In short: Shelf life in a perishable supply chain is a budget that starts being spent at harvest, and the hours between cutting and reaching temperature are usually missing from the receiving record. Quality loss roughly doubles for each ten degree rise in temperature, the Q10 convention, so thermal history explains more of the remaining life than the date printed on the box. Quality declines continuously and an expiry date is a projection of when a specification threshold gets crossed. Matching lots with different remaining lives to orders with different transit times and acceptance thresholds is a small assignment problem, and settling it on the morning call gives away margin.
The pallet books into the packhouse on Tuesday with fourteen days of shelf life printed on the specification. The planner treats that fourteen as a fact, allocates it to a customer four days away who requires five days of life on receipt, and the arithmetic works comfortably.
What the specification means is fourteen days from harvest at the correct temperature. This lot was cut on Monday afternoon, sat on the headland in the sun until the trailer filled at seven, reached the cooler at nine and came down to temperature around midnight. Somewhere between two and three of those fourteen days were spent before the product had a pallet label, and nothing in the receiving transaction records it.
The customer rejects the delivery on arrival for insufficient remaining life. Everyone downstream of the field did their job, and the loss was created in a six hour window that no system in the chain measures.
The shelf life budget is spent at a rate you control
Deterioration in fresh produce is a set of chemical and biological rates, and rates respond to temperature in a way that has been characterised for decades. The rough working rule is that respiration and the associated quality loss roughly double for each ten degree rise, which is the Q10 convention, and Labuza set out the general application of reaction kinetics to food deterioration in the Journal of Chemical Education in 1984.
That gives you an arithmetic you can actually use. Take a product with a fourteen day life at five degrees. Product sitting at twenty five degrees is ageing at roughly four times the reference rate. Six hours at twenty five therefore consumes about twenty four hours of the fourteen day budget. Two hours of that is the trailer filling and four is the queue at the cooler, and the entire day was lost before the product entered any system that tracks it.
Turn that round and the planning lever becomes visible. Cutting the delay from harvest to precooling from six hours to two returns roughly sixteen hours of shelf life to every pallet, every day of the season. On a chain where the rejection threshold is five days on arrival and a meaningful share of consignments sit within a day of it, sixteen hours is the difference between a claim rate that is annoying and one that is manageable.
The measurement that makes this operational is time and temperature from cut to cooler, by field and by crew. Most operations record the arrival time at the cooler and nothing before it, which measures the half of the interval that is already managed.
Quality is continuous, and the date on the box is a projection
An expiry date implies a switch. Fresh produce declines steadily: firmness, colour, sugar, water loss, and the point at which it stops being sellable is a threshold on a specification rather than an event.
Tijskens and Polderdijk set out a generic keeping quality model for vegetable produce in Agricultural Systems in 1996, in which quality declines according to a temperature-dependent rate and the acceptance limit is a parameter of the customer rather than of the product. That framing is more useful in planning than a date, because it makes remaining life a computed quantity that depends on where the product has been and on who is going to receive it.
The practical version is an accumulated thermal history per lot, converted into consumed shelf life, and a remaining life figure that differs by destination because different customers accept different thresholds. A retailer requiring seven days of life on receipt and a foodservice customer accepting three are looking at the same pallet and seeing two different products.
Building that requires the lot to carry its history, which means the traceability record and the planning record have to be the same object. Where the temperature data lives in a logger system and the inventory lives in a warehouse system with no key between them, the calculation cannot be done, and this is the common case.
Harvest timing trades yield against life
Unusually for a supply chain, the moment of production is a decision made weekly and it has two outputs that move in opposite directions.
Leaving a crop another week generally adds weight and improves size grading. It also advances maturity, and more mature produce has a shorter post-harvest life. On many crops the relationship is steep enough to be worth planning around rather than leaving to the field manager.
Put numbers on a case. Delaying a cut by seven days lifts marketable yield by eight percent, and shortens shelf life from sixteen days to eleven. If the eight percent is worth 4,800 on a block and the shorter life raises downstream loss from three percent to nine percent on a consignment worth 60,000, the loss increase is 3,600. The delay is still slightly positive, and the answer flips entirely if the destination for that block is the eight day export lane rather than the domestic one.
That is the point worth taking. Harvest timing is currently optimised for yield because yield is measured at the farm and loss is measured three parties later. Any operation that can attribute downstream loss back to a harvest date and a block will find the relationship in its own data within one season, and the decision moves from agronomy to planning without anyone needing a model.
