Relishing a challenge: How data science informs Gousto’s business model

x Sponsored content on Research Live and in Impact magazine is editorially independent.
Find out more about advertising and sponsorship.

A decade ago, if you wanted to have dinner delivered to you in the comfort of your own home, you would have been limited to whatever takeaway restaurants were in the vicinity – and the chances are, they wouldn’t offer particularly healthy choices. For that, you’d have to head to the supermarket to buy fresh ingredients, consult a cookery book, and do it yourself. But what if you could bypass the first two steps? 

Today, rapid disruption in the grocery sector, driven by advances in technology and consumer desire for convenience and wellbeing, has resulted in an alternative to supermarket food shopping – the rise of the cook-at-home meal kit. 

UK recipe box subscription service Gousto was founded in 2012 by Timo Boldt and his then partner James Carter. Though he loved food and cooking, Boldt found he had little time with his career as a hedge fund manager to cook, and when he did, he was wasting a lot of ingredients. 

These two pillars – convenience and food waste – are at the centre of the start-up’s approach to food delivery. Each box, selected by customers via the Gousto website or app, contains only the amount of ingredients required for the recipe, and customers can choose from 25 recipes each week.

The success of Gousto’s direct-to-consumer model is dependent on the company’s ability to understand what its customers want to eat and, as such, it has invested heavily in data tools and expertise, to the extent that Boldt describes it as “a data company that happens to operate in the food world”. 

Boldt says his vision for the company was to allow the customer to drive ideation around the products – for instance, monitoring which recipe boxes customers buy most to tailor the products to their preferences. “My philosophy was, can I build a framework that means innovation equates to ideation, plus selection, plus execution – and is there a way that the ideation can be done by the customer? Can I give customers recipes and see what they look at, what they buy, to fine-tune the menu to their tastes?”

The company looks at recipe feedback given by customers on its website, as well as data generated when orders are placed (metadata associated with each recipe includes ingredients, cuisine, cooking method, cooking time, and so on) to measure the performance of specific recipes and build an understanding of preferences. Boldt describes the data insights derived from the Gousto website as a dynamic segmentation model of sorts; by observing what ingredients people like and don’t like to eat, the company can start to understand their behaviour. 

This data is then merged with a cluster segmentation, based on people’s attitudes and attributes, such as favourite cuisine or how many people they are cooking for. 

“On top of this, we run two layers of deep learning, which allows us to build a personalisation engine that shows exactly what you are most likely to want to eat,” says Boldt. The model improves as more data is generated: “The more orders you place, the better our orders become.”

Though the company is biased towards funnelling spend towards data science and investing in building artificial intelligence and automation, it also uses focus groups to complement its strategy, particularly around new product development. 

Natalia Paine, the brand’s insights manager, says the company uses consumer research early in the process to give context on behaviour and motivations when coming up with a new concept. “Focus groups are most important in the early stages, when trying to explore an area more broadly – when we are exploring particular customer groups’ motivations and behaviours, or when trying out concepts for a new range of recipes to understand what people’s actual needs and wants are, along with how we are meeting them. Focus groups offer context for a lot of what we see in comments and through other means throughout the year. In them, we can probe the ‘why’.”

Softer feedback 

The new Boost and Balance range, for instance, was launched in January and offers recipe boxes for meals under 600 calories, and was informed by customer insights and ‘softer feedback’.

“We couldn’t possibly do this just based on machine learning and data insights,” says Boldt. “We went out to customers and tried to find out whether we should call it ‘healthy January’, or ‘vegan’, and we called it Boost and Balance, based on customer interaction. So, there’s a really important role for traditional marketing when it comes to proposition development for the core segments.”

But what people say and what they do don’t necessarily match – a constant issue for researchers, and particularly pertinent when it comes to food and health. “We often have focus groups tell us that they want healthier recipes, for example, and then we look at the data and the people who gave us those answers and what they buy – and they buy pork and sausage. Reconciling the two worlds is sometimes hard,” admits Boldt.

Another insight – that people want to cook a special meal once in a while – led to the launch of a premium range, Fine Dine In, marking a step away from recipes focused on healthy eating and convenience. “The main occasion of the midweek meal is very fast and nutritious; however, people still want a special night once a week,” he explains. “It allows customers to use Gousto in a slightly different way than they’re used to.”

Perhaps surprisingly, the brand’s customer demographic is skewed towards those aged 35 to 50, cooking for their families; it’s not necessarily the urban millennial working long hours who doesn’t have time to shop.

Boldt says the company isn’t trying to convert people to cooking, however. “I’m not in the game of telling you that cooking is great. I’m trying to help customers who already cook. If you look at who’s cooking every single day, it’s probably not the 25-year-old guy working at PwC in London. It’s more likely the 45-year-old woman with a young kid living outside Leeds.” He says the company’s customer demographic is skewed towards females: “Sadly, home cooking in the UK is still hugely traditional and done mostly by women, whether you like it or not.”

