The good news?
Route optimization algorithms solve most of the problem. And you don't need a degree in maths to know how.
Here is the breakdown...
What's covered below:
- Why Routing Became An Operations Problem
- The 4 Route Optimization Algorithms That Matter
- The Constraints That Break Most Route Plans
- How To Roll It Out Without Causing Chaos
Why Routing Became An Operations Problem
Routing was once the domain of the dispatcher with the best memory and a laminated map. No longer.
Prices have gotten high enough that guessing is costly. Research from across the industry pegged the average cost to operate a truck at $2.336 per mile in 2025. That's the highest level ever. Traffic complicates things further. Congestion by itself added $108.8 billion in 2022 to industry operating expenses. During that time, it also consumed roughly 6.4 billion gallons of diesel fuel.
Now here is the part most teams miss...
No algorithm is smarter than the vehicle data you input into it. Height, axle weight, turning circle, payload capacity, trailer length – each and every one of those figures determines what roads a vehicle is legally and safely allowed to take. Fleets operating specialist vehicles tend to pull those details directly from the build sheet provided by their commercial trailer manufacturer. Builders like Dennison Trailers also spec each unit to the job, meaning the dimensions input into the routing system are precise, not approximately correct. Feed it one incorrect number and it'll plot a route that takes a 4.9 metre trailer through a 4.2 metre bridge gladly.
Garbage in. Diverted truck out.
That is why route optimization sits with operations, not IT.
The 4 Route Optimization Algorithms That Matter
Hundreds of flavors are available. But just about every commercial routing program is based on four fundamental concepts.
Understand these four and you will understand every software demo you ever sit through.
Dijkstra's Algorithm: The Shortest Path Finder
This is the grandfather of them all.
Dijkstra's algorithm solves one basic question: how do you find the least expensive path between two points? The algorithm begins at the origin and traverses out along every connected path, tallying up the total "cost" of each. Cost doesn't always mean distance. It can also represent time, fuel, tolls or driver hours.
The lowest cumulative score wins once all options have been scored. Most sat navs use a quicker variation known as A*, which guesses directionality so that it need not waste time considering roads going away from its goal.
Best for: single vehicle, single destination, point to point runs.
The Travelling Salesman Problem
Now put 20 drops on one truck.
The Travelling Salesman Problem asks, instead: what order should those stops go in to minimise total mileage travelled? Simple question. Very difficult to answer. Ten stops have over 180,000 possible route combinations. Twenty stops have a number of combinations which contains 18 digits.
No computer on earth checks all of those.
So instead solvers employ heuristics - smart tricks to get a very good solution quickly rather than an optimal solution eventually. Nearest neighbour, 2-opt swaps and simulated annealing are all examples of this. Start with a rough solution, then make it better and better until it plateaus.
Best for: multi-drop delivery rounds on a single vehicle.
Vehicle Routing Problem Solvers
The VRP is the TSP with real life constraints bolted onto it.
Several trucks. Various capacities. Time-windows for deliveries. Legal driving limits per driver. Multiple depots to work from. The solver determines which vehicle to assign to which stops, and in which order – simultaneously.
This is the algorithm doing the heavy lifting inside most fleet software.
Good solvers understand trade-offs instead of optimizing towards one objective. Three trucks each half-full can make all time-windows. Two trucks full will use less fuel but might make one delivery late. The solver assigns scores to both and chooses according to your preferences.
Best for: any fleet running several vehicles out of one or more depots.
Machine Learning And Live Re-Routing
This is the newest layer, and easily the most oversold.
Machine learning doesn't replace the solvers above. Instead it feeds them better numbers. It learns by working through historical telematics data that a certain retail park takes 40 minutes to unload at on a Friday afternoon, rather than the 15 minutes currently sitting in the plan.
Live traffic and weather feeds then adjust routes mid-shift.
The catch? Machine learning requires good, reliable data to learn from. Fleets who missed steps one through four are served boldly inaccurate predictions.
Best for: established fleets with 12 months or more of reliable telematics.
The Constraints That Break Most Route Plans
Here is where good software quietly goes bad...
An algorithm will optimise exactly what you choose to make it optimise. If it's not in the system, it doesn't exist. Empty running is the poster child example — approximately 16.7% of miles are driven empty industry-wide — and the vast majority of that is a planning issue, not a driving issue.
Constraints worth loading before go-live:
- Vehicle height, weight and axle limits
- Loading and unloading times, site by site
- Customer delivery windows
- Driver hours and mandatory break rules
- Trailer type restrictions (curtainsider, refrigerated, tipper, low loader)
Skip a couple and drivers will silently disregard the plan. When that occurs, trust is lost and your pricey software turns into costly shelfware.
How To Roll It Out Without Causing Chaos
Start small. One depot, one region, four weeks. Then build outwards.
Follow this order:
- Scrub the master data first. Car specs, customer addresses, job times. Nothing else matters until you get this step correct.
- Establish a baseline. Document current miles per drop, fuel per route and on-time delivery rate. If you can't measure it, you can't prove it.
- Run side by side. Map out courses both directions for a period of two weeks and honestly compare notes.
- Include drivers early. They know what yard won't happen at 7am. That intel should be in the system.
- Review weekly. Find out where drivers diverge from the plan, then correct constraint that is causing divergence.
Teams that skip the baseline always end up arguing about whether the software worked.
The Bottom Line
Route optimization is less about software and more about how accurately your operation is quantified.
The algorithms themselves are mature and proven:
- Dijkstra and A* handle point to point
- Travelling Salesman solvers handle the order of stops
- Vehicle Routing solvers handle the full fleet
- Machine learning sharpens the inputs over time
Give them realistic vehicle capacity, truthful service times and policies your drivers understand and the savings will be realized in fuel, overtime and on-time performance.
Feed them guesswork, and you have simply automated it.
