The Real Math Behind Multi-Vehicle Route Planning

If you manage a fleet of more than a handful of vehicles, you've probably had this thought: "There has to be a smarter way to plan all these routes." You're right.

And the math behind it is fascinating, even if you never plan to solve an equation yourself.

Multi-vehicle route planning isn't just about plugging addresses into a map app and crossing your fingers.

And beneath the hood, it's a deeply complex optimisation problem that has occupied mathematicians and computer scientists for decades.

Understanding the basics of how it works can help you make better decisions about the tools you invest in, and ultimately, the service your customers experience.

Let's break down what's really happening when route optimisation software crunches the numbers for your fleet.

What Is the Vehicle Routing Problem (VRP)?

In formal logistics circles, multi-vehicle route planning is known as the Vehicle Routing Problem, or VRP.

It is a massive combinatorial optimisation challenge that seeks the perfect set of routes for a fleet serving a specific group of customers.

The objective is to drive down total costs by slashing distance, travel time, and fuel consumption while abiding by the heavy constraints of vehicle capacity, delivery windows, and shifting traffic conditions.

The real catch is that mathematicians classify VRP as "NP-hard." In plain terms, the number of possible route combinations does not just grow; it explodes.

For a modest fleet of 10 vehicles and 100 stops, the potential permutations are so astronomically large that checking every option to find the absolute "perfect" solution would take longer than the age of the universe, even on today's fastest hardware.

This is why modern route optimisation platforms have stopped chasing theoretical perfection.

Instead, they utilise state-of-the-art algorithms to deliver near-best solutions, which are routes within a fraction of a percent of the absolute best, in seconds rather than centuries.

In 2026, this speed is the difference between a profitable shift and an operational standstill.

How Does the Math Actually Work?

At its core, VRP models your delivery network as a graph. It is helpful to think of it this way: nodes represent your depots and every customer location, while edges join each pair of nodes with a specific cost, typically measured in travel time or distance.

The optimisation engine must decide which vehicle visits which customers, and in what order, to ensure the total cost across all routes is as low as possible.

The system does this while strictly enforcing the constraints you deal with during daily operations.

Every customer is visited exactly once to avoid double-ups or missed deliveries, and vehicle capacity limits are respected so no van is overloaded by weight or volume.

Time windows are honoured to ensure customers expecting a morning delivery are not left waiting until the afternoon, and every vehicle begins and ends its journey at the depot to complete a full loop.

For multi-vehicle scenarios, modern algorithms regularly partition the delivery area into regions of dominance where vehicles act independently.

Some state-of-the-art systems even use auction-like logic to assign stops, where vehicles essentially "bid" on tasks assigned on their proximity and current load.

This balances the workload so no single driver is overwhelmed while another sits idle, turning a confused pile of orders into a streamlined, high-performance operation.

The Algorithms That Deliver Route Optimization Software

Since brute-force calculation is off the table, route optimisation relies on layers of increasingly clever algorithms.

Think of it as a three-stage process: build a decent plan fast, then improve it, then improve it again.

Stage 1: Constructive Heuristics (Getting a Starting Point)

These are the quick-and-dirty methods that produce a workable route in seconds. They're not elegant, but they give the system something to improve on:

• Nearest Neighbour: Start at the depot, go to the closest unvisited stop, repeat. Simple and fast, but it tends to produce routes that zigzag inefficiently.
• Clarke-Wright Savings: Start with individual depot-to-customer routes (one vehicle per customer, which is obviously terrible), then merge pairs of routes that save the most distance when combined. This is smarter and tends to produce better starting solutions.

On their own, these heuristics typically produce routes that are 20 to 30% longer than optimal. Useful as a starting point, but you wouldn't want to stop here.

Stage 2: Local Search (Improving the Plan)

Once you have a feasible set of routes, local search methods chip away at inefficiencies using making small, targeted changes:

• 2-opt and 3-opt: These techniques look for places where route segments cross over each other and reverse or rearrange them to eliminate the overlap. If your route goes A → B → C → D, a 2-opt move might swap it to A → C → B → D if that's shorter.
• Swap and Relocation: Move a stop from one vehicle's route to another, or shift stops around within a single route, to find a better overall balance.
• Large Neighbourhood Search (LNS): This is a more aggressive approach. It deliberately destroys a chunk of the plan, say 30% of stops, and rebuilds that section from scratch using heuristics. This helps the system escape dead ends where small tweaks can't find further improvements.

Local search methods can deliver 15 to 25% improvement over the initial heuristic solution. That's significant when you're talking about fuel costs and driver hours across a fleet.

Stage 3: Metaheuristics (The Heavy Hitters)

For sizable operations with hundreds or thousands of stops, you need algorithms that can explore a broad solution space without getting trapped.

