A food manufacturer carries every traceability obligation a chemical or cosmetics plant does, and three more: allergen control, dated shelf life, and the ability to reconstruct a batch's full history quickly enough to matter during a recall. The first two shape how you record things every day. The third only gets tested when it is too late to fix.
One step forward, one step back
Most food safety schemes ask for the same core capability, described roughly the same way: for any batch you produced, you can identify the suppliers and lots of everything that went into it, and the customers who received it.
Backwards is usually the easier half, because bills of materials and production records already point that way. Forwards is where plants struggle, because it requires dispatch records tied to lots rather than to products — knowing not just that you shipped 400 cases to a distributor, but which lots were in them.
If your dispatch paperwork records product and quantity but not lot, you have half the chain. The half you are missing is the half a recall needs.
Allergens change what a changeover means
In a plant handling allergens, a changeover is not only a cleaning task, it is a record. What ran before, what cleaning was done, who verified it, what ran after. That sequence has to be reconstructable per batch.
Two practical consequences for how you set up records:
- Production order sequence has to be stored, not just planned. What you intended to run on Tuesday is not evidence. What actually ran, in what order, is.
- Cleaning verification belongs on the batch record, not in a separate logbook, because during an investigation you need them together and time-ordered.
Scheduling to minimise allergen changeovers is also the same problem as scheduling to minimise colour changeovers in soap — group like with like, and sequence from least to most. Scheduling around the bottleneck covers the general case.
Shelf life is a date, not a duration
The distinction matters more than it sounds. A raw material with a twelve-month shelf life is not useful information; the date it expires is. And the date that governs your finished product is derived from the shortest-dated input, not from the production date.
That derivation has to happen automatically, at the point of batch creation, from the lots actually consumed. Calculated by hand it will be right most of the time, and the exception will be the one that ships.
Expiry-first picking follows from the same data. It should advise rather than block — an operator who can see the system is asking for the older lot, and can override with a reason when the older lot is on the wrong side of the warehouse, will use the system. One that simply refuses gets worked around within a week.
The mock recall is the only real test
Every scheme asks for one and most plants treat it as an audit chore. Run it as a genuine test instead, unannounced, and time it.
Pick a finished lot from six weeks ago. Without warning anyone, ask for:
- Every raw material lot that went into it, with supplier and receipt date
- The quality results that released it
- Every customer who received any part of it, with quantities and dates
- What else was made from the same raw material lots
Time how long it takes and count how many people and systems were involved. Four hours and three people is a plant that will struggle on the day. Twenty minutes from one screen is a plant that will not.
Then repeat it for a lot that is still in stock, and check the physical quantity matches the system. The gap between those two numbers is the one that turns a targeted recall into a total one.