AI for catalog data: product width or package width?

Two widths, two catalog fields

In a hypothetical spreadsheet, a supplier uses the labels “product width” and “package width.” Each has a value and a unit. The company has separate catalog fields for both dimensions, but the import shortened both labels to “width.” The distinction was present in the source and disappeared during transfer.

A useful mapping suggestion retains the original label, value, and unit, with a link to their location in the spreadsheet. Alongside them, it shows the proposed destination field, such as “package width.” The specialist can check what the measurement describes before approving it. In this case, correcting the fixed mapping between columns and fields should be the first change considered.

The task becomes harder when suppliers describe the same attribute differently. One writes “product width without packaging,” while another places the width in a section headed “shipping packaging.” The word “width” alone is insufficient. A model can consider the full label and surrounding description when suggesting the appropriate catalog field.

Convert the unit after identifying the measurement

A value in centimeters can be converted to millimeters using a fixed rule. First, however, you need to establish what the number measures. Correctly converting package width does not turn it into product width. These are separate fields, even if both store a length in the same unit.

Akeneo describes a measurement as a value together with a unit. It groups units into measurement families with a standard unit for conversion. These capabilities in a product information management system, or PIM, are useful once the attribute has been assigned correctly.

If a supplier gives only “W” with no heading or explanation, the specialist may need confirmation. A missing unit also needs to be obtained from the source. The tool should show that uncertainty beside its suggestion rather than infer the meaning from other products.

How much can a mapping dictionary handle?

A consistent supplier template allows explicit mappings: a supplier's column corresponds to a particular company field. When names vary but their meaning is agreed, a dictionary can record the equivalents. Also check whether the current import preserves section headings and separates package dimensions from product dimensions. If that context is lost, the data transfer needs to be fixed.

An existing model is worth assessing where descriptions vary and specialists regularly have to read them in full. It receives definitions of the catalog fields and suggests assignments for review. Comparison with the dictionary shows whether it recognizes unfamiliar wording and whether it confuses attributes that simpler rules already handled correctly.

Fine-tuning means further training an existing model on reviewed examples. A trial may be worth considering if the model keeps making particular mistakes despite clear field definitions, and the catalog owner can identify the correct assignments. One example would be repeatedly overlooking “without packaging” when selecting a field. If an abbreviation's meaning is unknown, the supplier needs to explain it.

What the catalog owner needs to see

The trial should include material from suppliers not used during customization, with different ways of describing dimensions. The catalog owner checks which suggestions still need correcting and whether the source label is easy to find. For ambiguous names, they also check that the system leaves the issue open for clarification. An apparently complete product record with a dimension in the wrong field can conceal work that will have to be done later.

Compare the same material with the current mapping and the existing model. If the dictionary produces equally useful assignments, further training may not justify the maintenance cost. If a customized model needs fewer corrections, the team still needs to check that staff can review and approve them conveniently in their everyday tool.

Comparing the finished product page with supplier data is a later step, covered in our article on checking product listings before publication. Here, the decision concerns which field a supplied attribute belongs in as data enters the catalog.

Connect suggestions to the PIM review process

We propose an AI application for catalog data that shows the supplier's label alongside a suggested PIM field. The specialist approves or corrects the assignment while retaining a link to the source. Approved data can be passed to the catalog through the existing import process; product publication remains part of the company's established workflow.

Syntalith compares the current mapping with an existing model and assesses customization where errors in interpreting labels recur. Your team explains the catalog fields, maintains their definitions, and identifies difficult supplier labels.

For an initial conversation, describe an attribute that regularly ends up in the wrong field and how your team corrects it. We can establish whether the work calls for an import change, a mapping dictionary, or help interpreting varied descriptions. We will agree together on how examples are used. See Syntalith pricing for information about the offer.