At this year’s Delivery Hero Hackathon, the team “The Price is Right” secured third place with their fresh, AI-powered take on an age-old question faced by food vendors: “How can I set an optimal price for my fried chicken?”
The theme of the 2023 Delivery Hero Global Hackathon was “Igniting growth with AI”. Participants from around the world formed teams to ideate, execute and present their ideas within a span of just three days.
The team The Price is Right had a particularly diverse composition, representing four different entities. Despite geographical distances, they united behind a powerful idea: leveraging AI, specifically through Graph Neural Networks, to empower Delivery Hero vendors to set the best —and most profitable— prices for their dishes.
The Price is Right: helping vendors set the optimal price for their dishes
It’s all connected: utilising graph neural networks to predict dish prices
In an ever more competitive marketplace awash with data, Delivery Hero vendors need all the help they can get to satisfy customers and maximise revenue. With this context in mind, we came up with an ingenious solution: to harness the power of Graph Neural Networks —a form of Deep Learning applied to Network Graphs— as applied to Delivery Hero data on interactions between items, vendors and customers.
Using the Graph Neural Network to predict variables like item price, we captured information about how prices might vary based on factors such as item similarity, user baskets, vendor menus and more. This data was then utilised to generate predicted price elasticity curves for each dish, capturing the relationship between price and number of dishes sold.
But the model was only half the battle! With the predicted data in hand, we still needed to package this information in an intuitive UI, enabling vendors to set the optimal price for their dishes and reap the rewards.
What’s cookin’ good lookin’: a simple and intuitive interface streamlining price recommendations
We designed an intuitive visualisation of the price elasticity curve that integrates directly into the vendor portal:
- The vendor receives a notification that some of their items would benefit from a price review
- We display the price elasticity curve, highlighting the optimal price point that is likely to result in the most orders
- The vendor can then easily navigate to the menu management page, where the price recommendations are highlighted
Et voilà! In just a few easy steps, vendors can incorporate the recommendations of a powerful deep-learning model into their pricing strategy and maximise their revenue.
So is the price actually right?
Overall, the hackathon was a great experience for us. Being located in four different countries posed no barrier to developing a great idea and impressing the judges enough to secure third place!
However, we faced many challenges while tackling the problem:
- Graph scalability: expanding the network graph with increasing numbers of items and connections required more processing power than was readily available during the hackathon
- Model refinement: the tight time constraint meant that the Graph Neural Network was poorly optimized and could have performed better
- Data sparsity: data on transactions for a limited time period and a single market (as provided for the hackathon) meant that the model could have performed better
While the performance of the model on the day was less than ideal, the underlying concept certainly merits further exploration. So stay tuned: price optimisation might soon be a feature on a platform near you!
If you like what you’ve read and you’re someone who wants to work on open, interesting projects in a caring environment, check out our full list of open roles here – from Backend to Frontend and everything in between. We’d love to have you on board for an amazing journey ahead.
