In short: A temperature excursion changes whether stock can be sold without moving it, so the status has to be written against the batch at detection, in the field the availability calculation reads, rather than at disposition days later. Mean kinetic temperature, derived by Haynes in 1971 and used in the ICH Q1A(R2) guideline and USP chapter 1079, condenses a logger trace into one number that can sit beside the printed date as a planning figure. FEFO with a tolerance window recovers most of the write-off benefit while giving back most of the picking efficiency that strict FEFO costs. The disposition option set decays with the product, so pre-agreed rules for the common excursion classes are worth more than a faster investigation.
The reefer arrives at 06:40 and the logger download shows four hours above 8 degrees during a border delay two nights earlier. The receiving team flags it, QA opens an investigation, and the batch goes on hold in the quality system by mid-morning. In the ERP, the pallets book in as unrestricted stock at 07:15 and appear in available to promise for the rest of the week.
On Wednesday a planner commits that stock to a customer. On Thursday QA closes the investigation with a decision to downgrade, and somebody has to call the customer back. Nothing about the temperature was mishandled. The excursion was detected, investigated and dispositioned properly, and it still produced a broken promise, because the event lived in the quality system and the consequence lived in the planning one.
Most cold chain management programmes are built around proving that the temperature was controlled. The harder question is what the deviation did to your available stock, and how quickly the rest of the operation found out.
An excursion is a stock movement your planning system never sees
The physical inventory did not move. It is in the warehouse, it is on the count, it is in the on-hand balance. What changed is whether it can be sold, and the systems that know that fact are usually not the systems that calculate availability.
The gap is nearly always a timing one. Blocked and quarantine statuses exist in every serious ERP. The problem is that the status originates in a QMS or a LIMS running on its own clock, gets set by a person after a review, and reaches the planning record hours or days later. During that window the availability figure is wrong in the most dangerous direction, showing stock that is not sellable.
Two changes fix most of it. The excursion event has to write a status against the batch in the same field the availability calculation reads, at detection rather than at disposition, with a provisional hold that a later decision releases or confirms. And the event has to carry a quantity and a location, so the rest of the network can net around it while the investigation runs.
The second half matters more than it sounds. A batch going on hold is unplanned demand on everything else. If it was covering a customer commitment, that commitment now needs cover from somewhere, and the question of whether another site can supply it is a rebalancing decision (I4) that should start on the morning of detection rather than after the disposition meeting. What the promising logic does with a provisional hold is its own subject (I7), and it is worth checking which behaviour yours has, because some configurations treat a provisional block as available.
Remaining shelf life follows the temperature history
An expiry date printed at manufacture is a prediction made under an assumed storage profile. Product that spent thirty hours at 12 degrees has less life remaining than the label claims. Product held steadily at 2 degrees may have more, though you will never be allowed to sell it on that basis.
The kinetics behind this are old and settled. Most degradation reactions follow Arrhenius behaviour, and the usual rule of thumb puts the rate increase somewhere between two and three fold for each 10 degree rise, with the exact factor depending on the reaction. Taoukis and Labuza set out the practical version in the Journal of Food Science in 1989 when they analysed time-temperature indicators as shelf life monitors, and the same arithmetic underlies accelerated stability testing across food and pharmaceuticals.
Pharmaceutical distribution already has a summary statistic for this. Mean kinetic temperature is the single equivalent isothermal temperature that would produce the same degradation as the observed profile, derived by Haynes in 1971 and used in the ICH Q1A(R2) stability guideline and in USP chapter 1079 on storage and transport. It condenses a whole logger trace into one number that means something.
What almost nobody does is feed that number back into an inventory record. Batches carry a printed date and nothing else, so two batches with the same date and completely different thermal histories are interchangeable to every downstream system.
The workable arrangement keeps two numbers per batch. The printed date is the regulatory commitment and the only thing you can sell against. The computed remaining life, updated from logger data, is a planning number used for sequencing, for allocation and for deciding what to move first. Plan against the shorter of the two. Never extend a date because the model says the batch ran cold, since that direction requires stability data rather than an estimate.
First expiry first out, and what it costs on the floor
FIFO ships in receipt order. FEFO ships in expiry order. They diverge whenever receipts arrive out of expiry sequence, which happens with variable production dates, multiple sources, returns, and after any excursion recalculates a batch's remaining life. In a temperature-controlled range that is most of the time.
FEFO reduces write-off, and it also protects a second position that is easier to miss. Retail and hospital customers enforce minimum remaining shelf life at receipt, often as a share of total life or a contractual number of days, and a rejection at the door costs the product, the freight in both directions and the service failure on top. Shipping a batch that clears the threshold by a week while a shorter-dated one sits behind it converts a recoverable position into a rejection later.
The cost lands on the pick face. FEFO forces batch-level location discipline, which means more locations per SKU, mixed-batch pallets either avoided or tracked at case level, and pickers losing the freedom to take the nearest available pallet. Travel per line goes up, and the forward pick location has to hold a specified batch rather than any batch of that SKU, which constrains the slotting problem noticeably (X5).
