In short: A marine programme is a set of commitments that were feasible when they were made, and a chokepoint closure or a lost berth puts every commitment downstream of it in question at once. Experienced schedulers reach a workable answer quickly, and what the whiteboard version omits is a price on the alternatives, so the gap between the first feasible plan and the best one goes unmeasured. The reschedule is a constrained assignment problem over cargoes, vessels, laycans, berths and tank availability, which means the options can be costed rather than argued. Where the duration of the disruption is unknown, staging the decision by taking the actions that hold under either outcome and setting a date to resolve the rest beats committing to one optimised plan.
A marine programme is a set of commitments that were feasible when they were made. Cargoes assigned to vessels, laycans agreed with buyers, berths booked, tanks allocated for the loading sequence.
Then a strait closes, a canal restricts transits, a terminal loses a berth to weather, or a war risk premium makes a lane commercially unavailable. Every commitment downstream of that constraint is now questionable, and somebody has to work out which ones can still be kept and what the alternatives cost.
This gets done under time pressure, usually by a small number of people with deep experience and a whiteboard, and it gets done reasonably well. What it does not get is priced, and the difference between the first workable answer and the best one is frequently large.
The structure of the problem
Underneath the disruption is a fairly standard assignment and routing problem with constraints that make it specific.
Cargoes have to be matched to vessels, and the match is constrained by vessel size against berth and canal restrictions, by cargo compatibility with previous cargo carried, by vessel position and availability, and by charter economics.
Routes have costs that are not only distance. Bunker consumption at a chosen speed, canal or transit fees, war risk and insurance premiums that vary by lane and by week, and port charges at each call.
Time windows bind at both ends. A laycan with the buyer, a berth slot at the load port, a discharge window at the destination, and demurrage accruing whenever a vessel waits outside a window it cannot use.
Speed is a decision variable rather than a constant, which is the part most manual rescheduling handles least well. Bunker consumption rises steeply with speed, so arriving three days earlier can be expensive, and slow steaming to arrive exactly on time is frequently the cheapest option available. Getting that trade-off right across a programme of many vessels is not something a whiteboard does well.
Consumption goes roughly with the cube of speed for a given hull. That approximation is the one the marine operations literature has worked with since Ronen's 1982 paper in the Journal of the Operational Research Society on optimal ship speed under fuel prices, and Psaraftis and Kontovas surveyed the family of speed models built on it in Transportation Research Part C in 2013. Cubing turns a modest speed change into a large fuel bill.
Work it through. A vessel burns 45 tonnes a day at 12 knots and has 6,000 nautical miles to cover. Twelve knots is 288 miles a day, so the voyage takes 20.8 days and burns 937 tonnes. Lift the speed to 13.5 knots. The ratio is 1.125, cubed it is 1.42, so consumption goes to 64 tonnes a day. The voyage falls to 18.5 days and the burn rises to 1,186 tonnes.
That buys 2.3 days for 249 extra tonnes. At 600 dollars a tonne it costs about 149,000 dollars, or roughly 65,000 dollars a day of time gained. The decision then has an answer rather than an opinion. Speeding up pays if the demurrage exposure, the laycan penalty and the downstream cost of a late arrival together exceed 65,000 dollars a day, and it does not pay below that. Run the same comparison across thirty vessels on a disrupted routing and the aggregate difference between getting it consistently right and getting it approximately right is the whole prize.
What changes in a disruption
Three things move at once, which is why the problem is harder than ordinary scheduling.
The feasible set shrinks. Some routes are unavailable, some vessels are in the wrong place, some berths are gone. Any schedule that used them is invalid.
Costs reprice sharply. Freight rates on the alternative lanes move within days, insurance premiums jump, and bunker prices at particular ports move with demand. A rerouting decision made on last month's rate card will be wrong.
Commitments become renegotiable, at a price. A laycan can be shifted if the buyer agrees, and the agreement usually costs something. Treating every commitment as fixed produces an infeasible problem; treating them all as free produces an answer nobody will accept. The correct treatment prices each commitment change explicitly, so the optimiser trades a small laycan slip against a large freight saving and shows the trade.
That last point is what separates a useful rescheduling tool from an academic one. Most real disruption responses involve renegotiating something, and a system that cannot represent the cost of renegotiation cannot find the answer a human would find.
An example makes the shape of it obvious. Shifting a laycan by four days costs, say, 180,000 dollars in price adjustment and commercial goodwill. Holding the original laycan requires a rerouting that adds 2,900 miles plus a war risk premium on the alternative lane. If that rerouting comes to 700,000 dollars against the original plan, the laycan slip is worth 520,000 dollars and somebody should be making the call today. If it comes to 120,000, asking for the slip spends relationship capital to lose money. Both answers fall out of the same model, and the commercial team gets handed a priced request rather than a plea.
