AI for catalog data: product width or package width?

Two widths, two fields in the catalogue

A supplier gave the product width and the package width separately, each with a value and a unit. The catalogue has separate fields for both dimensions, but the import shortened the names and saved both values as „width”. An employee has to go back to the spreadsheet to work out which number belongs on the product card.

A useful proposal keeps the original label, the value and the unit, links to the place in the spreadsheet and names the target field. It gets harder when other suppliers describe the feature differently: one writes „product width without packaging”, another puts the width in a „transport packaging” section. The word „width” alone is not enough, and the model has to read the context.

How much a simpler change gives, and how much training adds

In this count a supplier name dictionary gives most of the effect for a fraction of the cost. Training makes sense at a larger scale or when the remaining errors are expensive. A missing unit, or the abbreviation „w.” with no heading, stays a gap for the supplier to fill, and the tool shows it next to the proposal.

The set of training examples has to be maintained: when the process, the naming or the base model changes, the evaluation is repeated on the same test set. We include that cost in the quote together with the training itself.

How much a supplier name dictionary solves

A fixed supplier template lets you record an unambiguous link: its column matches a given field. When the names differ but their meaning is agreed, a dictionary of equivalents is enough. Akeneo describes a measurement as a value with a unit and groups units into families with a shared standard unit. Converting centimetres to millimetres is a rule, but first you have to know what the number refers to. Checking product cards before publication is covered in the article on product card completeness.

Where to start

The team explains the catalogue fields, maintains their definitions and points out difficult supplier names. We run the trial on spreadsheets not used for adapting the model, comparing the dictionary, a ready-made model and an adapted model. We adapt models as part of private LLMs, and we build the app where an employee sees the result next to the source as part of AI apps. For the free analysis, a few examples of an error that keeps coming back and a note on who corrects it today are enough.