In short: The costs of sizing a fleet too small and too large are not symmetric, since one shows up as standing cost every week of the year and the other as spot hire and service failure in the weeks that matter most. Demand variability sets the number more than average volume does, so the useful input is the distribution of daily vehicle requirement, and the distance between its median and its 95th percentile is the capacity a flexibility strategy has to cover. The real question is how much of the peak belongs on your own balance sheet, with the rest covered by contract hire, spot hire or handing volume to a carrier. Vehicle mix is a separate decision from vehicle count, because a fleet of the wrong composition runs empty space and turns away loads at the same time.
The utilisation report comes round each quarter. Forty-four vehicles on the fleet. In week 47 the operation ran 42 of them and hired three more. Across the year the average is a little over 27 out on any given day, which is 62 percent.
Somebody eventually asks why 44. The answer, when it can be reconstructed at all, is that the fleet was 40 a few years ago, peak got tight, four more were added, and nobody has revisited it since. The fifteen or so vehicles between the average requirement and the peak requirement are standing capacity that exists for a handful of weeks, and they cost the same in February as they earn in November.
That gap is where the fleet decision lives. Sizing to the peak is a defensible answer, sizing to the average is not, and the interesting question is how much of the peak belongs on your own balance sheet.
What each error costs, and they are not symmetric
Undersizing shows up loudly. Overtime, split loads, spot hire at whatever the market is charging in week 47, missed delivery windows, and a transport planner spending the busiest weeks of the year rebuilding the plan by hand. The costs are visible and they land inside the period, which is why undersizing gets fixed fast.
Oversizing shows up quietly and never gets fixed. The standing cost is real: depreciation or lease, insurance, tax, maintenance, telematics, inspection, and the yard space to keep it in. Say an owned rigid costs 30,000 a year in your currency to have, before a driver and before a mile is driven. Fifteen of them is 450,000 a year of cost that exists to cover the top of the distribution.
There is a second effect of oversizing that is worth naming because it distorts everything else. Slack fleet absorbs the pressure that would otherwise force the route plan to improve. When there is always another vehicle, a badly built route never becomes anyone's problem, and the inefficiency compounds until it looks like a capacity requirement.
Variability sets the number more than volume does
Two businesses with identical annual volume can need very different fleets, and the difference is the shape of the week-to-week distribution rather than its mean.
Take a mean requirement of 27 vehicles a day in both. Business A has a coefficient of variation on daily vehicle requirement of 0.12. Business B has 0.38, which is ordinary for anyone with a promotional calendar and a heavy month-end. Size each to cover the 95th percentile of its own distribution and A needs about 33 vehicles, while B needs about 44. Same volume, eleven more vehicles, and the entire difference is variability.
Read that arithmetic backwards and it points at the cheapest fleet reduction available to most operations, which has nothing to do with vehicles. Levelling the demand pattern reduces the required fleet faster than anything else on the list. Moving customers off the two heaviest delivery days, staggering month-end ordering by account, spreading the promotional calendar so three brands do not all land in the same week, and putting delivery day incentives in front of customers who genuinely do not care which day they get: each of those cuts the tail without touching the mean.
One caution on the arithmetic. Daily vehicle requirement departs from normality in two ways that matter: it is right-skewed and it is autocorrelated, because peaks cluster into runs rather than arriving as independent spikes. Use the empirical distribution from a year of actual requirement rather than a normal quantile, and look at the run length of consecutive high days, because a five-day run above capacity is a different operational problem from five isolated days.
Own, contract, spot, or hand it over
The options are usually presented as a cost per vehicle per day comparison, which is the one framing guaranteed to give the wrong answer, because the whole point of the flexible options is what they do to the tail.
Owned or long contract hire. Lowest cost per vehicle-day when the vehicle is working, full cost when it is not, and a commitment measured in years. Residual value risk sits with you on an owned asset and with the lessor on a contract, which is part of what the contract premium buys.
Short-term rental with a driver. Highest unit cost, close to zero commitment, and one significant hidden risk: availability. Everybody's peak is the same week. The spot market is thinnest exactly when your demand for it is highest, so the option you priced at a rate may not exist at any rate on the day you need it.
Dedicated third-party capacity. Someone else's vehicles and drivers reserved for you, usually with volume bands and a margin on top. It converts a fleet decision into a contract negotiation and moves the utilisation risk depending on how the bands are written.
Pallet networks and per-consignment carriers. Worst unit economics, genuinely elastic, and useful as an overflow valve rather than as core capacity.