This is where metaheuristics come in, and they can approximate solutions within 0.5 to 1% of the theoretical optimum:

• Genetic Algorithms (GA): Inspired by evolution, these preserve a "population" of route plans. The best plans are combined (crossover) and randomly tweaked (mutation) over many generations, gradually evolving toward better solutions.
• Tabu Search: This method keeps a list of recent moves and forbids the algorithm from undoing them. That prevents cycling back to solutions it has already explored and forces it to keep searching new territory.
• Simulated Annealing: Borrowed from metallurgy, this approach occasionally accepts worse solutions on purpose, especially early in the process. This randomness helps it explore broadly before settling into a refined answer.
• Adaptive Large Neighbourhood Search (ALNS): A sophisticated version of LNS that learns which destroy-and-repair strategies work best for a given problem and adapts its method accordingly.

Combined, the full pipeline from heuristic initialisation through metaheuristic refinement can deliver 20 to 40% total savings in mileage and time compared to manually planned routes.

What Happens When Things Change Mid-Route?

Everything we've discussed so far assumes you know all your stops, constraints, and travel times before the day begins. In the real world, that's rarely true. Traffic jams, last-minute orders, driver delays, and cancellations happen constantly.

This is the domain of the Responsive Vehicle Routing Problem (DVRP), and it's where route planning gets really interesting.

Modern systems handle dynamic conditions via several strategies:

• Reoptimisation Triggers: Rather than recalculating everything constantly, the system watches for threshold breaches. If a delay pushes an ETA beyond an acceptable window, it triggers a recalculation, but only for the affected downstream legs, not the entire fleet. This keeps computation manageable.
• Region-Based Policies: The delivery area is partitioned into zones, and each vehicle is assigned to its zone's centre point. Routes are optimised locally within each zone. Research shows this approach is provably optimal in lighter delivery loads.
• Predictive Layers: The best systems don't just react to problems. They anticipate them. By ingesting historical travel time data, weather forecasts, and even driver behaviour patterns, machine learning models have the ability to refine estimated travel costs before disruptions actually occur.

The most effective platforms use a composite approach: heuristics build the initial plan, metaheuristics refine it before dispatch, and AI-based reoptimisation adjusts routes live throughout the day.

Fleets using this kind of layered system are reporting 15 to 40% mileage savings and 95%+ on-time delivery rates.

Handling the Messy Reality of Mixed Fleets

Real fleets are almost never uniform.

You might be juggling vans with different payload capacities, trucks with strict height restrictions, or electric vehicles that require dedicated charging stops.

In 2026, the mathematics of logistics is required to account for every one of these variables to maintain a viable operation.

Smart solvers now assign tasks allocated on individual vehicle profiles, matching capacity and type to the exact demands of each stop.

For EV fleets, this means the algorithm adds charging station nodes to the graph and factors in green windows, which are the specific times when charging is available and the vehicle can afford to stop without blowing the schedule.

Furthermore, instead of only chasing the shortest distance, modern objectives focus on workload balancing to minimise the maximum time any single driver is on the road.

This avoids the resentment caused by lopsided rosters and is a critical lever for driver retention in a market where turnover is increasingly expensive.

These are not theoretical additions; they reflect the daily operational reality of managers who must balance vehicle variety, driver availability, and customer expectations to stay profitable.

Frequently Asked Questions

Why can't route optimisation software find the "perfect" route?

The Vehicle Routing Problem is NP-hard, meaning the number of possible solutions increases exponentially with each added stop or vehicle. For any real-world fleet, checking every possible combination is computationally impossible. Instead, modern software uses innovative algorithms to find solutions that are within 0.5 to 1% of the theoretical best, which is more than good enough to deliver major cost savings.

How much can route optimisation actually save?

Results vary depending on fleet size, delivery density, and current planning methods, but fleets commonly see 15 to 40% reductions in total mileage and pronounced improvements in on-time delivery performance. Even moving from basic heuristics to layered metaheuristic approaches can unlock 20% or more in additional savings.

What's the difference between static and dynamic route planning?

Static route planning calculates all routes before the day begins and doesn't adjust them. Dynamic route planning continuously monitors current conditions like traffic, delays, and new orders, and recalculates routes on the fly. For fleets confronting unpredictable conditions, dynamic planning is necessary for maintaining service levels.

Does route optimisation work for mixed or electric vehicle fleets?

Yes. Modern solvers can handle heterogeneous fleets by factoring in each vehicle's capacity, speed, and type. For electric vehicles, the algorithm includes charging station locations and battery restrictions to guarantee routes remain feasible without manual workarounds.

Ready to See What Smarter Routing Can Do for Your Fleet?

Understanding the math is one thing. Seeing it work on your actual routes is another. If you're curious about where your current operations stand and where the biggest opportunities for improvement might be, we can help.

Or, if you'd opt for a hands-on conversation: book a free operations review with our logistics experts and uncover quick wins for your fleet.