FEST is Europe’s central trade body representing the national associations of wholesalers engaged in the distribution of plumbing and heating products across Europe.


In order to save costs and to ease the portability for product data from manufacturer to wholesale, downstream to retail and (end) customers, there was a strong need for standardization of product data in the plumbing and heating, or – more general – the Heating, Ventilation Air conditioning (HVAC) branch. The lack of a standard leads to duplication and unnecessary transformations of product data. Moreover, having to meet the needs of the different wholesale and retail organization requirements, leads to costly errors in product data and makes the expansion to other markets and countries more difficult.

ETIM is the standard classification model for technical products and is chosen as the standard for the HVAC branch. This means that manufacturers in this branch will have to convert their own meta data definition (e.g. manufacturer specific product classes, feature names, feature values and units of measure) to the ETIM standard.

FEST is operating in eighteen different countries, and just over half are participating in ETIM for the HVAC sector. Thus, there is strong a need to encourage the remaining countries and belonging manufacturers to convert their manufacturer specific product data to the ETIM standard. This transformation process can be quite labor intensive: products have to be classified according to the ETIM class definition, and manufacturer specific feature names/values/units of measure have to be matched with and converted to the ETIM standard.


The usage of modern Artificial Intelligence (AI) techniques, however, can ease these transformation processes significantly so that less time is required to convert a manufacturer’s own meta data definition to the ETIM standard. After a short assessment of the available data, Squadra Machine Learning Company proposed to use the Powerconvert.ai tool for the (1) classification of products, (2) feature name matching and (3) feature value conversion.

  1. ETIM Class prediction

The ETIM class prediction algorithm predicted the ETIM class given a manufacturer name + product description. The service was offered in two forms: a bulk classification service which inputs an excel file with multiple products and outputs the excel with the predicted ETIM class added to it, and a ‘google like’ web search page which enables a user to enter one manufacturer + product description and get the predicted ETIM class as a result.

How to deal with other languages
The algorithm needs to be trained with already ETIM classified product data. Therefore, the algorithm works with product descriptions in the language the algorithm is trained with. This is a potential issue, since the initial scope is intended for manufacturers from Southern Europe and Eastern Europe and there are not yet data pools available in the language of these countries.

However, recent research has shown that there is an alternative way to deal with this. The Finnish, Dutch or German training sets can be translated with Google Translate, and a new model can be trained using the translated training set. The research project has shown that this method results in nearly the same accuracy as the algorithms using the native language training set (only a 1-2% difference).

So based on the initial dataset plus usage of the Google translate service, Squadra Machine Learning Company was able to create ETIM Classification algorithms for every (European) language.

  1. Matching manufacturer feature names to ETIM feature names

The ETIM feature matching algorithm provided a tool to intelligently match manufacturer’s non-ETIM features to the ETIM format for a manufacturer.

The service was offered as part of the Powerconvert.ai web application where the manufacturer could upload a dataset in a standardized excel format, and the algorithm predicted the matched features. The user was additionally able to validate or adjust these mappings per feature. In this way, all the feature name matchings were done within the web application and this data could be used to teach the algorithm, making it improves over time.

  1. Conversion of feature values  / units of measure

Once the feature name matches were found and validated, the feature values were converted to the proper format. The manufacturer  for example used inox steel instead of stainless steel. If the feature value was not yet in the ETIM format the algorithm proposed a matching feature value. For features values where the unit needed to be changed, for example going from dm3 to m3, the algorithm could detect these.


With the help of Powerconvert.ai, FEST is now able to convert their manufacturers’ meta data definition to the ETIM standard. Based on the initial dataset and the help of Google Translate, Squadra MLC has assisted FEST to better cope with different languages, which assists them in their international online presence. Lastly, by using this software, the brand has saved a lot of valuable time and money to manually convert product data.