An Insight into Gousto’s Data Kitchen

When I started my job at Gousto, I honestly struggled to get my head around how data-driven the company is, with every single team leveraging data in some way. Since joining, we (the analytics team) have undertaken some seriously cool projects across the entire business — a few examples being:

  • Building a predictive model to calculate the likelihood of a customer buying their next Gousto recipe box
  • Analysing customers’ online behaviour to make the website user experience seamless and as easy to use as possible
  • Working closely with Gousto data scientists to optimise the performance of our automated ingredient picking operation

With this in mind, it’s easy for our colleagues to sometimes forget that we don’t have a magical crystal ball to answer their questions. Instead of supernatural powers, we prefer a mixture of SQL and Python and tools like Periscope to solve Gousto’s challenges.

The area of analytics I want to focus on in this blog post is really at the heart of the business and is the main reason I (and probably most of my colleagues) joined Gousto — FOOD. Given the level of investment into data science and analytics elsewhere in the business, it should come as no surprise to learn that we crunch some serious numbers to understand what it is that tickles our customers’ taste buds and keeps them coming back for more. In particular, I will explain how we use data to inform the recipe development process here at Gousto. I’ll then showcase some of our tastiest dishes, before finally revealing some fun facts and interesting stats about our recipe collection.

Whilst I can’t divulge too much in this post, it’s no secret that to measure the performance of a recipe, we look at data from recipe specific customer feedback — on our website and app you have the option to rate each recipe you taste out of 5 and let us know what you think. And whilst the average taste score of a recipe is an important metric to look at, you don’t need a PhD in statistics to understand that it’s also essential to consider the underlying distribution of those scores. A dish with an average taste score of 4 may have done so by scoring 4’s across the board or say 3 x 5’s and a 1. The graph below demonstrates quite an extreme example of this.

If we are to nail the recipe development process, then we also need to listen carefully to feedback from those few 1’s to ensure we are delighting absolutely EVERYONE.

So what does a successful recipe look like?

Whilst we recognise that there’s always room for improvement, with 5 years’ practice under their belts, by now our food team know a thing or two about developing delicious recipes. So without further ado, allow me to introduce one of our customers’ favourite ever dishes (and my personal favourite to date): The Pulled Chicken Teriyaki Donburi Bowl.

Not only did this recipe excel in terms of taste scores, it accounted for nearly 8% of all recipes sold that week. That’s nearly twice as high as you would expect if all 22 recipes on offer performed equally. Chef Jordan, take a bow son!

Going further than just feedback

Because our customers are a busy bunch and don’t all have time to rate recipes, we also make use of the data generated when they place their orders and choose from the 22 (and counting) recipes on offer each week. Each recipe has associated metadata (cuisine, cooking method, cooking time etc.), not to mention ingredients, that we can explore to gain a deep understanding of customer taste preferences.

But in the world of big data, instead of considering all customers equivalently, there’s tremendous value in studying customers as groups or cohorts. We cluster customers together according to different attributes like their favourite cuisine or how many people they’re cooking for. In doing so, we can better understand why it might be that they loved the chilli prawn linguini or that nutty goats’ cheese salad.

Generating tasty insights

Now I’ve given you an idea of what’s possible with our data platform, let’s put that into action and unearth some insights based on real-life Gousto data. Here are some more of our customers’ favourite recipes, served alongside some tasty facts and figures.

This Indian Spiced Carrot and Lentil Soup scored well across the board, but astonishingly the average taste score for families was 70% higher than it was for couples. We suspect that this is because to families, time is a particularly precious commodity and this soup bears minimal extra work for every additional mouth to feed (plus it was delicious).

It’s not hard to imagine why a light, fresh salad is so appealing on a hot summers day, but to what extent do we actually prefer a salad when the sun is shining? Well we served this Provincial Chicken & Bacon Salad once in February and once in July and witnessed a whopping 130% increase in popularity (that’s sales relative to other dishes). By that logic, we’ve concluded that salad is 130% more appealing in summer than winter!

Our final fun fact takes the form of a quiz question — which of the two below vegetarian dishes do you think Gousto meat-eaters preferred?

If you said feta & sweet potato taquitos, give yourself a pat on the back! Though there’s no hard and fast rule for this, what we frequently see (as in the case of the taquitos) is that carnivores go mad for the veggie dishes that are presented in a form they’re accustomed to (think curries, lasagne, enchiladas).

Future challenges for data-driven product development

With our recipe library currently in the thousands and growing week on week, managing product development is inevitably becoming more challenging. We’re tackling this issue head-on by employing principles of data democratisation; educating other teams and sharing insights company-wide. A pertinent example of this undertaken by the analytics team is the creation of several automated emails to share live company KPIs. Our hope is that by making a point of sharing our insights, teams will feel more empowered to make decisions for themselves and ultimately be better equipped to home in on the optimal solution.

Come and join us

If you have an insatiable appetite for data and bottomless curiosity then we would love to hear from you! Keep an eye out for data roles on our jobs page or drop us an email at

Edward Cardy,
Data Analyst

Originally published at on July 7, 2017.