When it comes to evaluating the food trends on which to base new recipes, the company starts by examining what customers are searching for on its website. It usually sees surges in traffic before trends are covered by the media and will build new propositions based on the trends it deems substantial enough. Gousto recipes include vegetarian, plant-based, gluten- and dairy-free options, and though it hasn’t yet ventured into kosher and halal ranges, Boldt says this is something the team “constantly re-evaluates”. 

Data mining in this way also helps to give insights into people’s relationship with food – for instance, when it comes to gluten-free, some consumers see it as a trend or health benefit, whereas, for others, it is a dietary requirement that has an impact on how food is handled in a warehouse. 

Success criteria 

When trialling a new range, the brand initially launches one single meal and invests some marketing spend on promoting it. It then looks at the take-up ratio compared with other meals to infer if the recipe appeals to one customer segment or all of them, with the aim of maximising the recipe’s relevance for all segments. “There’s quite a bit of prioritisation framework in the middle, based on our customer understandings,” says Boldt. 

The success criteria for new product development is how well the firm retains customers, he adds. “If they don’t retain well enough, but we can get sufficient amounts of them, we then start to ask: could we retain them better if we simply had twice the choice? Is it because the meals are too long? It’s a combination of data richness and automated process, but following up and going much deeper into behavioural questions.” 

Being able to access in-the-moment insights through automation creates a short cycle for research and development, allowing the company to identify quickly whether or not an idea has scale. Boldt attributes this to Gousto building “very narrow data IP around forecasting, automation and personalisation to the customer”. 

Often, ideas don’t pan out as expected, but it can be just as beneficial for the company to understand what doesn’t work.
At any given time, it runs a number of new product development, marketing channel and customer tests, while website user experience is tested once a week.

Net Promoter Score is also managed by collecting weekly feedback on particular recipe box numbers. Paine explains that as well as looking at the overall score – calculated by taking the percentage of promoters and subtracting the percentage of detractors – the company also breaks down the complaint areas that are preventing a perfect score. “This allows us to understand what areas have the biggest impact on customer satisfaction,” she says. “These categories include ‘range of recipes’, ‘delivery’ and ‘ingredient quality and shelf life’.” 

Gousto can then track how changes improve complaints, she says. “The more points off in a category, the more important it is to tackle. We also look at this by box size, diet type and different factors.” For example, when it was continually noticed that ‘range of recipes’ was a substantially bigger cause for complaint by customers, the company increased the number of vegetarian recipes by a third, adding plant-based recipes. 

The firm keeps all its work on NPS and insights in house. Boldt says: “I’m a big believer that if you want to build a long-term successful business, you have to obsess about your moat strategy – how do you protect the castle in the future? To widen the moat, you have to own the data science and the technology.” 

Data insights are available to every team within the company in real time, with teams able to create their own dashboards. Boldt gives the example of someone working on the procurement operation, responsible for buying onions. “They have all the information available to create dozens of dashboards on onions. That person can really understand everything, from shelf life to consumer behaviour to working capital by supplier, freshness and supply-chain risk.” 

Paine adds that people across the business gather their own insights, ranging from the digital product and customer services teams to the SEO manager. “We’re all sharing what we learn with each other, so that we can build upon existing knowledge.”

As well as helping to predict what recipes customers will buy, the company uses data to inform its own purchasing. Through a centralised data warehouse, a forecasting algorithm uses information about past orders to predict how much of each ingredient it will need to fulfil forthcoming orders, so reducing food waste.

The system gives the company an advantage over supermarkets, many of whom are grappling to retro-fit data models on top of legacy systems. “Their challenge is that they have no idea what’s in store,” says Boldt. “It takes them 48 hours to find out what was sold and another day to ship it to the store.”

Gousto isn’t the only brand offering recipe-box delivery in the UK, with its closest competitor being HelloFresh. One criticism of the subscription box market has been its long-term viability: while it’s easy to draw consumers in with discounted initial offerings and one-off purchases, is it sustainable in the long-term? Boldt thinks so. 

“The barriers to entry in this market are quite low so lots of new competition are entering the market, but barriers to scale and to overcome the retention topic are significant. We’ve invested millions into capability, because if you don’t, you will have issues.

“I think what you see today is the absolute tip of the iceberg. There are roughly one billion meals eaten in the UK every week, and we’re selling a tiny, tiny percentage of them. This market is so enormously large and, for the past 50 years, supermarkets have built up a supply chain that is no longer fit for purpose for the next 50 years.” 

This article was first published in Issue 21 of Impact.