From shortest path to best plan
The simplest routing question is how to get from A to B. Navigation apps answer it well. Freight and delivery operations face a harder question: given many stops, several vehicles, delivery windows and load limits, which vehicle should visit which stop, and in what order? Route optimization is the use of mathematical and computer methods to answer that.
The problem has a long history in mathematics and is known to be hard: as the number of stops grows, the number of possible orderings grows explosively. For even a few dozen stops, checking every combination is impractical. So practical systems use methods that find very good plans quickly, rather than guaranteeing the single best one.
What a routing method has to weigh
A route that is shortest by distance may not be the best, because real operations have constraints.
- Distance and drive time. These are not the same. A longer road with free-flowing traffic may be quicker than a shorter, congested one.
- Time windows. Many customers accept deliveries only during set hours. A route that ignores them risks missed deliveries and repeat trips.
- Vehicle capacity. Weight and volume limits restrict how many stops a vehicle can serve before it must return to reload.
- Vehicle types. Some stops need a lift gate, a refrigerated unit or a smaller vehicle that can enter narrow streets.
- Driver rules. Driver hours and required breaks limit how long a route can legally run.
- Service time. Unloading takes time that varies by stop, and underestimating it throws off the rest of the day.
- Pickups and deliveries. Some routes combine both, with the constraint that an item must be picked up before it is dropped off.
- Costs. Fuel, tolls, driver pay and vehicle costs can be built into the objective, so the plan minimizes total cost rather than just distance.
The method works by choosing an objective, such as lowest total cost or fewest vehicles, and searching for the plan that meets all the constraints while scoring best on it.
Common techniques, in plain terms
Without going deep into mathematics, a few families of approach are widely used.
Constructive methods. Build routes step by step, for example by repeatedly adding the nearest unvisited stop, or by starting with each stop on its own route and merging routes when it saves distance. These are quick but may leave gains on the table.
Improvement methods. Start with a workable plan and make small changes, such as swapping two stops or moving a stop to another vehicle, keeping changes that improve the result. This continues until no easy improvements remain.
Metaheuristics. More sophisticated search strategies that avoid getting stuck in a mediocre solution by sometimes accepting temporary worsening. They are common in commercial software.
Exact methods. For smaller problems, techniques exist that can prove a plan is the best possible. They tend to become too slow as the problem grows.
Dynamic re-optimization. Rather than planning once in the morning, the system revises routes during the day as new orders arrive, traffic changes or a delivery fails. This depends on live data from vehicles, such as that supplied by Internet of Things devices and mobile apps.
Learning-based estimates. Machine learning can improve inputs, such as predicting how long stops will take or how traffic typically behaves at certain times. This is related to broader supply chain AI work, though the routing step itself is still usually built on the methods above.
What the benefits can look like
Results depend heavily on the starting point. A fleet that already plans carefully may see modest gains, while one that routes by habit may see more. Reported benefits typically fall into these areas, though no particular figure should be assumed.
- Fewer miles and less fuel, because stops are grouped sensibly and backtracking is reduced. That in turn lowers emissions from the fleet, a point relevant to measuring freight emissions.
- Fewer missed deliveries, since time windows are part of the plan rather than an afterthought. Reliable windows help with service levels and customer satisfaction.
- Better use of vehicles and drivers, meaning fewer vehicles for the same work, or more stops handled by the existing fleet.
- More predictable days, because the plan accounts for breaks, loading and service times.
- Faster planning, since software generates a plan in minutes rather than a dispatcher spending hours.
- Better data for review. Comparing planned with actual routes shows where reality differs from assumptions.
These gains matter especially in last-mile delivery, where many small stops make routing cost a large part of the total.
A small hypothetical example
Imagine, purely as an illustration, a driver with six stops in a town. A simple rule, always going to the nearest remaining stop, might end the day far from the depot, forcing a long empty drive back. A better plan might visit a distant stop earlier and finish near the depot, accepting a slightly longer first leg for a much shorter return. If one customer also accepts deliveries only between 9 and 11 in the morning, the best plan has to put that stop early, even if it is not the nearest. Optimization means weighing these trade-offs together rather than one at a time.
Limits and cautions
Data quality. Plans depend on accurate addresses, map data, service times and vehicle information. Poor inputs produce plans that look neat but fail on the road.
Real life is messy. Accidents, road closures, weather, parking problems and absent customers all disrupt plans. Good systems allow for re-planning, and good operations leave some slack.
Driver knowledge. Experienced drivers often know details a system lacks, such as a loading dock that is hard to reach at certain hours. Plans that drivers can adjust, and feedback that improves the data, tend to work better than rigid ones.
Fairness and workload. A plan that minimizes cost may give some drivers far heavier days than others. Many organizations add balancing rules.
Over-optimizing. A very tight plan has no buffer. One delay then ripples through every later stop.
Fit with the rest of the system. Routing is more effective when connected to order, capacity and dispatch information. A transportation management system often includes routing, while some operations use specialist routing software.
Getting started sensibly
Organizations that adopt routing tools commonly begin by measuring the current state: miles driven, stops per route, on-time rate and time spent on planning. They then pilot the software on one depot or region, compare results with the old method, and gather driver feedback before widening use. This keeps expectations tied to evidence from their own operation, which is more reliable than general claims about what optimization can achieve.