Today the key to more customers and to a significant increase of sales is “recommend” instead of “selling”!

On-line shopping pioneers such as Amazon.com did it and proved it: depending on the category of products between 10 to 20 per cent increased sales in the Web-shop by means of purposeful, automated article recommendations in the “Long Tail” of the total assortment. In the meantime German enterprises also recognized this enormous potential. E.g. Quelle.de increased the portion of cross-selling-articles in the market place demand even more than ten times after the introduction of its recommendation-engine-solution in the beginning of 2006.

The METRO group makes one step further and examines the real-time-recommendation-solution for the offline purchase world in its future-store. Through so-called personal shopping assistants (PSA) and displays installed in shopping carts, the corresponding articles are chosen and scanned by the customers themselves. At the same time similar articles are offered to the customers.

How do IT-supported recommendation engines get the knowledge about group purchases and demand chains, about preferences, favors and usage habits?

In order to identify group purchases, on the one hand the shopping-cart analysis can be made (with temporal purchase sequence analysis or the sequential shopping-cart analysis) on the other hand the methods of the so-called collaborative filtering (CF) can be applied. At the same time the article affinity of each customer for each article (customer-article-matrix) is computed on the basis of similar articles or similar customers.

• Product-based approach: based on a product A and customer B the approach looks for a list of products X,Y,Z, … which customer B bought, clicked, rated, etc. in past and which are most similar to A. Based on a weighted average of these similarities the affinity A-B will be calculated.


• Customer-based approach: based on a product A and customer B the approach looks for a list of customers C,D,E, … who are most similar to B. Based on a weighted average of these similarities the affinity A-B will be calculated.

If you would like to know in detail which solution is successful in practice, what are the further aspects like e.g.

• A/B- and multivariate tests (“lab on the Web”),
• Similarity and distance metrics,
• Role of article-category/-hierarchies,
• Handling with sparse customer-article tables,
• Collaborative filtering (CF) of new customers – what to do?,
• Consideration of the article characteristics (brand, author, composer,…),
• Considering the affinities of the customers (color, size, model,…),
• Availability check and business rules,
• Item prices and price optimization (PriMini - price mining),
• ...

which factors are responsible for the increase of turnover to 10-20 per cent in webshops, then please click here.

You can find further scientific literature under this link.