GermanyRetail and e-commerce2021 – ongoingFlaconi
Real-time personalized product recommendations for an online beauty shop
Flaconi cut recommendation predictions from 20 minutes to 300 milliseconds on a Databricks data and ML platform
Problem
Flaconi, a European online beauty retailer with more than 55,000 products and over 2.3 million customers, had very large volumes of streaming data. Its infrastructure was costly to scale, and data analysts and data scientists needed different setups.
Approach
Flaconi built streaming data pipelines and machine learning on Databricks, with MLflow tracking model versions and experiments. The models serve real-time personalized product recommendations to shoppers. It also used Amazon SageMaker with the DeepAR algorithm to forecast demand against historical trends.
Results
Takeaway
Flaconi's main gains were in how quickly models could be built and served. The 5% order income gain was an expectation, not a measured result.
Categories
- Industry
- Retail and e-commerce
- Business function
- Marketing and sales
- Innovation type
- Product and service innovation
- Adoption
- Off-the-shelf tool or platform
- Evidence
- Vendor case study
Sources
Source summaries
Databricks describes how Flaconi moved its streaming data and machine learning work onto the Databricks platform. It reports real-time recommendation predictions in 300 milliseconds instead of 20 minutes, 200x faster time-to-market for new models, and infrastructure setup cut from two days to 15 minutes. It also gives an expected 5% rise in net order income and a 40% cut in staff costs.
Full details are in the original sources.
