9,175 supplier spellings brought back to 1,493 for SPIE Building Solutions
SPIE’s asset register had been fed by maintenance systems, contractors, and manual entry for years. Pearstop matched every manufacturer against SPIE’s own approved list, scored each record for confidence, and gave the team a ranked list of where to review first.
One manufacturer, a dozen spellings
SPIE Building Solutions manages a large portfolio of building assets across many sites. Over the years its register was filled from different maintenance systems, contractor uploads, and manual entry.
The result: the same manufacturer under a dozen spellings, and equipment types missing or recorded wrongly. The lists worked for a mechanic on site. They did not work for planning maintenance or pricing bid risk across a portfolio.
Every asset carries a manufacturer and a type, the same way every purchase order line carries a supplier and a description. The engine that cleans spend data works on an asset register without changes.
Matched against the list SPIE already trusts
Early on we saw general AI models guess or invent supplier names. So every row was matched against SPIE’s own approved supplier list, and nothing else.
2 · Cross-check with a language model
3 · Route conflicts to a person
Deterministic matching first
Exact, prefix, edit-distance, and typo-correction matching against SPIE's approved suppliers, wherever a rule could decide.
An independent second opinion
A large language model checked each match on its own. Where the two agreed, confidence went up.
People decide the hard ones
Conflicting matches went to SPIE's master data team, and every decision fed the next iteration.
Five spellings of one pump maker, resolved
One family of circulation pumps showed up under five supplier spellings, one a plain typo. The type field was messier still: a bare family name in one row, a full model code in the next.
Names changed. The shape matches the real register. A supplier team now sees this exact pump family, in these exact configurations, instead of “pumps, all sorts”.
A ranked review list instead of a wall of rows
Every record got three scores: supplier confidence, type confidence, and review priority. SPIE’s team started at the top of the list and worked down.
The confidence scoring meant our team knew where to spend their review time first, instead of starting from scratch on 200,000 rows.
Bart van PeijHead of Master Data Management, SPIE Building SolutionsWhen a client has an approved list, match against that list
Matching only against SPIE's validated suppliers raised both accuracy and the team's trust in the output. It is now standard on every Pearstop engagement with a defined supplier base.
From servicing assets to advising on them
With a structured register, SPIE can look at a whole equipment family and spot repeat failures across a client's portfolio. A technician can say:
“This pump family fails at this rate across your sites. Here is what we recommend.”
“Pearstop's manufacturer matching gave us the right starting point to load data back into the ERP. This project taught us a lot about data cleaning with AI.”
Same problem in your asset register?
Send us a sample of your asset or equipment data. We match it against your own reference list and send it back scored, so you see a first pass on your own records.
