Recommendation Engine Analytics

In the digital age customers of retail and media have an unprecedented power of choice. Shopping around, making comparisons, searching for the best reviewed, lowest priced, most recommended item is now the norm. Loyalty to one retailer or brand is becoming a thing of the past and customers move fast.

The pressure is on to provide more than the time-tested qualities of great products, exemplary service and coompetitive pricing. Retailers and media providers must find more effective ways to retain customers and hold their attention. In the online environment particularly the integration of a recommendation engine allows for collection and analysis of customer choices and behaviour, which results in actionable insight, in turn providing the customer with a more personalized, relavant experience.

Depending on the environment in which it is deployed, a good recommendation engine will identify and recommend a next likely purchase or a bespoke offer or a matching item based on the behavior of a customer and the behavior of others like them. Using data such as recent purchases, online browsing, use of loyalty cards, viewing habits and combining this with customer segmentation data results in well-targeted recommendations and increased take up.

Common scenarios where a recommendation can be used include:

  • Online fashion stores: Recommend matching items to create a whole look, suggest items based on browsing and purchases, show best reviewed products.
  • Music and film: Recommend song or film based on recent listening or watching patterns, suggest similar music or films by the same director / featuring the same actor, show award-winning or well reviewed items.
  • Advertising revenue: The longer you can hold your customer's attention, the more targeted advertising you can place.
Challenges facing recommender systems

Real-time results refresh: Given the volume of data (1000’s of content x millions of Users), refresh of the recommendation results as the dataset is augmented may take days.

Lack of Data: How to address the 'cold start' problem for both new content and user due to lack of data.

Diversity of recommendations: Bias towards popular content, dealing with difficult content, ranking large catalogs

Recommendation Engine solution
Portrait of a couple holding color swatch near the wall. I like it.
The Key Pillars of the Fuzzy Logix Recommendation Engine

Efficency

  • Hybrid Model incorporating techniques like content filtering, collaborative filtering, behavioral segmentation, etc.
  • Able to tackle problems like 'cold start', eccentric content, popularity bias.
  • Special ‘Oldies’ recommendations provided to tackle the long tail phenomenon.

Immediacy

  • GPU-based recommender system. Usually 1000x faster.
  • True real-time results. Refresh of the recommendations possible as soon as the dataset is augmented.
  • Virtuous engine cycle ingesting more data, refreshing results, learning more; thus becoming more efficient faster.

Data Model

  • Fuzzy Logix Media & Retail Data Model.
  • Extensive data model schema already in place to be fed to the engine.
  • Able to integrate all the different sources of data like content management systems, web navigation, sales ledger, user information, user-content engagement, etc.

Ease of Deployment

  • Recommendations can be put live on your system in less than a week.
  • We just need API/FTP access to the data sources and the results will be embedded in your system
  • Infrastructure hosted and managed by Fuzzy Logix in the cloud (can also be done on-premise).