
Knowledge AI series 7/7 · (1) How Apparel Orders Are Split Between Knowledge Services and Manufacturing · (2) Where Information Changes Hands in the Supply Chain Workflow · (3) Why the Same Order Number Is Entered Again and Again Until Shipment · (4) Why Production Data Measured on the Floor Gets Lost · (5) Why CM Is the Only Shared Variable in FOB Cost · (6) Why Vendor and Factory Calculate the Same CM in Opposite Directions · (7) How Measured Values Become a Company Asset
KEY POINTS
When an order is finished, the process knowledge gained along the way stays in people’s memories, mailboxes and personal folders, and it leaves with them when they move on. For records to become a company asset, they have to go through four stages: measure values on the floor as they are, standardize the same things under the same names and attributes, define the relationships between operations, materials, machines and people, and reuse those relationships for sourcing and quoting the next new order. The tool used in the third stage is an ontology.
A month has passed since WJ-2411 shipped. A new woven jacket style for next season lands on the vendor’s costing manager’s desk. Let’s call it WJ-2502 (another made-up number). The fabric and the details are similar to WJ-2411. The costing manager opens the quotation from the last order. The CM and efficiency are written down, but not the reasons behind those numbers.
Why two operators were put on the front pocket operation, which line had puckering (a defect where the seam gathers and wrinkles) with that fabric, and which attachment was changed as a result: the merchandiser who handled WJ-2411 and the factory’s IE engineer remember all of this. But the merchandiser moved to another team last week. In this last post of the series, we go through the four stages that keep measured values as a company asset instead of in one person’s memory.

Where process knowledge is kept
The philosopher Michael Polanyi started from the fact that “we can know more than we can tell” and argued for rethinking human knowledge.1 The management scholar Ikujiro Nonaka called this kind of knowledge tacit knowledge, and explained that because it is personal, it is hard to formalize and pass on to others.2 Knowledge that has been put into documents and procedures, on the other hand, is called explicit knowledge.
Most knowledge on a sewing floor is tacit. How many seconds an operation takes, and which fabric caused problems on which line, is kept in people’s memories. What they remember shows up in quotations and work orders only as results, and the reasons behind the decisions are not recorded. This is where that knowledge is kept today.
| What builds up | Where it is kept today |
|---|---|
| Measured time by operation | People’s memories |
| Difficulty by style | Personal folders |
| Machine and operator performance | Month-end reports |
| Problems and fixes | Messenger chats |
Knowledge built up with each order, and where it is kept today
Knowledge lost when employees leave is already treated as a serious business risk. One five-year longitudinal study reported that the gaps left by knowledge loss are not easily filled even over time.3 That is why, when the person in charge changes, even a familiar style has to be worked out again from scratch.
Four stages that turn records into assets
Having records does not automatically make them an asset. A monthly report is a record, but it cannot be used again to quote the next order. The data management field treats data as an information asset, and defines data management as the work of delivering, controlling, protecting and enhancing the value of data and information assets throughout their life cycles.4 A record has to be reusable before it can be counted as an asset.
SIJE sees the process of turning measured values into assets as four stages.
| Stage | What it involves | Example on a sewing floor |
|---|---|---|
| Measure | Collect values from lines, operations and machines as they are | Cycle time by operation, hourly output |
| Standardize | Organize the same things under the same names and attributes | One name for an operation that each factory called something different |
| Relate | Define the relationships between operations, materials, machines and people | This fabric caused this defect on this machine |
| Reuse | Use the relationships built up for sourcing and quoting the next new order | Starting point for quoting a similar style |
The four stages from measuring to reusing have to be put in place in order, starting from the first
The first stage, measuring, was covered in the earlier post Why Production Data Measured on the Floor Gets Lost. Only values measured for every cycle on the same basis can move on to the next stage. This post starts from the second stage.
