From a part name and a SKU to a classified bearing catalog.

The challenge

Where things stood

Stock reaches this US bearing and MRO distributor from many unrelated sellers. Most listings carry little more than a product name and a SKU, with no structured dimensions and no bearing type, so every new lot has to be identified and reconciled before it can be listed and priced.

How Claro helped

What changed

Claro built a bearing-type classification from the listings themselves, then defined the attributes common to every bearing and the ones specific to each type. It filled those attributes from ranked sources, manufacturer pages first, with marketplace listings ranked lower and never accepted as the only evidence. Every value carries its source and a confidence level, and low-confidence items go to review instead of straight into the catalog.

In the data

How the catalog was structured

One attribute model, layered by category, type and series. Specs shared by every bearing sit at category level. Each bearing type adds only the dimensions and ratings that apply to it, so nothing is asked of a product that cannot have it: a tapered roller has no seal field, a thrust ball no radial clearance.

Table showing one source width column meaning three different dimensions: ring width B 8 mm for deep groove ball bearing 6000, total assembled width T 15.25 mm for tapered roller bearing 30204 with cone width B and cup width C missing, and bearing height H 9 mm for thrust ball bearing 51100 — **“Width” arrives as one field and means three dimensions.** In the source data one width column carries ring width for ball bearings, total assembled width for tapered rollers and height for thrust bearings. Split into B, T and H, a filter on “width 9 mm” no longer puts a thrust bearing’s height next to a ball bearing’s ring width.

Three cards on empty and filled bearing attribute cells: seal or shield on tapered roller 30204 is not applicable and excluded from the completeness score; cone width B and cup width C on 30204 are missing but findable and filled from the manufacturer datasheet with the page cited; radial clearance CN on thrust ball bearing 51100 is filled but not applicable, so it is held for review and kept off the filters — **Blank, not applicable, or wrong.** A completeness score treats every empty cell as a gap. On bearings an empty cell can be correct, findable on a manufacturer datasheet, or filled with a value that does not apply, and each one needs a different action.

Bearing catalog filtered by bore d equals 10 mm, returning five products from three categories in one comparable table: deep groove ball bearings 6000, 6000-2Z and 6000-2RS1, self-aligning ball bearing 1200 and thrust ball bearing 51100, with outside diameter D, width keeping its letter B or H, seal type, radial or axial load rating and number of brands — **The “10 mm bearing” search, compared across categories.** Once every family shares one attribute model, a single bore filter returns ball, self-aligning and thrust bearings in one comparable view. Width keeps its letter, and thrust ratings are marked axial so they are never ranked against radial ones.

Illustrative examples on standard bearing designations, shown to explain the model. They are not extracts of the customer’s data.

Results

What it looks like today

One attribute model for every bearing family: shared specs at category level, type-specific dimensions and ratings below it.

Fields that mean different things by bearing type, such as width, split into B, T and H so customers only compare like with like.

Every empty cell sorted into not applicable, missing but findable, or filled but not applicable, each with its own action.

A source and a confidence level on every value, with low-confidence items held for review.

What we took from this engagement

On bearings, “empty” and “filled” each hide three different situations. A catalog is only as good as its ability to tell them apart.

Claro's conclusion, not a customer statement.