Blog
A customer asks which material went into this batch. How many hours until you can answer?
Traceability is not a paperwork question. It is a question about how wide the damage gets when something goes wrong.
Published
The customer calls on a Monday morning. Parts shipped two months ago have a problem. They want to know which supplier's material went into the affected lot, when it arrived, which machine ran it, on whose shift, and where else material from that same batch was shipped.
A plant that can answer within two hours recalls only the lots concerned.
A plant that cannot recalls everything shipped in that window, because there is no way to prove which lots are unaffected.
The difference between those two outcomes is not manufacturing quality. It is whether the data exists.
Why traceability is usually incomplete
Most plants hold the data in segments. Goods receipt sits in purchasing. Production records sit in a folder at the machine or in a shift leader's spreadsheet. Inspection results sit in the quality system. Despatch notes sit in sales.
Every segment is accurate. None of them shares a key with the others, so joining them is manual work done by whoever remembers what was happening at the time.
When that person leaves, the ability to trace leaves with them.
What a traceability system actually needs
An identifier that travels with the material. The material lot has to be recorded as it goes into the machine, not written up at the end of the shift from memory. Scanning at the machine works, and takes seconds per event if it is designed properly.
Recording at the moment of the work. Every time recording happens later, accuracy drops. Not because anyone intends to be wrong, but because that is how memory works.
A link to inspection results. Results tied to a lot let you establish whether a problem was visible in the plant or emerged after despatch, which shapes both the fix and the conversation with the customer.
Search that is fast enough. A system holding complete data that takes two days to query is not meaningfully different from having none. In an incident, speed of answer is the value.
The benefit nobody planned for
Plants that build traceability to satisfy a customer requirement usually find the same data answers other questions.
Which supplier's material correlates with higher scrap. Which machine produces more defects than its identical neighbour on the same job. Which shift has recurring trouble. All of it sits in the same dataset, and none of it was visible while the data lived in four places.
Where AI fits
Plainly: traceability is not an AI problem. It is a problem of capturing data correctly and joining it.
AI has a place afterwards. With several years of joined-up production data, finding relationships between process parameters and scrap rates is something a model does better than a person. But the data has to exist first.
Proposing AI to a plant still recording lots on paper is selling something that cannot work.
What GIPSIC does here
We work with industrial manufacturers, muRata among them, and build systems that join data at the machine to systems in the back office.
We will raise the awkward point when there is one. If we can see that the place the data will break is an extra scan asked of operators during an already pressed shift, we will say so before we start, and set out the alternatives and what each one costs.
If you would like to talk about your line, get in touch.
About the author
Film — Wisit. A businessman who still does his own BA work more often than he probably should, and writes a fair bit of code, front and back. Runs two or three small businesses. Follows technology and business obsessively, in Thailand and everywhere else. Off the clock: physics, astronomy, DIY, and anything to do with networks. Music always on, though he cannot sing. Plays instruments anyway, badly. Plays a lot of sport, racket sports above all. Not much of a traveller by himself, but happy to take Mint anywhere in the world. A man who fears — sorry, loves — his wife. One flaw: he barely touches video games.
Written with Claude Opus 5