Deep Recommendations Deep Recommendations:
A Generational Shift in Recommendations
Personalize from Day one   |   No More Relying on Historical Data
store-assistant

Suggest products like store assistants

Often the only way to sufficiently understand a user’s affinities is by looking at the products that interest them. No amount of tags or words can help to predict what they’re likely to buy. Visual AI uses deep learning to extract features from images, and generate recommendations that visually match the seed product. These recommendations help shoppers build conviction and make a decision, and serve both visually similar products, or complementary products that complete-the-look.

Visually Similar

Visually Compatible

We instinctively knew that using visual aspects of a product for recommendations is effective in fashion and lifestyle business - it's much closer to the expertise of our merchandisers. I am excited with early results - our engagement is up 40% over our merchandising rules, and revenue per 1000 impressions has increased by 19%, compared to the other recommendation models.

Sylvain Lys - Head of Omnichannel Customer Experience

1
promod-logo
analytics

Personalize without historical and behavioral data

Traditional recommendation engines work well when you have tons of behavioral data for each product. New, seasonal and long tail products, however, do not get picked up in recommendations due to lack of past events. NLP powered recommendations go past this constraint, and start relating products immediately, so you can start personalizing from Day 1, and give shoppers the joy of discovering unique items.

Deep recommendations with NLP is right now the top strategy, and is delivering average attributable sales of Eur 10.68 per click. The results are scaringly good. Without RichRelevance, we wouldn’t have used these innovative AI technologies that differentiate us, and help us grow.

Anton Paasi, Head of E-commerce

1

Retailers that are exploiting the first-mover advantage