In short: In agricultural supply planning the quantity and quality of supply are set by the harvest, so the plan allocates what arrived rather than deciding how much to make. Intake capacity binds hard for the five to eight weeks the crop comes in and is irrelevant for the rest of the year, which makes segregation the decision that sets the year's options. Blending two quality bands into one bin is irreversible, and it removes premium sales that were available that morning. The carry decision, whether to hold crop or sell it, has a clean answer from the theory of storage for the part of the position that is genuinely optional.
Intake starts on the fourteenth of August and by the twenty-second there are 40 lorries in the queue before seven in the morning. The lab is running moisture and protein on every load, the drier is at capacity, and the intake manager has fourteen bins and eleven quality bands he would like to keep separate. He has already blended two of them together because bin six filled faster than anyone expected, and that decision has quietly removed a milling contract from the year's options.
Six weeks later the whole crop is in, and the business now knows what it has. From that point until next August the supply plan is an allocation problem: 180,000 tonnes of a known quality distribution, against twelve months of orders, with almost no ability to acquire more of the good stuff at a sensible price.
Almost every planning habit imported from manufacturing gets this backwards, because it assumes supply is something you decide.
Supply is an input with a distribution, and the plan starts after it lands
In a factory, the supply plan answers how much to make. Here it answers how to spend what arrived. The quantity is set by area planted and yield, the quality is set by the weather in a handful of critical weeks, and both are known with any precision only once the crop is in the shed.
That inverts the usual sequencing. The demand plan and the supply plan stop being two things to reconcile and become one allocation: which customers, which specifications and which months get served from which part of the crop. The decision variables are segregation, blending, drying and the carry.
One practical consequence is that the planning year runs harvest to harvest and the financial year usually does not. A crop year that starts in August and a reporting year that starts in January means every set of accounts cuts a crop in half, and the two halves are drawn from different harvests with different qualities and different costs. Businesses that run their supply plan on the financial calendar spend a lot of time explaining variances that are really just the join between two crops. Running the plan on the crop year and mapping to the financial one for reporting is the arrangement that causes fewer arguments.
It also changes what a forecast is for. Forecasting demand still matters, because you are choosing between customers. Forecasting supply matters more, and it matters earliest, because the contracting decisions that determine how much crop you will handle are made months before harvest. The in-season estimation methods that narrow that uncertainty are their own subject (A2).
Intake capacity is the constraint for six weeks and irrelevant for forty-six
Annual throughput is never the problem. The problem is that a year's material arrives in five to eight weeks, and the weighbridge, the sampling lab, the tipping pits and the drier all have hourly rates.
The cost of getting this wrong is subtle, because it lands on the grower rather than on you. A grower who queues three hours at your gate during the only dry week of the harvest remembers it when contracting season comes round, and the supply you lose next year does not appear in any variance report about this year's intake. Businesses that measure gate-to-gate time during harvest and publish it to growers tend to hold their catchment better than those that measure only tonnes received.
The planning question is how much peak capacity to build for a peak that happens once. A drier sized for the ninetieth percentile harvest week sits idle most of the year. The honest framing prices the alternatives against each other: additional drier capacity, temporary external storage, paying growers a premium to deliver in a wider window, or accepting queues. Only the last one is free, and it is free to you rather than to the system.
Segregation is the decision that sets the year's options
Grain, oilseed and most stored crops arrive as a quality distribution rather than a grade. Moisture, protein, specific weight, admixture, oil content and contamination all vary load to load, and once two loads are in the same bin the decision is irreversible.
More bands means more optionality and more bins. Fewer bands means blending, and blending is asymmetric: mixing a 13.2 protein wheat with an 11.4 gets you something around 12.3, which sells against the lower specification, and the premium on the better half is gone.
Work the arithmetic. Say 180,000 tonnes arrive with 22 percent of it above the milling protein threshold, carrying a 24 a tonne premium. Segregate it properly and that is 39,600 tonnes earning 950,000 of premium. Run out of bins in week three and blend half of it into the feed pool, and you have given away 475,000, most of it during a fortnight when nobody had time to think about it.
The way to stop losing that money is to decide the segregation policy before intake rather than during it, using the previous three harvests' quality distributions to size the bands, and to hold one bin empty as a swing for the band that fills unexpectedly. That is a planning decision made in June about a constraint that binds in August, and it is worth more than most of what gets optimised later in the year.
Drying and contamination are where a hard limit sits
Moisture at intake decides drying cost and storage safety. Drying costs energy and shrinks weight, so over-drying gives away tonnes you were paid for. Under-drying risks spoilage across the whole bin, which is a loss of a different magnitude.
