In short: Bunker fuel procurement bundles three decisions, where to lift, how much to lift and against which specification, and they are usually made by different people in different sequences. A greedy rule, filling up wherever fuel is cheap relative to the next credible option, is close to optimal on the quantity decision and can be applied without a model. Fuel in the tanks adds displacement and displacement costs consumption, so carrying a cheap lift past a cheaper port has a running cost. Bunkers are bought against ISO 8217, where the parameters are constraints on machinery rather than commercial adjustments to trade away.
The vessel is on a fourteen day passage with 620 tonnes remaining and two credible bunkering options. One is 45 dollars a tonne cheaper and sits 180 nautical miles off the direct track. The chartering desk wants the cheaper price, the operations team is protecting a laycan, and the technical superintendent has just circulated a note about cat fines from a previous lift at that port.
All three are arguing about different terms in the same calculation, and the calculation is short enough to settle the argument in an afternoon if anyone writes it down.
Three decisions inside one purchase
A bunker purchase is three decisions bundled together, and they get made by different people in different sequences.
Where to lift, which is a network decision involving deviation distance, port cost, availability and the route ahead. How much to lift, which trades the price differential against the cost of carrying fuel and the risk of arriving with too little. And which grade, which is a compliance and machinery decision constrained by the regions the vessel will trade in.
Treating these as three separate approvals produces the common failure, where the cheapest price per tonne is secured at a port whose deviation and port cost consume the whole saving, and the grade lifted turns out to be marginal on a parameter that costs a liner overhaul eight months later.
The greedy rule that is close to optimal
There is a clean piece of theory here that is worth knowing because it produces a rule of thumb you can apply without a model.
Khuller, Malekian and Mestre analysed what they called the gas station problem in ACM Transactions on Algorithms in 2011: given a route with a set of stations, each with its own price, and a tank of fixed capacity, find the cheapest way to travel from origin to destination. Their central structural result gives the shape of the optimal policy.
At any station, look ahead as far as a full tank will carry you. If there is a cheaper station within that range, buy only enough to reach it. If every station within range is more expensive, fill the tank completely here.
That rule is exact under the simplifying assumptions of the model, and it is a good heuristic for the real thing. It also explains behaviour that looks inconsistent from outside. A vessel lifting a small stem at an expensive port and a full stem at a cheap one is following the policy correctly, and an instruction to always lift at the cheapest available port would do worse.
Besbes and Savin extended the treatment to joint route selection and refuelling with uncertain prices in Manufacturing and Service Operations Management in 2009, which is closer to the operating reality where the price at the next port is not known when you leave this one. The qualitative conclusion survives: the decision is about where the next cheap opportunity is relative to your range, rather than about the price in front of you.
Deviating for a cheaper port, worked
Put the numbers on the case at the top.
The saving. A 2,000 tonne lift at 45 dollars a tonne below the alternative is 90,000 dollars.
The deviation. An extra 180 nautical miles at 12 knots is 15 hours. At a consumption of 28 tonnes a day that is 17.5 tonnes of fuel, worth about 9,600 at 550 a tonne. Fifteen hours of time charter equivalent at 22,000 a day is 13,750. Add an incremental port call cost of 12,000 covering agency, pilotage, towage and dues, and possibly a barge waiting charge.
Total cost of the deviation is roughly 35,400, against a 90,000 saving. The deviation pays, with about 55,000 net.
Now change one input and watch the answer move. If the price differential is 15 dollars a tonne rather than 45, the saving is 30,000 and the deviation loses money. If the vessel is on a tight laycan and the 15 hours risks missing it, the demurrage or the cancelled fixture dwarfs everything on the sheet. And if the lift is 700 tonnes rather than 2,000, the saving is 31,500 and the deviation again fails.
The break-even is the useful output. Divide the deviation cost by the stem size and you get the differential you need: 35,400 over 2,000 tonnes is 17.70 a tonne. Any differential below that fails for this deviation at this stem size. Computing that number for your standard deviations and putting it on a card removes the recurring argument entirely.
Carrying fuel costs fuel
The term most often left out is that fuel in the tanks adds displacement, and displacement costs consumption.
The relationship comes from the Admiralty coefficient, the long-standing naval architecture approximation in which required power varies with displacement to the two-thirds power and with the cube of speed. So the fractional increase in consumption from carrying extra weight is the displacement ratio raised to two thirds.
Take a vessel at 60,000 tonnes displacement carrying an extra 1,000 tonnes of bunkers. The ratio 61,000 over 60,000, raised to the power two thirds, is 1.0111. Consumption rises by 1.11 percent. Over a twenty day passage at 28 tonnes a day, that is 560 tonnes of base consumption and 6.2 tonnes of extra burn, worth about 3,400.