The way to price this properly is a newsvendor calculation rather than a rate card comparison. For each candidate owned fleet size, run the year against the empirical daily requirement distribution: owned vehicles cost their standing charge whether they move or not, everything above the owned count costs the spot rate, and days where spot capacity is unavailable cost you a service failure at whatever you believe that is worth. Plot total expected annual cost against owned fleet size, and it is a shallow U with a minimum somewhere between the mean and the peak.
The decision rule underneath is a ratio you can compute on the back of an envelope. Owned standing cost of 30,000 a year is roughly 577 a week. If a spot vehicle costs about 900 a week more than the marginal cost of running one you already own, then the Nth vehicle is worth owning when you need it more than 577 divided by 900, or about 64 percent of weeks. Run that test down the fleet and the answer for the last several vehicles is usually clear, and usually different from what you have.
Two adjustments make it honest. Availability risk on the spot side raises the effective spot cost above its quoted rate, which pushes the answer toward owning. And availability can be bought separately: a standby agreement with a haulier, paid as a retainer for guaranteed access to a stated number of vehicles in stated weeks, is frequently cheaper than owning those vehicles and removes the risk that priced the spot option badly in the first place.
Vehicle mix is a different decision from vehicle count
Fleet composition and fleet size get bundled together in most reviews and they answer different questions. Golden, Assad, Levy and Gheysens formalised the fleet size and mix vehicle routing problem in 1984, and Hoff, Andersson, Christiansen, Hasle and Løkketangen's 2010 industrial survey makes the point that composition and routing are coupled in reality and solved separately almost everywhere in practice.
A mixed fleet serves a wider demand distribution. Small vehicles reach the sites large ones cannot and carry small drops without wasting capacity; large vehicles handle trunking and the days when volume concentrates. The cost is scheduling flexibility. Each class is its own pool, a breakdown in one pool cannot be covered from another, driver licensing narrows who can take what, and the buffering advantage of a large interchangeable fleet erodes as it fragments. The same pooling arithmetic that governs inventory buffers applies to vehicle pools, and splitting one fleet of forty into two of twenty costs you real coverage.
The workable rule is to add a vehicle class only when it solves a structural constraint no existing class can: an access restriction, a temperature requirement, a licensing threshold, a weight limit on a bridge you cross daily. Adding a class because a size looks more efficient on a spreadsheet usually costs more in fragmentation than it recovers in payload.
Optimise the routes, then size the fleet
Required fleet size is an output of the route plan. If your current plan was built by hand and grown by accretion, your current vehicle count overstates what the operation actually needs, and sizing off it commits that overstatement for the length of a lease.
The sequence that works is to fix the plan first (X1), run it against a full year of real historical orders, record the daily vehicle requirement the fixed plan produces, and size against that distribution. Skipping the first step is how businesses end up buying capacity to protect an inefficiency.
The routing change that removes vehicles is usually not a better sequence within the day. It is a better allocation of customers across days, because moving accounts off the heaviest day flattens the requirement distribution and the fleet is sized against the top of that distribution. Territory boundaries feed the same calculation (X2), since a territory that straddles two depots generates vehicle requirement neither depot can absorb efficiently.
The limit: an eight-year commitment on an eighteen-month forecast
Contract hire on a heavy vehicle runs five to seven years. The demand forecast you can defend with a straight face runs to about eighteen months, and beyond that you have scenarios with widening bands rather than a number.
There is no analytical trick that closes that gap. What you can do is change the shape of the commitment so being wrong is survivable in both directions.
Stagger the expiries so a predictable share of the fleet rolls off every year, which converts one large irreversible decision into a series of small reversible ones. Size the owned core to a level of demand you are confident about rather than to the expected level, treating the confident floor as the thing you own and everything above it as the thing you buy. Price break clauses and flex options explicitly instead of accepting the standard term, because a lessor will quote them and the quote tells you what the flexibility is worth. And keep the residual value assumption visible, since any powertrain currently subject to regulatory review carries more uncertainty in its resale value than a standard depreciation schedule reflects.
The uncomfortable part is that scenario ranges only tell you how wrong you might be. The choice of where to sit inside that range is a judgement about your own tolerance for standing cost against service failure. Making that judgement explicitly, and writing down what it assumed, at least gives the next review something to argue with.
Start with the distribution you probably do not have. Take a year of actual orders, compute the vehicle requirement each day under your current plan, and plot it; the distance between the median and the 95th percentile is the exact amount of capacity your flexibility strategy has to cover, and most operations have never seen that number written down.