Contamination is where this stops being an economic trade. EU maximum levels for mycotoxins in food, in force under Commission Regulation (EC) No 1881/2006 and recast in 2023, set limits for deoxynivalenol, ochratoxin A, aflatoxins and others, with separate provisions for animal feed. A lot above the limit for a food outlet cannot be sold into it. There is no price at which that transaction is available.
That belongs in the plan as a filter applied before any allocation runs, generated from the lab results attached to each bin. Never encode it as a cost with a large penalty, because a penalty is a price and an optimiser will eventually find a case where paying it looks attractive, producing an allocation nobody can execute. Build the feasible set first, from the analytical results and the outlet specifications, and optimise the allocation inside it.
The interesting planning content sits just inside that boundary. A bin testing close to a limit has a distribution of true values around the sample result, so treating it as compliant on a single test is a bet. Sampling more heavily on the marginal bins and less on the clearly clean ones is a cheap improvement, and the sampling plan is itself an optimisation with the cost of a test on one side and the cost of a rejected consignment on the other.
The carry decision has a formula and it is usually ignored
Holding crop rather than selling it is the oldest question in this business, and it has a clean answer for the part of the position that is genuinely optional. The theory of storage, set out by Kaldor in the Review of Economic Studies in 1939, developed by Working in the American Economic Review in 1949 and by Brennan in the same journal in 1958, says that carrying pays when the forward premium exceeds the full cost of carry.
Put numbers to it. Spot is 210 a tonne, the six month forward is 222, so the market offers 12 to carry. Against that, physical storage runs 1.6 a tonne a month, which is 9.6. Finance at six percent on 210 for half a year is 6.3. Shrink and handling loss at 0.4 percent is 0.8. Insurance and fumigation add a little more. Total cost of carry is around 17.5, against a 12 spread, so carrying loses about 5.5 a tonne. On 60,000 tonnes that is 330,000 that the business will lose slowly and never book as a decision.
The reason processors carry anyway is that they are not traders. A mill that runs out closes, and the value of having material available exceeds the spread arithmetic. That premium is convenience yield in the literature and operational cover in the plan, and the two should be separated explicitly: the volume needed to keep the plant running is a cover decision sized against consumption and replenishment risk, and everything above it is a position with a profit and loss. Most businesses hold one number and defend it with whichever argument is convenient that month.
Contracting is where the real uncertainty gets shared
The commitments that decide how much crop you handle are made before anyone knows the yield. Fixed tonnage contracts, area contracts, pools and minimum price arrangements all divide the risk differently, and they are usually chosen by habit rather than by analysis.
The exposure worth measuring is the joint one. A fixed tonnage contract in a poor year means buying on a rising market to cover it, so your cost goes up exactly when your volume goes down, and the two losses compound. An area contract passes yield risk to you and price risk stays wherever the pricing mechanism put it. Running your contract book against the last fifteen harvests, jointly on yield and price, tells you which combination actually breaks, and the answer is frequently a modest-looking exposure that turns severe in the two worst years of the fifteen.
Ahumada and Villalobos reviewed planning models across the agri-food chain in the European Journal of Operational Research in 2009 and their observation still holds: the models that get built concentrate on the harvest and post-harvest stages, and the contracting stage upstream of them, where most of the risk is allocated, is comparatively neglected.
Where this stops
The allocation model is only as good as the quality data attached to the stock, and in most operations that data degrades after intake. Bins get topped up, material gets moved for operational reasons, and the recorded quality of a bin drifts away from the sample results it was built from. Weighted averaging on transfer helps, and it accumulates error. Any business planning allocations against quality that has never been re-tested after storage is working from a model of the crop rather than from the crop.
The carry calculation also assumes a liquid forward market for something close to your product. It works cleanly for major grains and oilseeds, and it degrades quickly for crops with thin or non-existent futures markets, where the forward premium has to be inferred from forward physical offers with a wide bid-ask. In those markets the decision is closer to judgement, and the discipline available is to write down the assumed forward price at the time of the decision so the review afterwards is about the decision rather than about the outcome.
The last limit is that none of this makes a bad harvest good. The value of planning well here concentrates in the middle of the distribution, in the ordinary years where segregation, drying targets and allocation between outlets are genuinely choices. In the worst year the crop decides and the job becomes triage.
Take the last three harvests, plot the intake quality distribution for your main crop against your current bin plan, and count how many tonnes of premium material would have had nowhere to go.