Walk any multi-site maintenance operation and the symptoms repeat. The same bearing stocked under four item numbers in three languages, because each site described it in its own words. Stock-outs on one shelf and dead stock of the identical part on another. Free-text requisitions that purchasing cannot match, so the part is bought again, at list price, with expedited freight. A steady flow of returns because the description and the part disagreed.
None of this appears as a budget line. It is paid anyway: in carrying cost, in duplicated inventory, in search time, in production risk. A quiet tax, levied by bad data.
Segment first: runners, shakers, movers, sleepers, dead stock
An inventory of fifty thousand items is unmanageable as one list and obvious as five. Built on the classic ABC and criticality analyses, our segmentation names the segments for how they behave. Runners move constantly and earn tight replenishment and sharp pricing. Shakers are the high-value, specification-critical items where the award deserves real care. Movers turn steadily and reward standard contracts. Sleepers barely move, but some protect uptime and must be held regardless; the question is how many, where, and at what stocking logic. Dead stock moves never, and every month it stays on the shelf it costs money and space while pretending to be an asset.
Each segment has its own sourcing logic and its own stocking logic. Treating them identically is how organisations end up with safety stock of dead items and stock-outs of runners.
OEM where it binds, commodity where it does not
The second segmentation is OEM versus commodity. Where warranty terms or technical criticality bind, the OEM part is the right buy, and the model should respect that constraint rather than chase a false saving. Where they do not bind, the same fit-for-use part exists as a commodity at a fraction of the price, and only clean technical attributes let you see the equivalence. Technical life belongs in the same decision: an asset near end of life changes the stocking question entirely, sometimes down to a final last-time buy.
Clean data, then let the model work
The cure is unglamorous and compounding: identify, classify and cleanse the item base; impose naming conventions; enrich technical attributes; translate once, centrally. Deduplication then surfaces substitutes across sites, and a free-text request can be matched automatically to an equal or better part already on the shelf or already under contract.
The returns are not marginal. In well-led programmes like ours, carrying costs fall twenty to forty percent, item counts drop ten to twenty percent through duplicates and obsoletes, and a quarter to half of the planning and buying labour is recovered. In one such programme, a large manufacturer lifted inventory turns by seven percent, cut inventory by four to twelve percent per site, and recorded a multi-million saving within eighteen months, several times the programme cost.
And for procurement, clean data is leverage: properly classified parts with real specifications can finally be aggregated, competed and awarded across the whole base. The first step is unglamorous and vendor-neutral: clean the data, whoever does it for you. And if you want it run as an engagement, bring us one messy parts category and we will show you whether, and how much, the quiet tax costs you.