Standardize: calling the same thing by the same name
Even when measured values are collected, they cannot be compared if the names are all different. If the Korean head office writes “밑단 말아 박기”, the English operation breakdown says “Hemming Bottom”, the Indonesian factory writes “Heming Bawah” and the Vietnamese factory writes “May gấu đáy”, there is no way to put the operation times from the four sites side by side. The UN Statistics Division’s data interoperability guide also explains that data can be cleanly compared and combined only when there are standard classifications.5
Three concepts are used in standardizing. In the terms of the W3C, which sets standards for data on the web, a class is a group of things that have something in common, and each individual thing that belongs to that class is an instance. The characteristics of a thing, or the relationships between things, are called properties.6
A cookie cutter makes this easier to picture. The cutter is the class, and each cookie cut from it is an instance. Whether a cookie is chocolate or vanilla, and how many grams it weighs, are properties.
| Concept | Definition | Sewing example |
|---|---|---|
| Class | A group of things that have something in common | Hemming Bottom operation |
| Instance | One thing that belongs to a class | The Hemming Bottom operation on WJ-2411 |
| Property | A characteristic of a thing, or a relationship between things | Standard time, machine used, operators assigned, attachment |
Organized into classes, instances and properties, the same operation can be compared side by side across different factories
The industry has made efforts like this before. GSD, a standard time system for apparel, defines sewing motions as 39 motion codes and combines them to describe thousands of operations. Whoever measures, the same code gets the same time value.7 Bringing the four names above together under one entry is what SIJE’s sewing process glossary does. We have built it so that the same operation is called by the same name, and it now covers 2,186 entries in five languages: English, Korean, Indonesian, Vietnamese and Japanese.8
Relate: connecting operations, materials, machines and people
Standardizing gives you a well-organized dictionary. But a dictionary explains words one at a time and says nothing about how they relate to each other. Knowledge such as “this fabric puckered on this machine, so we changed to this attachment, and efficiency on this line went up” is a relationship between several things.
The tool for defining these relationships is an ontology. The computer scientist Thomas Gruber defined an ontology as “an explicit specification of a conceptualization.” A conceptualization here means an abstract, simplified view of the world you want to represent for some purpose.9 Later, Studer and colleagues added the word “shared” to this definition, meaning that an ontology should be knowledge agreed on by a group of people.10
Put simply, if a dictionary is a word list, an ontology is a subway map. A list of station names does not tell you where to change trains, but with a map you can see at a glance how the stations connect. The properties mentioned above also come in two kinds: object properties that link one thing to another (fabric is used in an operation), and data properties that attach a value to a thing (the standard time of an operation is so many seconds).6

The difference between a table and relationships shows up on the next order. In a table, you have to find the WJ-2411 row to see the information. With relationships, you can find WJ-2411 starting from “this fabric”, “this machine” or “this defect”, and you can find other orders that used the same fabric at the same time. Industrial ontologies like this are also being set out in international standards. ISO/IEC 21838 sets the requirements for top-level ontologies that connect ontologies from different fields,11 and the Industrial Ontologies Foundry (IOF) is building open reference ontologies covering all areas of digital manufacturing.12
SIJE has built an ontology that organizes apparel manufacturing processes this way, at a scale of more than 470,000 classes.13 Because it defines how operations, materials, machines and people relate to each other, products, factories and countries can be compared on the same basis even when they differ.
Reuse: knowledge as the starting point for the next order
Knowledge organized as relationships can be used again on the next order. When the new jacket from the start of this post (WJ-2502) comes in, past styles with similar fabric and details can be found through their relationships, and those styles’ actual labor minutes, efficiency, performance, and problems and fixes can be used as the starting point for the quotation.
| What builds up | What it does on the next order |
|---|---|
| Measured time by operation | Becomes the basis for standard times and line targets |
| Difficulty by style | Becomes the starting point for quoting similar styles |
| Machine and operator performance | Becomes the reference for line setup and operator allocation |
| Problems and fixes | Keeps the same defect from coming back |
How the knowledge in the earlier table is used on the next order
The costing manager can quote from the values built up, without having to search their memory. The more orders build up, the more accurate that basis becomes, and it stays with the company even when the person in charge changes. The style and factory properties covered in the process analysis series also need to be organized as relationships like this before they can be used again in the next quotation.