The middle ground most operations settle on is FEFO with a tolerance. Enforce expiry order at the location level, but let the picker take any pallet whose expiry falls within a set window of the earliest available. A window of a few days recovers most of the write-off benefit while giving back most of the picking efficiency. Set the window per product family from the actual spread of expiry dates in stock, and tighten it for the products where customers enforce receipt thresholds hardest.
The disposition decision, and how fast it decays
When an excursion is confirmed there are four outcomes: release, downgrade, quarantine pending testing, and destroy. Each carries a value and a cost. Testing costs money and time, and the product continues to age during it. Downgrading realises a lower price, or redirects the stock to a channel with a shorter remaining-life requirement. Destruction has a disposal cost and sometimes a reporting obligation.
Structurally this is an expected value calculation over the probability that the product still conforms, and it runs inside a hard filter. Where the excursion falls outside the validated stability data or the terms of the marketing authorisation, the economics are irrelevant and the answer is fixed. Freezing damage to adsorbed vaccines is the clearest case: the damage is irreversible, invisible on inspection, and treated as a discard condition, which is why WHO publishes a shake test for suspected freeze exposure. Matthias and colleagues reviewed freezing in the vaccine cold chain in Vaccine in 2007 and found exposure documented in both storage and transport across the studies they examined, in high-income and low-income settings alike, so this is not a rare edge case.
Speed is the variable most operations underuse. The option set decays with the product. A batch that could have been downgraded and sold at day one has less remaining life at day ten, and an investigation that takes a fortnight converts a downgrade into a write-off without anyone deciding to write it off. The fix is to pre-agree disposition rules with QA for the common excursion classes, so routine cases clear automatically against the documented allowable limits and only the unusual ones queue for a person. That queue is an exception queue with all the properties one needs, including a decide-by time (X7).
Many pharmaceutical products already carry the right structure for this in the form of a documented cumulative time out of refrigeration. The batch has a budget, each excursion spends part of it, and the remaining budget travels with the batch. Treating it as a depleting resource per batch, visible to whoever is allocating stock, converts a series of isolated compliance events into a running balance somebody can plan against.
Monitoring density is a value calculation
The choice runs from one logger per shipment, through one per pallet, to one per case, and then to connected telemetry reporting in transit. Each level buys something different, and the differences are quantifiable with your own numbers.
Per-shipment monitoring tells you a shipment had an excursion without telling you which pallets experienced it, which usually forces you to hold the whole shipment. Per-pallet monitoring lets you release the pallets that stayed in range, converting a whole-shipment problem into a partial one.
Run the arithmetic. A shipment of 20 pallets at 4,000 each carries 80,000 of stock. If shipment-level monitoring holds all 20 while pallet-level monitoring would have held 4, the difference in exposed value is 64,000 for that event. The honest input is expected loss rather than full value, since holds often end in release or downgrade, so take a third of it and call the avoided loss 21,000 per excursion. At one excursion per 200 shipments that is about 107 of expected saving per shipment, against 300 for twenty single-use loggers at 15 each. Single-use loses on those numbers. Reusable loggers amortised over twenty trips cost nearer 60 a shipment and win comfortably, and the answer flips again on a lane running one excursion in 60, where even single-use pays for itself.
Real-time telemetry earns its premium differently. A logger read at receipt can only classify a loss that already happened. An alert at hour one of a reefer failure, reaching a driver who can divert to a cross-dock, prevents it. That value only exists if somebody is on the other end with the authority to act and a route to act through, so the case for telemetry rests on the response capability sitting behind it (X7).
Before adding sensors anywhere, map the storage areas. WHO published a technical supplement on temperature mapping in 2014 alongside the model guidance in TRS 961 Annex 9, and the reason it comes first is that a probe in the wrong place produces confident wrong data. Hot spots near doors and lights and cold spots near evaporator discharge are ordinary, and a chamber monitored at a single convenient point can pass continuously while a corner of it sits out of range.
Where this stops
The regulatory constraint is a hard filter and nothing here overrides it. EU Good Distribution Practice guidelines (2013/C 343/01), WHO TRS 961 Annex 9 for time and temperature-sensitive pharmaceutical products, and the FDA sanitary transportation rule under FSMA for food all define what may be released and under what evidence. Where release is not permitted, no expected value calculation makes it permitted. The optimisation lives strictly inside the permitted set, and its job is to find the best option among the ones compliance allows and to find them faster than a manual process would.
The kinetic estimate has its own limits. A remaining-life model is only as good as the stability data behind it, and for a product with one accelerated study and no real-time data at the temperatures actually experienced, the extrapolation is weak. Use it to prioritise what moves first and what gets tested. Do not use it to argue with a printed date.
Logger data quality is the quiet failure. Devices get activated late, placed on top of a pallet in the airflow rather than inside the load, left in the cab, or read incorrectly at receipt. A trace from a badly placed logger produces a mean kinetic temperature that is precise and wrong, and it will be believed because it has decimal places. Audit placement practice before trusting any of the downstream arithmetic. The equivalent disposition economics for ambient goods run on different constraints and are covered separately (I5).
Take the last twelve months of excursion records and measure the hours between the logger read and the stock status changing in the system your planners look at, because that gap is where the avoidable losses are.