidation logic.

Other wins included mode-shift recommendations from air to road based on real cost-service trade-offs, multi-pick and multi-drop routes that reduced empty miles, and dynamic routing optimized daily based on changing network conditions.

But here's what matters most, the system doesn't just solve for one best plan. It lets users simulate different business policies, run what-if scenarios, and push optimized recommendations directly into execution systems.

What's Next: GPU Acceleration Changes the Game

As powerful as this approach is, we're not done pushing the boundaries.

We're now collaborating with NVIDIA to integrate CuOpt, their GPU-accelerated solver for routing and fleet optimization. CuOpt brings massively parallel processing that slashes solve times from hours to seconds.

What this enables:

Interactive planning. A planner tweaks a constraint, maybe adds a stop or changes a delivery window, and sees optimized results update in real-time in seconds, not minutes.

Dynamic replanning. A truck breaks down or a customer changes their delivery window? The system recalculates routes on the fly.

Fine-grained last-mile optimization. For urban logistics and final-mile delivery, where every minute and every mile counts, GPU acceleration makes it feasible to optimize at a granularity that was previously impossible.

We're experimenting with interfaces where planners can literally drag a shipment node on a map and watch the optimization recalculate instantly. It's the difference between "submit a job and wait" and "explore possibilities in real-time."

We're not going to pretend this is simple. Transportation optimization at this scale is genuinely hard; there's a reason most software can't do it. We built Lyric specifically to crack problems like this. 

“With partners like NVIDIA pushing the frontier on GPU acceleration, we're making what seemed impossible five years ago routine today.”

Why This Matters

Here's the thing: legacy transportation and planning tools were built for a world that doesn't exist anymore. They assume your network is stable, your lanes are predictable, your capacity doesn't fluctuate wildly, and you can plan once and execute for weeks.

None of that is true today.

Port congestion. Driver shortages. Fuel price swings. Customer delivery windows that change by the hour. The problems you're solving on Monday look completely different by Wednesday.

“You don't need a system that gives you the answer. You need a system that helps you continuously explore better answers as conditions change.”

That requires domain-specific modeling that understands transportation constraints, custom solver architectures that can break problems apart intelligently, cloud-native scale that grows with your business, and GPU acceleration that makes real-time replanning feasible.

Lyric Studio combines all of this. The result isn't just faster solutions, it's better decisions, more often.

The Bottom Line

Transportation optimization at this scale isn't a theoretical exercise anymore. It's a business imperative.

Every percentage point you shave off transportation costs drops straight to the bottom line. Every hour you save in planning time means faster response to market changes. Every truck you fill instead of running half-empty is money you're not burning.

With Lyric's combination of intelligent decomposition, parallel solving, and GPU acceleration through NVIDIA CuOpt, what was once "too big to solve" is now not only solvable, it's strategic.

In a world where every load counts, every hour matters, and every constraint collides with another, Lyric delivers clarity, control, and competitive advantage.

"The era of static shipment plans is over. The future is fast, dynamic, and powered by intelligent decomposition."

Ready to rethink how you plan transportation? Let's talk.