What Route Optimization Actually Does (and What It Cannot Fix)

Route optimization is the practice of deciding which vehicle visits which stops, in what order, at what time, subject to constraints that cannot all be satisfied at once. That last clause is the whole subject. Finding the shortest path between two points is navigation and your phone does it for free. Deciding how to cover 180 stops with nine vans, honouring delivery windows, vehicle capacities, driver shift limits and the fact that one customer's loading bay is only open between 10am and noon, is a different kind of problem.

The distinction matters commercially because it decides what you are buying. If your stops are fixed, sequential and few, you need navigation. If your constraints conflict, you need something that chooses which constraint to give up, and you need to know which one it gives up first.

Route optimization, route planning and navigation are three things

These get used interchangeably in vendor copy, and the difference is worth holding onto.

A planner who assigns stops to vans by postcode and then uses a tool to sequence each van's list is doing route planning with navigation underneath. The assignment step — the one that decides van three takes the north run — is the part optimization replaces, and it is the part where most of the available gain sits, because a badly assigned route cannot be rescued by good sequencing.

The constraints that define the problem

Every optimizer accepts some subset of these. The useful question in an evaluation is never "does it support time windows" — they all claim to — but what it does when two constraints cannot both hold.

• Time windows: the interval a stop will accept service, sometimes two per stop (a morning window and an afternoon fallback)
• Service duration: how long the vehicle is stationary at the stop, which is distinct from drive time and is the input fleets most often get wrong
• Vehicle capacity: weight, volume, pallet positions, or a count of units, often more than one at once
• Vehicle compatibility: height and weight limits on the roads to a stop, refrigeration, tail lift, access restrictions
• Driver skills and certifications: who is licensed or trained for a particular load or site
• Shift limits: start time, maximum duration, break requirements — and for regulated drivers, the federal hours-of-service ceilings
• Depot structure: one start point, multiple start points, or start-from-home
• Horizon: one day, or a week where a stop may move between days
• Precedence: a pickup that must precede its delivery, which turns one order into two linked events

Those last two are where products genuinely diverge. Single-day, single-depot optimization is close to commodity. Multi-day planning and multi-depot assignment are usually the features that sit on a higher tier of a rate card, which is worth knowing before you compare an entry-level quote against a full-featured one.

Hours of service is a hard constraint, not a preference

For drivers subject to the federal rules, the shift constraint is not a company policy that an optimizer can stretch. 49 CFR 395.3 sets the limits for property-carrying vehicles, and they are specific:

A plan that puts 11.5 hours of driving on a regulated driver is not an aggressive plan; it is a plan that cannot be executed lawfully. The 14-hour window is the one that catches planners out, because it keeps running through loading, waiting and breaks — so an hour lost at a depot in the morning removes an hour of driving capacity at the end of the day even though the driver has driven nothing extra.

If your drivers are on electronic logging devices, the available-hours data already exists in a system you own. Whether your route optimizer reads it is a separate question and a good one to ask. Our glossary covers hours of service and the ELD in more detail.

Why the problem is hard, briefly

The academic name for it is the vehicle routing problem, and the reason it resists brute force is combinatorial. With a handful of stops you could in principle evaluate every sequence; the number of possible sequences grows factorially with the number of stops, so at realistic sizes — a hundred stops, a dozen vehicles, capacity limits and time windows — enumerating them is not an option on any hardware.

What commercial optimizers do instead is search: build a reasonable solution quickly, then improve it by making local changes — move a stop to another route, swap two stops, reverse a segment — and keep the changes that help. That is why two products given the same stops produce different answers, why running the same optimizer twice can produce slightly different routes, and why none of them will tell you a plan is optimal. They will tell you it is the best they found in the time allowed.

The practical implication for an evaluation: give both candidates the same real day and compare the plans they produce, rather than comparing feature matrices. And look at what each one did with the stop that could not be served inside its window, because that reveals the priority order built into the product.

