Traceability is one of those requirements that feels like paperwork until the day it does not.
The day it matters, the question is narrow and urgent: a defect has been found, and you need to know which batch it came from, what went into that batch, and where it went. How quickly and precisely you can answer determines whether you recall one lot or everything shipped that month.
The difference between those two outcomes is usually not a compliance system. It is whether a few links in the chain were recorded at the time.
The two directions
Traceability runs both ways, and they answer different questions.
Backwards — from a finished item to its inputs. A customer reports a problem with a specific unit. Which batch was it, which raw materials went into it, from which suppliers, on which date, on which machine, with which operator?
Forwards — from an input to everything it touched. A supplier notifies you of a problem with a delivered lot. Which of your batches used it, and which customers received those?
Most small manufacturers can partially answer the first with effort. Very few can answer the second at all — and the second is the one that determines recall scope, because it defines the boundary of what is affected.
The links that have to exist
Traceability is a chain, and it is only as good as its weakest link. Four have to be recorded, at the time:
Goods in, by lot. Material received, recorded against a supplier lot or batch reference. If incoming material is recorded only as a quantity, the chain has no starting point and nothing downstream can be traced back.
Consumption against a work order. Which lots were issued to which job. This is the link most commonly missing, because material is issued informally — taken from the store as needed. Without it, backwards traceability stops at "some of the material we held that week".
Output identified by batch. Finished goods carry a batch or lot identifier that ties back to the work order.
Dispatch by batch. Which batch went to which customer. Without this, forwards traceability fails at the last step, and every customer becomes potentially affected.
Break any one link and the chain gives you a range rather than an answer — which in practice means recalling more than necessary.
Why the consumption link is usually the broken one
Recording which lots went into which job requires action at the moment material is issued, by someone who is trying to keep production running.
This is the same design constraint that governs a busy counter: if recording the truth is slower than not recording it, the record will be optimistic. In practice that means issuing has to be scanning rather than writing, defaults have to be right most of the time, and the awkward cases — a partial lot, a substitution, a top-up mid-run — must be as fast as the normal path. The principle is in how Truffaire builds software.
It also connects to something with immediate commercial value rather than only compliance value: recording consumption against jobs is what makes yield measurable. Factory operations: what to digitise first covers that argument — which is worth knowing, because a traceability system justified only by compliance tends to be maintained grudgingly, while one that also shows material variance gets used.
Batch size is a business decision
An under-appreciated point: your batch definition determines your recall exposure.
Large batches are efficient and mean a defect implicates more product. Small batches cost more in changeovers and record-keeping, and confine problems.
Most manufacturers set batch size purely on production efficiency, without considering that it is also setting the size of the worst-case recall. It is worth deciding deliberately rather than inheriting.
What "recall-ready" actually means
The useful test is not whether you hold records. It is how fast you can answer, under pressure, on a day when this is not the only thing happening.
A practical drill: pick a finished item shipped three months ago. Without warning anyone, establish which batch it came from, which raw material lots it used, which suppliers those came from, and every other customer who received that batch.
If that takes hours of assembling paperwork, you are not recall-ready — you have records. The distinction only becomes visible when it is tested, and most operations test it for the first time during an actual incident.
What we would say honestly
Truffaire has delivered production monitoring and factory operations systems, so the structure above is work we build. We have not published case studies in regulated food or pharmaceutical manufacturing, and we will not imply sector-specific compliance experience we cannot substantiate.
What we can say is that the pattern is consistent: traceability fails at the consumption link, and it fails because recording it competes with running production. Systems designed as compliance documentation get filled in retrospectively, which produces records that satisfy an auditor and cannot survive an actual investigation.
The version that works captures the same data as a by-product of issuing material — which requires the system to be fast at the point of use, and requires the operation to get something back from it beyond compliance.
Frequently asked questions
Do we need barcodes for this?
Something that makes recording fast. Barcodes or QR codes on material lots and work orders remove transcription errors and make issuing quick enough to be done honestly. Handwritten lot numbers copied under time pressure are where chains break.
How long should we keep records?
Determined by your sector's obligations and product shelf life, and worth confirming specifically rather than assuming. Storage is cheap; discovering you discarded records you needed is not.
Is this only relevant to food and pharma?
Those have the strictest obligations, but any manufacturer facing a defect benefits from a narrow recall rather than a wide one. Automotive, electronics and components all have real exposure.
What if we make continuous product rather than batches?
You define time-based lots — a shift, a run, a period. The principle is unchanged: a bounded quantity with identifiable inputs.
Can we manage this on paper?
For a very small operation, yes, and it will be slow to query. The failure mode of paper is not accuracy, it is retrieval speed — which is exactly what matters on the day it is needed.
Where to start
Run the drill. Pick something you shipped three months ago and trace it, both directions, timed.
Whatever step takes longest is your weakest link, and it is almost always material consumption against jobs. Fixing that single link improves both your recall position and your understanding of yield — which is why it is worth doing even before any obligation requires it.
If you want a read on where your chain breaks, get in touch.