The fifth layer: an AI agent
On top of the knowledge that has become an asset through the four stages, one more layer can be built. It is an AI agent that makes decisions based on productivity, quality, cost and delivery (PQCD) and carries those decisions through to action. The agent can only base its decisions on this evidence once measured values have been standardized and organized as relationships. SIJE is currently preparing this agent for launch. We will share its name and full features at launch.
Wrapping up the series
Over seven posts, we followed a single order as it moved forward. We started with how the buyer, the vendor and the factory divide the work, then covered the points where people move information by hand, why the order number is entered again in each tool, why numbers from the floor are not used for the next order, what CM means on the cost sheet, and the different assumptions the vendor and the factory use when calculating CM. The problems in all seven posts have the same cause. The three companies write the same numbers from the same order in separate places, and cannot use those numbers again for the next decision.
In the first post of the series, the buyer was deciding which vendor to give the order to, and the vendor was deciding which factory to place it in. The basis for decisions like these should sit in the company’s records rather than in one person’s memory. Even when people change, the company’s basis for making decisions should stay.
How SIJE handles this
SIJE’s products each take on part of the four stages. Monolog measures time and output by operation on the line, and the sewing process glossary and the ontology standardize those values and define the relationships between them. In Monolis, when you upload a work order and TechPack, AI builds the operation breakdown, then simulates CM by factory based on the relationships built up and reflects it in the cost sheet. You can see this process with screenshots in Process Analysis Is Cost Design: Six Steps Where AI Builds the CM.
Knowledge AI series
1. How Apparel Orders Are Split Between Knowledge Services and Manufacturing
2. Where Information Changes Hands in the Supply Chain Workflow
3. Why the Same Order Number Is Entered Again and Again Until Shipment
4. Why Production Data Measured on the Floor Gets Lost
5. Why CM Is the Only Shared Variable in FOB Cost
6. Why Vendor and Factory Calculate the Same CM in Opposite Directions
7. How Measured Values Become a Company Asset (this post)
References
- Michael Polanyi, The Tacit Dimension, University of Chicago Press, 1966, p.4. “we can know more than we can tell”. ↩
- Ikujiro Nonaka, A Dynamic Theory of Organizational Knowledge Creation, Organization Science 5(1), 14~37, 1994. Definition of tacit knowledge and the difficulty of formalizing it (p.16). ↩
- Peter Massingham, Measuring the impact of knowledge loss: a longitudinal study, Journal of Knowledge Management 22(4), 2018. Knowledge loss from employee departures and gaps that are not filled (abstract). ↩
- DAMA International, What is Data Management?. Delivering, controlling, protecting and enhancing the value of data and information assets throughout their life cycles. ↩
- UN Statistics Division, Data Interoperability Guide, Chapter 3: Standard classifications and vocabularies. Standard classifications and data comparability. ↩
- W3C, OWL 2 Web Ontology Language Primer (Second Edition), 2012 and W3C, RDF Schema 1.1, 2014. Definitions of classes, instances and properties, and of object and data properties. ↩
- Coats Digital, GSDCost. 39 motion codes and fixed time values (vendor material). ↩
- SIJE internal material, sewing process glossary v4.4. Number of unique entries and languages covered. ↩
- Thomas R. Gruber, A Translation Approach to Portable Ontology Specifications, Knowledge Acquisition 5(2), 199~220, 1993. Definitions of ontology and conceptualization. ↩
- Rudi Studer, V. Richard Benjamins, Dieter Fensel, Knowledge Engineering: Principles and Methods, Data & Knowledge Engineering 25, 1998. “Shared conceptualization” and knowledge agreed on by a group. ↩
- ISO/IEC 21838-1:2021, Information technology, Top-level ontologies (TLO), Part 1: Requirements. Requirements for top-level ontologies. ↩
- Industrial Ontologies Foundry. Open reference ontologies across digital manufacturing. ↩
- SIJE internal material, SIJE company knowledge base v2.0. Scale of the apparel manufacturing process ontology. ↩