So carrying an extra 1,000 tonnes for twenty days costs roughly 3.40 a tonne of fuel carried. If the differential you are capturing is 45 a tonne, that is noise and you should lift the maximum you can. If the differential is 5 a tonne, the carrying cost eats most of it, and there is also the freight you displaced by loading fuel instead of cargo on a vessel that is deadweight-limited, which on some trades is the largest term of all.
That last case is worth checking rather than assuming. On a fully laden voyage where every tonne of bunkers is a tonne of cargo forgone, the opportunity cost is the freight rate per tonne, which can be an order of magnitude above the consumption effect and inverts the whole decision toward lifting the minimum safe quantity.
Specification is a constraint before it is a price
Bunkers are bought against ISO 8217, and the parameters in that standard are constraints on machinery rather than commercial adjustments.
Two failure modes cost more than any price differential.
Catalytic fines. Aluminium and silicon particles carried over from refinery catalytic processes are abrasive, and they cause liner and fuel pump wear that shows up months later. ISO 8217 permits residual grades up to 60 milligrams per kilogram of aluminium plus silicon at delivery, while engine designers' guidance for the fuel actually entering the engine is roughly a quarter of that figure. The gap between those two numbers is closed by the vessel's own separators, so a delivery at the top of the specification is compliant and still demands that the purification plant be working properly. A lift accepted at 55 milligrams per kilogram on a vessel with a marginal separator is a compliant purchase and an expensive one.
Stability and compatibility. Low sulphur residual fuels are blends, and blends from different suppliers can be individually stable and produce sludge when mixed. The remedy is segregation of stems in separate tanks, which requires tank capacity that the fuel planning has to account for. A vessel with limited segregated capacity has fewer purchasing options than its total bunker capacity suggests.
Both of these argue for sampling and testing discipline that most operators have on paper. The representative sample taken at the vessel's manifold during delivery is the document that decides any subsequent dispute, and testing turnaround usually means the fuel is already aboard and possibly in use by the time the result arrives. A policy of holding a new stem out of service until the analysis returns costs tank flexibility and prevents the expensive class of failure.
The compliance layer now prices the molecule
The purchase price of a tonne of bunkers is now a smaller share of its landed cost than it used to be, and the additional terms are regulatory.
The IMO's global sulphur limit of 0.50 percent mass by mass has applied since January 2020 under MARPOL Annex VI, with 0.10 percent inside emission control areas. That set the grade structure the market trades: very low sulphur residual, marine gas oil, and high sulphur fuel for vessels with exhaust gas cleaning.
On top of that, the European Union brought maritime transport into its emissions trading system from 2024, phasing in coverage at 40 percent of verified emissions for 2024, 70 percent for 2025 and 100 percent from 2026, applied to 100 percent of emissions on voyages between EU ports and 50 percent on voyages with one end outside. And Regulation (EU) 2023/1805, known as FuelEU Maritime, has applied since January 2025, setting a declining limit on the greenhouse gas intensity of energy used on board with penalties for exceeding it.
Work the arithmetic, because it changes the ranking of ports. The default carbon factor for residual fuel oil in the EU monitoring rules is close to 3.11 tonnes of carbon dioxide per tonne of fuel. At an allowance price of 70 euros, that is about 218 euros of carbon cost per tonne of fuel burned on a fully covered voyage, against a fuel price of perhaps 520. Substitute the current allowance price rather than that one.
Two consequences follow immediately. A 45 dollar a tonne price differential between two ports is smaller than the carbon cost attached to the same tonne, so any procurement process that optimises the fuel price without the compliance cost is optimising the smaller number. And the effective price of a tonne now depends on where it will be burned, so the same physical stem has different economics depending on the voyage plan, which means the bunker decision and the voyage plan have to be made against a single model rather than sequentially.
The price risk this creates across a fleet is a separate discipline covered in N17.
The limit
Everything above assumes the price you were quoted is the price you pay for the quantity you receive, and the quantity is the part that is genuinely contested. Bunker quantity is determined by tank measurements on the delivering barge, and disputes over short delivery are a persistent feature of the trade. Mass flow metering has improved this materially where it is mandated, and where it is not, the vessel's own measurement discipline during delivery is the only protection. A procurement analysis accurate to a dollar a tonne on a quantity that may be one percent optimistic is measuring the wrong thing first.
Counterparty risk is the second limit and it is not symmetric. A supplier offering a price meaningfully below the local market is telling you something, and what it usually tells you is about credit terms, quality, or the reliability of the barge showing up in the window you need. The cheapest quote in a port is frequently a real opportunity and occasionally an operational failure that costs a day of hire.
And the theory in the middle of this piece assumes prices at future ports are known. They are not, and the volatility over a two week passage is comparable to the differentials being chased. That argues for making the decision on the current spread against a break-even you have precomputed, rather than on a view about where prices will be when the vessel arrives.
Compute the break-even differential for your three most common deviations, at your standard stem sizes, and give the number to whoever approves the next lift.