A technician is waiting for a mechanical seal to get a pump running again. The system shows zero in stock. Meanwhile, two boxes of the same seal sit at the back of the store, recorded under another code, typed in four years ago by a colleague who has since left. The part is ordered as an emergency, express shipping is paid, and the pump stays down for two days.
The scene is ordinary in a company that sells, installs or services industrial equipment. The item master grows over the years: new brands, new suppliers, references typed by hand, descriptions abbreviated differently each time. Nobody has time to clean it, because cleaning it is in nobody's objectives. This article is part of our guide to AI and automation for industry.
Why it costs more than it seems
Spare parts stock ties up cash, and part of that stock is no longer used. A 2025 survey by SPARETECH, a software vendor, of 300 industrial executives in the United States and the German-speaking countries of Europe, estimates that on average 22% of spare parts stock has not moved in more than five years. This is vendor data, from markets that differ from Tunisia: it gives an order of magnitude, not a measure of your store. We found no equivalent Tunisian data.
Duplicates make the problem worse in both directions. They inflate stock, because you buy again what you already have. They also cause apparent stock-outs, because the part exists but under a code nobody searches for.
What a tool can do from a plain export
The starting point is neither a sensor nor new software. It is an export of your item master and, if possible, of your stock movements over a few years.
- Match references that are probably the same part. The tool compares descriptions, manufacturer part numbers, dimensions and suppliers, including when they are written differently ("BRG 6204 2RS", "bearing 6204-2RS1"). It proposes groups of likely duplicates, each with the reason for the match.
- Find dead stock. From the movements, it ranks items by date of last issue and by value tied up. You see on one page what no longer moves and what it represents.
- Flag poorly covered critical parts. For the parts you mark as critical, it compares available stock with past consumption and supplier lead time, and prepares a reorder alert before the stock-out.
- Prepare corrections without applying them on its own. It produces a list of proposed merges, parts to write down and thresholds to review. Your team approves, then the corrections are made in the ERP, by import or through its API.
The assistants we build use AI for the most tedious part: reading badly written descriptions and understanding that they refer to the same part. To find a technical specification in supplier catalogues, see also an assistant that knows your catalogues and technical manuals.
What about maintenance?
Spare parts and maintenance are linked: a critical part is often missing because the job was not planned. According to NIST, in US manufacturing, plants that rely mainly on preventive and predictive maintenance have markedly fewer unplanned stoppages than those that repair after failure. A Fluke survey (vendor, more than 600 respondents, 2025) reports that 61% of manufacturers had an unplanned stoppage during the year.
What we do: the software side of preventive maintenance. A schedule per machine, reminders before due dates, a history of interventions and parts used, and the link with stock so the part is there on the planned day.
What we do not do: install sensors or predict failures. Predictive maintenance needs sensors, a history of measurements and often an automation integrator. We do not promise it.
The limits, stated plainly
- We do not replace your ERP. We work on its exports or its API. It is the same approach described in the ERP trap: integrate rather than replace.
- The result depends on your data. An item master with no manufacturer references and no movement history gives a thinner analysis. We tell you so after looking at the export, before committing.
- A person decides. The tool proposes duplicates; it merges nothing on its own. A wrong merge on a critical part costs more than a duplicate.
- No promised savings figure. The gain depends on your stock. We measure it with you on a first, limited scope.
How to start small
Pick one family of parts: bearings, seals, or the parts for one range of machines. Set a simple indicator at the start, for example the number of confirmed duplicates, or the value of stock with no issue in more than three years. A first pass on that family quickly shows whether the approach is worth extending to the rest of the file.
If supplier documents are still retyped before they reach your stock, the neighbouring topic is covered in purchase orders, delivery notes, supplier invoices: stop retyping.
If your store holds more codes than real parts, let's talk.
Do you have this problem? Describe it in a few lines: what happens today, who is involved, what it costs you in time. We get back to you quickly, usually within 48 working hours, for a short discussion: to understand your project and see whether we are the right partner.
Describe your problem · WhatsApp · All our solutions for industry
Sources
- SPARETECH survey 2025 (vendor, 300 industrial executives, United States and DACH), via Reliamag: reliamag.com
- NIST, manufacturing machinery maintenance (US manufacturing): nist.gov, PMC
- Fluke, unplanned downtime (vendor, more than 600 respondents, 2025): fluke.com