Running it as a decision rather than a plan
Two properties make the output usable in the room where the decision is made.
Options rather than an answer. A single reoptimised schedule invites an argument about assumptions. Three or four schedules with different postures, protecting delivery dates at maximum cost, minimising cost at the expense of some late arrivals, and a balanced option, let the commercial and operating sides see the trade-off in terms they each recognise. The differences between the options are the actual decision.
Deltas against the current programme. Nobody wants a fresh schedule. They want to know which twelve of forty cargoes change, what each change costs, and who has to be called. A rescheduling output that is not expressed as a change list will be translated into one by hand, badly, under pressure. The format that works is one row per changed cargo carrying the old and new vessel, the old and new dates, the cost delta, and the name of the person who has to agree to it. Anything that asks a reader to diff two schedules themselves will get diffed wrong at two in the morning.
Stability deserves an explicit weight in the objective. A schedule that saves two percent by rearranging thirty cargoes is worse in practice than one that saves one and a half percent by rearranging four, because each change carries coordination cost and risk that the model does not see.
Preparing before the disruption
The response quality is largely determined before anything happens, by whether three things exist.
A current, accurate picture of vessel positions, cargo assignments and commitments in one place. This sounds basic and is frequently the binding constraint, because the information lives across chartering, operations, terminal and commercial systems that were never joined.
A cost model that can be repriced quickly. Freight rates, bunker prices, transit fees and insurance premiums as parameters that can be updated in an hour rather than assumptions embedded in a spreadsheet somebody built two years ago. A cheap test of this: ask for the current programme repriced at today's bunker and freight assumptions and time how long it takes to come back. Under a day means you have a cost model. A week means you have a spreadsheet, and in a disruption a week is the entire decision window.
Pre-computed alternatives for the known chokepoints. There is a short list of transit points whose closure would be materially disruptive, and the alternative routings for each are knowable in advance. The Panama Canal Authority's transit restrictions during the 2023 and 2024 drought gave months of warning to anyone watching lake levels, and most of the programmes affected were still working out their options after the restrictions took effect. Working the option set out during the crisis wastes the days when the option set is widest.
The failure mode to look for is a position picture that is accurate and stale. Vessel positions updated from a daily noon report are adequate for planning and useless in the first 48 hours of a disruption, when the question is which vessels are still upstream of the constraint and which have already committed to a transit. The symptom is an operations team keeping a parallel whiteboard because they do not trust the system's positions, and that habit tends to get read as diligence rather than as the data problem underneath it.
One diagnostic worth a day of somebody's time: take the three largest deviations from the original programme over the last two years, reconstruct the alternatives that existed at the moment each decision was made, and reprice them against what actually happened. The output measures how wide the option set was and how much of it was visible at the time, which is more useful than a verdict on the decisions themselves. The visibility half is the part you can change before the next one.
The wider consequence
A marine disruption is rarely only a marine problem. Cargoes that arrive late are feedstock that does not reach a refinery, or product that does not reach a customer, and the cost of the marine decision is partly incurred somewhere else.
The version of this that is worth building connects the schedule to the downstream consequence, so that a delayed cargo shows its effect on plant operation or on customer service rather than only its freight cost. Otherwise the marine team optimises freight, the plant absorbs the disruption, and the total cost is higher than either would have chosen.
A rough connection beats none. If a late feedstock cargo forces a 200 thousand barrel a day crude unit to cut rate by fifteen percent for four days, that is 120 thousand barrels of lost throughput, and at a six dollar margin it is 720 thousand dollars. Set that against the 149,000 dollars the earlier speed calculation would have cost to arrive on time and the ordering is obvious, which is the point. The downstream figure is frequently several times the freight figure, and it is available from the plant's own economics in about ten minutes.
The limits
Optimisation over a disrupted programme assumes the disruption has a known shape. In practice the duration is the largest uncertainty and it is usually unknown at the moment the decision has to be made. A closure lasting five days and one lasting five weeks call for different responses, and committing to the five week response on day two is expensive if it reopens.
The reasonable treatment is to decide in stages: take the actions that are correct under both scenarios first, defer the ones that differ, and set a date by which the deferral has to resolve. That is less satisfying than a single optimised plan and it matches the actual information available.
There is also a limit on what any model knows about commercial relationships. Some buyers can be asked to shift a laycan and some cannot, for reasons that are about the relationship rather than about a contract clause. That knowledge sits with the people in the room, and the right design surfaces the priced options for them to choose among rather than attempting to encode the judgement.