Under 20 vehicles the answer is not a percentage — it is a structural question about stop density, and you can answer it from records you already keep.
Key takeaways
Under 20 vehicles, route optimisation pays off when your stops change most days and each route has enough stops that a human cannot hold the sequence in their head. It does not pay off when routes are contractually fixed, when you run under roughly a dozen stops per vehicle, or when the real constraint is not the sequence at all but time windows you do not control.
That is a structural answer rather than a percentage, and deliberately so. Any published saving figure comes from someone else’s stop density, and stop density is the variable that decides the whole question. What follows is how to work out your own number from data you are already required to keep.
Optimisation is not one thing. Four distinct mechanisms are bundled under the label, and a small fleet usually benefits from only two of them.
You do not need a pilot to estimate the prize; you need your own distance data. If you run interjurisdictionally and hold an IFTA licence, that data is a legal obligation already. Ontario’s Interjurisdictional Carrier’s Manual requires distance records maintained on an individual vehicle basis, and lists what each trip record must contain: start and end dates of the trip, point of origin and destination, list of highways used, odometer or hubodometer readings at the beginning and end of each trip and at each jurisdiction’s borders, total trip distance, distance travelled in each jurisdiction, power unit or vehicle identification number, fleet or unit number, and the name of the registrant.
Those records must be kept for four years from the return due date or the filing date, whichever is later. So a carrier with an IFTA obligation has four years of per-vehicle, per-trip distance sitting in a filing cabinet or a database. That is the baseline. Divide distance by completed stops for a month and you have your current cost per stop in kilometres — the only number a vendor cannot argue with.
If IFTA does not apply to you
Ontario’s IFTA guidance says a qualified motor vehicle is one with two axles and a gross or registered gross vehicle weight of more than 11,797 kg, three or more axles regardless of weight, or a combined weight over 11,797 kg with a trailer — and that light trucks and vans at or below that weight, including delivery vans and courier services, are not required to register. A van fleet therefore has no IFTA distance record to mine, and should build the same baseline from odometer readings at fuelling.
Optimisation that ignores duty-time rules produces plans that cannot legally be run. For vehicles above 4,500 kg registered gross vehicle weight — the threshold at which the Commercial Vehicle Drivers Hours of Service Regulations define a commercial vehicle — the driver’s available hours are a hard constraint on the plan, not a preference.
Local fleets get a meaningful simplification here. The electronic logging device and daily record obligations do not apply where the driver operates within a radius of 160 km of the home terminal, returns to the home terminal each day to begin a minimum of 8 consecutive hours off duty, and the carrier keeps accurate and legible records showing the cycle followed and on-duty times for at least 6 months. A great many small delivery and service fleets sit entirely inside that radius. If yours does, the optimiser needs to respect that 160 km boundary as a constraint — a route that pushes one vehicle to 175 km changes that vehicle’s compliance regime for the day.
Where the regime does apply, note that the carrier must monitor the compliance of each driver, take immediate remedial action on non-compliance, and record the dates and the action taken. Optimisation software that plans hours it cannot legally use creates that record for you, in the wrong direction.
An 11-truck HVAC service fleet in Alberta runs about 60 jobs a day, five to seven stops per truck, with two-hour customer windows. It is considering an optimiser.
The honest analysis: at six stops per route, sequencing gains are small — a competent dispatcher is nearly optimal at that size. The real inefficiency is assignment. Two trucks are territorially locked to the north end and routinely finish early while the south trucks run over. The right test is therefore not “does the optimiser find a shorter route” but “does it rebalance stops across the whole fleet against the two-hour windows”.
The measurement is: take four historical weeks, feed the actual stops and windows into the tool, and compare planned kilometres and planned finish times against what actually happened. If the tool cannot ingest history and replay it, that is a meaningful signal about the product. Run the comparison before you buy, not during a trial where dispatchers are being watched.
Note what has not been claimed here: a percentage. Yours will depend on your stop density, your window width and how much slack your current territories carry. Someone else’s figure tells you nothing.
Be willing to reach it. Optimisation does not pay when routes are contractually fixed with the customer; when the sequence is dictated by an external constraint such as a dock appointment or a clinic’s collection schedule; when stops per route are in single digits and territories are already balanced; or when your dispatcher’s real job is negotiating with customers rather than sorting addresses. In those cases the money is better spent on notification, proof of delivery and exception handling.
Two things, both about the inputs rather than the algorithm. Service-time prediction — learning from your own history that a particular site takes 40 minutes and not the 15 in the standing record — is the single largest source of plan-versus-actual drift in small fleets, and it is a straightforward learned estimate. And unstructured-instruction parsing: turning “buzz 4B, loading dock closes at 3, ask for Dev” into structured constraints the optimiser can honour.
The routing engine itself has been solved arithmetic for decades. The improvement available to a small fleet is almost always better inputs, not a better solver. And the plan remains a proposal: the dispatcher accepts, edits or overrides it, and the driver’s hours and safety judgment are not the software’s to decide.
Vehicle count is the wrong unit. Stops per vehicle per day, and how much those stops change week to week, are the variables. A five-van fleet doing 80 changing stops each benefits more than a 25-truck fleet running fixed contracted routes.
It reduces distance, and fuel tax follows distance, but the filing obligation is unchanged. Ontario’s manual still requires quarterly returns due April 30, July 31, October 31 and January 31, with a penalty of 5 per cent of the net tax due where a return is late, incomplete or unpaid, and interest based on the Canadian federal Treasury Bill rate plus 2 per cent, adjusted quarterly.
Good ones can, and for vehicles over 4,500 kg it must, because the hours-of-service regulations cap what you can lawfully plan. Ask the vendor to demonstrate it with a driver mid-cycle, not with a fresh driver at the start of a day.
Not to optimise, but you need honest actuals to prove it worked. If you have no telematics, use odometer readings and stop timestamps from your dispatch records for the baseline — less precise, sufficient for a purchase decision.
A 30-minute call is enough to tell you whether AI pays for itself in your back office.
