A plant that cannot say which batch went to which customer should not be spending money on automation. Not because automation is wrong, but because the first serious question it will be asked — which units are affected? — is one it still cannot answer, only now faster and at greater volume.
Traceability is the least exciting thing a manufacturing business can invest in and the one that determines what every later investment is worth. It is also, in most mid-market plants we have looked at, the thing that exists on paper and nowhere else.
The recall question
Here is the test, and it is worth running as a drill rather than a thought experiment.
A customer reports a defect. It traces to a raw material lot received eleven weeks ago. Which finished units contain material from that lot, where are they now, and which customers have them?
Answer it with a stopwatch running. In most plants one of three things happens. Somebody goes to a cupboard. Somebody calls a supervisor who remembers roughly when that drum was opened. Or the honest answer emerges: we would have to recall everything from that window, because we cannot narrow it.
That last answer has a price, and it is computable. It is the difference between recalling four hundred units and recalling nine thousand — a number most plants have never calculated, because the day they needed it they were too busy to do arithmetic.
The genealogy that has to exist
Traceability is one unbroken chain of links, and it breaks at whichever link is kept only on paper.
| Link | What it must record | Where it usually breaks |
|---|---|---|
| Receipt | Supplier lot number against your internal lot | Recorded on the GRN, never entered anywhere queryable |
| Issue | Which lot went to which work order, and how much | Issued in bulk; the lot identity is lost at the store |
| Production | Which input lots made which output batch | Shift log on paper, in a folder, by date not by batch |
| Quality | Test results tied to the batch, with the release decision | Lab keeps its own register, keyed differently |
| Dispatch | Which batch, in which consignment, to which customer | Invoice records quantity and grade, not batch |
The last row is the one that undoes all the others. A plant can maintain immaculate internal genealogy and still be unable to answer the recall question, because at the moment of dispatch the batch identity was dropped and only a quantity and a grade were recorded. Everything upstream becomes unusable for the purpose it was kept for.
If the full chain is out of reach this year, build the last link first: record batch numbers on the dispatch document. It is a small change, it is usually the least automated step, and it converts “recall everything from that window” into “recall these consignments”.
Traceability pays back from the customer end inwards, not from the raw-material end outwards. Most projects are sequenced the other way and run out of money before reaching dispatch.
Paper is not the problem. Paper that is never keyed is
Shop-floor paper is durable, cheap, works when the network does not, and survives conditions that kill tablets. The failure is not that records are on paper. It is that the paper is filed rather than entered, so the information exists but cannot be queried — which for a recall is the same as not existing, because you have hours and the answer is in a cupboard organised by date.
The realistic pattern is to keep paper where it works and add one keying step at a natural pause — end of shift, or at the quality gate — so the data lands in a queryable record while the paper stays as the physical backup. This is less satisfying than a fully digital floor and it is what actually gets adopted, because it does not require the operator to change what they do while they are doing it.
Why this must come before automation
Three reasons, in ascending order of how much they cost to learn the hard way.
Automation without traceability multiplies the blast radius. A faster line produces more units per hour from the same undifferentiated pool. When something goes wrong, the quantity you cannot narrow down is larger.
You cannot improve what you cannot attribute. The case for automating a step rests on knowing that step’s scrap rate, cycle time and rework. If output is not tied to inputs and process conditions, those numbers are plant averages — and plant averages hide exactly the variation the investment was meant to remove. Automating on an average is how a business spends heavily and moves its aggregate figures by nothing.
Customers are increasingly making it a condition. Supplying into automotive, pharmaceutical, food or export chains brings lot-level traceability requirements with audits behind them. A plant that cannot demonstrate genealogy is not a plant with a compliance gap; it is a plant that is not eligible for that customer.
What a traceable plant looks like from the inside
It is undramatic. Every lot received gets an internal identifier at the gate. Issues to production record lot and quantity. Each batch records its input lots, its operator, its shift and its process conditions. Quality results attach to the batch and the release decision is a state change, not a signature on a form. Dispatch records which batch went in which consignment to which customer.
Nothing there is technically difficult. The difficulty is organisational — it requires the store, the floor, the lab and dispatch to agree on one identifier for a physical thing, when today each of them has their own and translates at the boundary.
That agreement is the actual deliverable. The software is the easy half, and it will not survive the absence of the agreement. Which is why the first two weeks of this kind of work are spent in the plant with the four people who each have a different name for the same drum, getting them to pick one.