How can companies predict what consumers want? It’s a burning question the food industry – indeed, most industries – would like answered, and one that Gousto hopes to solve with the implementation of nifty tech.
One of the leaders in the UK’s meal-kit market, Gousto now offers its customers 30 recipes a week, having recently diversified their offering to include gluten-free and plant-based meals. This March, it managed to raise £28.5m from venture capital investors to put towards its goal to provide 400 million meals by 2025.
That’s a big target, which is why the group is working with graph database platform Neo4J to help it offer recommendations to its users.
“The sweet spot between convenience and choice is personalisation,” says Irene Iriarte, data scientist at Gousto. “We’re committed to using technology to make our customer interactions relevant, making it easier for them to find the dishes they want.”
An explanation of how graph databases work can quickly become very complicated, but essentially they help identify connections between individual items. This information can then be used in everything from fraud prevention to social media analysis.
For the purpose of shopping, however, it’s most commonly employed to help companies make useful recommendations to customers. Walmart is one of the largest retailers to utilise graph database technology, but the applications are effective across any scale of business.
Gousto has been using Neo4J to develop its ‘recipe similarity measure.’
“A graph database allowed us to easily capture how interconnected our recipes and ingredients are, capture customers’ needs and then analyse the relations between this data,” says Iriarte. “Neo4J allows us to explore the data in a visual way, allowing us to build a recommendation service we can adapt and change.”
To tailor products to individual needs, the meal-kit maker focuses on two particular data sources: previous interactions with the menu – including recipes chosen and ratings of purchased meals – and information on upcoming recipes.
“This allows us to create a recipe similarity measure, which provides us a good indication of which recipes we think each customer will enjoy most,” notes Iriarte. “So next time they choose their recipes, we use this understanding to help them navigate our menu, with recommendations that make meal planning even easier.”
As an example, if an individual rates Gousto’s Asian cuisine highly but also orders burgers from the menu regularly, it makes sense to suggest that they will love the Asian chicken burger with sesame fries.
The graph path to success
While Gousto is still only in the testing phase with graph database technology, the initial results have been promising. In fact, there’s been a 30% increase in the number of customers selecting recommended recipes, indicating a more accurate appraisal of what users want.
So what is most popular with Gousto customers? Proving that convenience is king, the Ten to Table range (featuring meals that take just 10 minutes to prep, cook and serve) is being chosen by nearly half of customers every week.
Most highly rated, however, are the Fine Dine In recipes, which offer premium, specialty ingredients and take a little longer to make. Clearly, when they have the time, people like to show off with dishes like the pan-fried salmon, pea velour, roast asparagus and Hasselback potatoes.
Beyond the initial data-gathering phase, Gousto already has ideas about where to take the tech next.
“There’s so many directions we could go,” says Iriarte, “from recommending recipes in order of performance, to using the tool to predict the uptake of recipes – which would help us improve our forecasting and continue to battle food waste.”