How to clean a material master: a practical plan

A material master clean-up fails in one of two ways. It is attempted by hand and abandoned at the ten-thousandth record, or it is done by a script that merges things that only looked alike. A workable plan sits between the two: machines do the reading and matching, people make the decisions that carry risk.

This is the order of work that holds up in practice.

01Step 1: decide the scope before touching a record

Which plants and material types are in. Spares and consumables are usually the place to start: high count, high duplication.
What counts as the same item. Same form, fit and function is the usual test; brand alone is not.
Who signs off a merge, and what happens to stock, open orders and history on the code that is retired.

02Step 2: classify everything

Put every record into a class: bearing, motor, valve, gasket, cable. Classification decides which specifications matter. A bearing is defined by bore, outer diameter, width and type; a motor by power, speed, voltage, frame and mounting. Without the class, two descriptions cannot be compared fairly.

03Step 3: pull the specifications out of the text

The description "MOTOR 5HP 3PH 1440RPM FOOT MTD TEFC" holds six facts. Extract them into named fields, convert units to one standard and record which facts are missing. This is the step that turns free text into something a system can match on.

Keep the original description. The extracted fields sit beside it, so a reviewer can always see what the system read and from where.

04Step 4: find the duplicates, with a verdict for each

Identical on every critical specification: a duplicate. Propose a merge.
Identical in function, different in a tolerance or a brand: a possible substitute. Flag it for an engineer.
Alike in words but different in a critical specification: not a duplicate. Record why, so it is not raised again.

05Step 5: review in order of value

Nobody can review everything at once. Sort the proposals by stock value, by spend or by how often the item is ordered, and work down. The first few hundred decisions usually cover most of the money.

06Step 6: put a gate on new codes

A clean master stays clean only if a new request is checked against the existing ones at the moment it is raised. The requester sees what already exists and has to say why a new code is still needed. Without this step the clean-up has to be repeated.

In short

Agree scope, the meaning of "same item" and who signs off merges before any record changes.
Classify first: the class decides which specifications define an item.
Extract specifications from free text into fields, keeping the original description beside them.
Give every match a verdict, review by value, and gate new codes so the work does not come undone.

Questions

How long does a material master clean-up take?

It depends on the number of records, how many classes they span and how fast your engineers can review proposals. The machine work is quick; the review sets the pace, which is why reviewing by value matters.

Can it be done without stopping operations?

Yes. The analysis runs on a copy of the data. Changes are loaded back in agreed batches, with retired codes blocked for new use rather than deleted.

What happens to stock on a code that is merged away?

That is a business rule to agree in step one: usually stock and open orders move to the surviving code and the old code is blocked, with the link between them kept for history.

Does this need our data to leave the plant?

No. The whole process can run on your own infrastructure, with nothing sent to an outside service.

Sounds like your problem?

Tell us about it. We'll say honestly whether the swarm can help, and what it would take.

Talk to us hello@quantumbeetle.ai

How to clean a material master: a plan that survives contact with a real plant