A discovery tool to increase user engagement and promote broader recipe exploration
...flip recipe views from just the most popular 5% to the full range of 13,000 recipes?
...encourage people to uncover new cuisines beyond what they would normally come across?
...a discovery tool that suggests new ingredient pairings and recipes to users
Optimizely; React; Serverless; Sketch
The Recipe Finder was a web based discovery tool built for BBC GoodFood to drive engagement across their full range of 13,000 recipes and promote broader food exploration for their users.
The Recipe Finder was live on the BBC GoodFood site with 10,000+ daily users and saw a 64% engagement uplift. 49% of recipes clicked into were unique, showing that users were able to discover a broader range of content.
Results from A/B test
Presented with a visual starting grid of seasonal ingredients, users could begin with what appealed to them or what they already had in their fridge. Suggested ingredient pairings were then presented alongside a full list of recipes that featured their selection of ingredients. These ingredient pairings were organically suggested by our machine learning engine which could find novel flavour pairings based on their co-occurrence in existing recipes.
The system was built using a classifier that pulled out ingredient names and quantities from the corpus of recipes. An ingredient embedding space was created to generate a graph of connected and similar ingredients that could be queried in real time as the user edited their selection.
Using the Recipe Finder to discover new meals to try out