What it changes, and what we are not going to claim

The mechanisms by which optimization reduces cost are straightforward and worth stating plainly: fewer miles for the same stops, fewer vehicles for the same workload, fewer failed deliveries from arriving outside a window, less planner time spent assembling routes by hand, and fewer overtime hours from plans that were never achievable.

What we are not going to do is attach a percentage to any of that. FleetOpsClub runs no testing programme and holds no customer dataset, so any savings figure here would be either a vendor's marketing claim repeated as fact or a number we made up. Both are worse than nothing. The figures that do circulate in this category — the ones in the 10 to 30 percent range — almost all originate with vendors describing their own customers, and the baseline they improved on is rarely stated.

If you want a defensible number for your own operation, the only way to get one is to run a parallel period: plan one week by hand as usual, plan the same week in the tool without releasing it, and compare the two on miles, vehicles used, and planned finish times. That takes a week and costs nothing, and it produces a figure you can actually put in a business case.

The part nobody warns you about: the inputs

Optimization amplifies the quality of its inputs, in both directions. Three inputs account for most disappointed rollouts.

Addresses that do not geocode to the right place. A plan built on a geocode 400 metres from the actual loading entrance looks fine and fails on the road, and the error is invisible in the planner because the pin is in roughly the right neighbourhood. This is the one to fix first, and it is a data-cleaning job rather than a software job.

Service durations that reflect hope rather than history. If a stop takes nine minutes and the system is told five, a 20-stop route is 80 minutes short by the end of the day. Every optimizer will then produce a plan that is 80 minutes wrong and present it with confidence. If you have timestamps from a driver app or a previous system, derive service durations from them per customer rather than using one fleet-wide figure.

Order cut-off times that sit later than the planning run. Optimization needs the day's demand before it can assign it. Operations whose orders keep arriving until an hour before dispatch end up re-planning continuously, which is a different product requirement — real-time re-optimization — and usually a higher tier. Fixing the cut-off time is often cheaper than buying the feature.

How this category is sold, and why quotes do not compare

Route optimization is priced in at least four incompatible units, which means two quotes for the same operation are not comparable until you normalise them. The figures below were read from each vendor's own pricing page on the date shown, as part of our check of all 38 vendors we cover.

Seven of the 25 products we list in this category publish a rate at all. The rest either state no price, serve a pricing page that could not be retrieved, or appear only in third-party estimates. The full count across all 38 vendors we cover is in our pricing transparency index.

The consequence for a shortlist is specific. You cannot rank this category on cost, because cost is available for fewer than a third of it. Rank on fit, get to two or three candidates, then ask each of them for a rate in the same unit and in writing. Our head-to-heads work through what that looks like: OptimoRoute against Route4Me is a published per-driver rate against a verified absence, and Onfleet against Routific is two vendors who both publish but meter different things.

Do you need it?

A rough test, in the order the answers matter.

• Does a human currently decide which vehicle takes which stops? If no — the stops are fixed routes that never change — you need navigation, not optimization.
• Do your stops have time windows, or capacity limits, or both? If neither, sequencing software is probably enough.
• Does the day's demand vary enough that yesterday's routes are wrong today? Stable demand is the case where manual planning holds up longest.
• Is a planner spending more than an hour a day assembling routes? That hour is the clearest cost to measure against a subscription.
• Are drivers regularly finishing late, or missing windows? If so, the fault may be service durations rather than sequence, and that is worth checking before buying anything.

If the answer to the first two is yes and the planner hour is real, this category pays for itself on planner time alone at most published rates. If your stops are fixed and your demand is stable, it will not, and the honest answer is to spend the money elsewhere.

When you get to the software question, the products themselves are compared on our route optimization page, and the adjacent categories are worth a look too: delivery platforms if the recipient experience is part of the job, and dispatch software if the real problem is assigning work to people rather than sequencing stops.