Traceability: deciding how narrowly you could bound a problem

Granularity is a commercial decision, not a technical default

The chain can be held at individual item level, at batch level, at a shift or a day of production, or not at all beyond the delivery note. Each step finer reduces exposure when something goes wrong and adds marking, recording and handling effort to every unit made. The way to choose is to price the alternative: estimate the quantity you would have to reach if the boundary were coarse, and what reaching it would cost in replacement, freight, customer disruption and reputation. For low-consequence goods the coarse answer is right; for anything whose failure injures somebody, it is not.

Backwards to inputs, forwards to customers — and both must work

Two directions are needed and they fail independently. Backward traceability answers what went into this item: which material lot, which sub-assembly, which tool, which settings, which operator. Forward traceability answers where this input went: given a suspect material lot, which finished items contain it and which customers received them. Plants often build the backward chain well, because it follows the production record, and neglect the forward one, because it requires the dispatch and delivery records to be linked to the same identifiers. The forward gap only becomes visible during a containment, when it is expensive.

The process has to preserve identity, not just record it

Traceability places real constraints on how work is done. Material cannot be tipped from several containers into one hopper without merging lots. Parts cannot be left unmarked between operations, or moved between trolleys, or returned to stock after a rework without their history. Cleaning down between batches becomes a traceability step as well as a contamination one. Where a process physically destroys identity — bulk melting, continuous mixing, plating baths — the honest position is that the lot begins there, and the chain upstream describes inputs to the batch rather than to the item.

Where the chain quietly breaks

Failures cluster at the same handful of points: rework and repair loops that put an item back into flow without recording what was replaced, parts scrapped and substituted at assembly, samples pulled for testing and returned, bulk hardware and consumables used untracked, and packing operations that combine several production lots into one shipment without recording which. Add supplier deliveries where the certificate references a lot number that does not appear on the containers. Testing the chain periodically — picking a shipped item and walking it backwards, then picking a material lot and walking it forwards — finds these long before an incident does.

What customers and downstream users actually require

Requirements arrive from several directions and rarely agree: a customer contract specifying identification and record retention, sector expectations for goods with safety consequences, and your own need to bound a problem. The prudent approach is to design for the strictest genuine requirement across the product family rather than maintaining several regimes, because mixed practice on one shop floor reverts to the loosest. Whatever level is chosen has to be visible in the work instructions and enforced at the line, since traceability maintained by the office and ignored by production is only a record of intentions.

Frequently asked questions

How do we prove our traceability works without waiting for a real incident?

Run a drill. Pick a finished item at random, then require the team to produce its inputs, process records and release evidence within a stated time. Then reverse it: nominate a material lot and ask which customers received product containing it. Record how long each took, what could not be found, and where the chain broke. Repeating this a few times a year, with different products and different people, is the only honest test.

Is item-level traceability worth it for low-value components?

Rarely on value alone, and sometimes on consequence. A cheap fastener in a structural joint, or a small moulding in a medical assembly, can carry recall exposure out of all proportion to its price. The question to ask is not what the part costs but what it would cost to reach every assembly containing an uncertain batch of them. Where that number is large, the marking discipline pays for itself the first time it is needed.

What identifier should we use to tie everything together?

One that survives the process physically and appears on the paperwork the customer will quote back at you. A code that is laser-marked or moulded in survives handling, washing and packing; an adhesive label may not. Whatever is chosen, it must be present on the container, the production record, the inspection result and the dispatch document, because a chain that uses different references at different stages requires a person to reconcile it under time pressure.