How Machine Learning can address the product data syndication challenges

Squadra has been asked by Stibo Systems as guest speaker to share our vision and best practices on this topic during  the Alliance meeting of Stibo Connect in Kopenhagen. The interchange of product data in the value chain is one of the biggest challenges manufacturers, wholesalers and retailers are struggling with. This is confirmed by the Product Information Management survey Squadra is conducting on annual basis as well as from our talks with many organizations regarding the so-call “product data onboarding challenges”. 

PIM and MDM solutions provide limited support to cater with the lack of use of product data standard between all parties in the value chain. Squadra promotes the use of standards like GS1 and ETIM, but see that many companies struggle to adopt these standards or to convert their product data into these standards.

Although product data standards like GS1 and ETIM are available, they are only used to a limited extend for the exchange of product data. “The standards do not full cover all the features of our product range”, “We partially use standards but more than half of our assortment is not covered”. “Our customers have their own proprietary standard.' As such the data landscape of all parties in the value chain is very scattered. It takes a lot of time for manufacturers to deliver the data that is often stored in their own data model format into the many different formats of their customers. Some customers ask them to provide data via datapools or at least according to the above mentioned data standards. Most customers ask their suppliers to deliver it into a proprietary format. The same applies to  other parties further down the value chain. retailers ask distributors and wholesalers to deliver information in their own format, implying that wholesalers struggle with the same challenges as manufacturers do.

Of course there are tools that facilitate the mapping of classifications and features, sometimes even including some basic ETL functionality to transform the values. This works fine for organizations who have to deal with a limited set of products. With the grown relevant of digital commerce, assortments have grown a well, implying that this manual mappings and transformation is blocking factor to grow business.

Macihine learning (ML) techiques do address tis issue: instead of the traditional business rule driven mapping (if this, than..) the ML algorithms are ‘trained’ with proven data, going well beyond e.g. mapping of features that have the same name. Even if classes with a classification structure of feature names in two different data model do have completely different names, the algorithms are able to match them based on so-called trustability levels. For more info, please feel free to download the presentation, including the link to the video that demonstrates the above.

Download the file here

This site is protected by reCAPTCHA and the Google
Privacy Policy and
Terms of Service apply.

Do you want to know more? Feel free to contact Squadra.
Send an e-mail to (+31 6 53352945)