Mapping GS1 and UNSPSC: The Agent Still Needs Clean Data

An agent that maps a GS1 code to UNSPSC returns the quality of the record you hand it. That is the point of this session with Stephan Spijkers, Isak Marais and Stiaan Maree. AICA Data ran a live classification. The oversight stayed with people.

The recording:

At the microphone:

• Stephan Spijkers, Co-Founder, PIMvendors.com
• Stiaan Maree, CTO, AICA Data
• Isak Marais, CEO, AICA Data

What the session covered:

• Agentic classification was presented as a way to map codes with less manual work, including catalogs that keep changing. The concrete example was the live demo of AICA's classification, which the speakers presented as accurate on the examples shown in the room.
• Hand mapping falls behind once the catalog grows and the number of target codes grows with it. That is the speakers' case for a model.
• Data quality remains the constraint. The session said bad input still produces a bad class. Categorising well still matters in e-commerce, because the class is how a channel and a buyer find the product.
• The agent adds capacity. The product-data team stays accountable. In the speakers' words, the better results come from the model plus someone who can overrule it.
• Scaling that pattern across more systems was left as the next pressure.

The standards themselves, and how GS1 sits next to ETIM and ECLASS, are in product classification. The rule the model is applying still needs an owner, which product data quality and governance treats as an operating problem. What an agent downstream can actually do with the class is the subject of AI-ready product data.