Ripening rooms are the only place in the chain where you control the clock
Several important categories arrive unripe by design, and the ripening or conditioning step converts a durable input into a perishable output on a schedule you set. That makes it the most valuable planning asset in the chain, and it usually gets managed as a facility rather than as a decision.
The room is a batch process with a cycle of a few days, a capacity in pallets, and a set of temperature and ethylene regimes that produce different arrival profiles. Filling it is a scheduling problem against a demand forecast three to seven days out, and the constraint is that once a batch is started it cannot be paused for long.
The scheduling insight is that green stock is inventory with a long life and ripe stock is inventory with a very short one, so every room start converts safety stock into commitment. The right buffer sits on the green side, and the room schedule should be driven by orders in hand plus a short forecast rather than by keeping the rooms full. Operations that run rooms to a utilisation target reliably produce ripe fruit on the days demand is low.
The pack decision commits shelf life as well as format
Raw intake is flexible. Once it is packed it is committed to a format, a customer specification and frequently a label, and in prepared lines the packing operation itself accelerates deterioration by cutting surfaces and raising handling temperature.
That gives the packhouse a version of the postponement decision (G4) with an unusual property: waiting costs shelf life at a measurable rate. Hold 40 tonnes of intake unpacked from six in the morning until orders firm at two in the afternoon and you keep format flexibility for eight hours, at a cost of eight hours of ageing if the holding area is cold and considerably more if it is not. Commit at six and you have optionality of zero and a full remaining life.
The way to make that a decision rather than a habit is to price both sides in the same units. Convert the holding delay into consumed shelf life using the same Q10 arithmetic, convert the format commitment into expected mismatch cost using the historical difference between the six o'clock forecast and the two o'clock order book, and compare. Businesses that run this find the answer differs sharply by line: bulk formats with wide customer tolerance should be committed early, and prepared lines with narrow order windows are worth holding, which is the reverse of how most packhouses sequence their morning.
Allocating remaining life across destinations is an assignment problem
At any moment you hold a set of lots with different remaining lives and a set of orders with different transit times and different acceptance thresholds. Matching them is a small optimisation with a hard feasibility constraint, and in most operations it is done by whoever is loudest on the morning call.
Lay it out. Four destinations at one, three, five and eight days transit, each requiring five days of life on arrival. A lot with ten days remaining is feasible for the one, three and five day lanes and infeasible for the eight day one. A lot with fourteen days is feasible everywhere. Send the fourteen day lot on the short lane because it was nearest the door, and the ten day lot has nowhere legitimate to go, so it either travels and gets rejected or it gets marked down at home.
Solving it properly is a transportation problem with a feasibility mask, it runs in under a second on any realistic size, and it recovers value that is invisible in every existing report because a rejected consignment gets recorded as a quality failure rather than as an allocation error. What to do with stock that has no feasible destination left is a separate decision with its own economics (I5).
Nahmias reviewed the perishable inventory literature in Operations Research in 1982, and the structural result that still matters is that policies for perishables depend on the age distribution of stock rather than on the total, which is precisely the information a standard on-hand balance discards.
Where this stops
The thermal history calculation depends on data density and on the loggers actually being read. A single logger on a trailer describes the trailer, and the pallet by the door has a different history from the one in the middle. How much monitoring to install is a real cost question with diminishing returns, and it is the design decision covered elsewhere (X8). What is worth saying here is that the field to cooler interval is usually the largest single gap and the cheapest to instrument, because it happens in one place under your control.
The keeping quality models also need parameters per crop, per variety and sometimes per growing region, and calibrating them takes a season of destructive testing. A business with forty product lines will not calibrate all of them. Calibrate the six that generate most of your claims, use a simple degree-hours proxy for the rest, and accept that the rest are being managed on a rule of thumb.
The harvest timing argument has a limit that is contractual rather than technical. Growers are frequently paid on delivered weight, which means the yield side of the trade lands in their account and the shelf life side lands in yours. No analysis changes that behaviour, and the businesses that have moved the decision have generally done it by changing the payment basis, which is a commercial negotiation the planning team can inform and cannot conduct.
The FAO estimated in its 2019 State of Food and Agriculture that around fourteen percent of the world's food is lost between harvest and retail. A meaningful share of that sits in the hours this article is about, and almost none of it is currently attributed to anybody.
Record cut time and cooler arrival time for a fortnight, convert the gap to consumed shelf life at a Q10 of two, and compare that number against the life you are promising your customers.