A page from the printed catalogue: item numbers, diameters and lengths set as typography, with hex bolt line drawings.
The catalogue of record. 241 pages of this.
← All work
Building Materials & Manufacturing

Product data nobody can actually build on

A print catalogue from the 1990s turned into a product data model. A distributor whose catalogue of record was a document designed for print in the 1990s, and the data model that had to exist before anything else could.

The problem

The catalogue existed, but only as layout. A specification sat in a table cell because that is where it fit. A variant was implied by a footnote. Units were written however the page had room that day. None of it could be queried, filtered, priced or synced, because none of it was structured as data in the first place. Building a storefront on top of that would have meant building on sand, and no amount of front-end work would have hidden it.

What we built

  • Worked outward from the print catalogue and the ERP to a real attribute model, deciding what was genuinely a product and what was a variant of one
  • Normalised specifications and units, so that two products described by different people a decade apart became comparable
  • Modelled the attributes customers actually filter and search on, rather than the ones that happened to get printed
  • Made the model the shared source of truth, so the storefront, catalogue search and ERP agree instead of each holding its own version
  • Set it up so a new product is entered once, as data, and appears everywhere it belongs

Result

  • A print document became a queryable product data model
  • The ecommerce store runs on structured data rather than transcribed pages
  • New products are entered once instead of maintained in several places
241 → 1
pages of print catalogue, now one queryable model
3 → 1
places a product is maintained
~65%
less handling per new product
Estimate Three systems of record collapsed to one, so a product is entered once instead of three times. Assumes comparable effort per system.

What one printed line actually contained

Scan of the printed catalogue, page 77: Cap Screws, Grade 2 hex head, with item numbers and sizes set in columns.

One line. Eight facts, none of them addressable. Repeat for 241 pages.

Product data modelvalidated
sku
043-0175string
family
Cap Screwenumderived
head_style
Hexenumderived
grade
Grade 2enumderived
finish
Zinc Platedenumderived
thread_size
7/16-14thread
length_in
1.750decimal
conforms_to
ANSI/ASME B18.2.1specderived
Now filterable byDiameterLengthGradeFinishHead styleMaterial
Four of these eight fields were never printed. They were implied by where the row sat on the page, and someone had to decide what they meant before anything could query them.