Retail

Waitrose turns to AI to create recipes for successful food products


Under fake pink cherry blossom, guests sipped House of Suntory cocktails and picked at plates of chicken karaage, prawn gyoza and cauliflower tempura from a kaitenzushistyle conveyor belt … This was the London launch of Waitrose’s new Japanese range.

But without knowing it, and even if you live hundreds of miles away, your food choices may have had a hand in shaping the supermarket’s 26-dish Japan Menyū range. That is because it was developed with input from Tastewise, an artificial intelligence (AI) platform that analyses menus, social media and online recipes to pinpoint food trends.

While many businesses and individuals are concerned that AI is going to eat their lunch rather than set the menu, the technology is becoming more prevalent in the food industry, with its use doubling since 2017, according to McKinsey’s 2022 Global Survey on AI.

This is probably because it offers under-pressure retailers and food manufacturers an understanding of what fickle shoppers will want to buy in the future. It takes a year to perfect a new food project, but even so most of them miss the mark, and in recent times, companies have instead been forced to play catch-up with trends that have exploded on social media.

Martyn Lee, Waitrose’s executive chef, whose career includes a long stint writing menus for a national restaurant chain, says until recently the only way of getting trend information was to use market research agencies. “That didn’t mean that the information wasn’t great, because it was: the problem was, everyone used the same people,” he says.

“We are a premium retailer. We’re not the biggest, but our food is all about quality,” he says of the UK’s eighth largest supermarket, with a market share of 4.6% against Tesco’s 27.2%. “Customers come to us for innovation, newness and excitement. If we’re seeing the same information as everybody else, that gets very hard and, because of our size, the risk of getting it wrong is really high.”

Traditionally, researchers would spend days studying restaurant menus and combing social media to identify trends, often laboriously inputting data into spreadsheets tracking the rise of an ingredient or dish. Now Israeli company Tastewise and rivals such as London-based BlackSwan provide a short cut.

Tastewise data helped Waitrose to decide to put its money where its mouth is: social discussions about Japanese cuisine were up 15%, it said, while over a nine-month period there had been a 5% increase in the number of restaurants adding it to menus. It even identified yuzu and ponzu as popular flavours. (Menyū includes cauliflower tempura with a soy yuzu dip and “shredded cabbage and edamame slaw” with a ponzu dressing.)

“AI can be quite a scary term … but it’s a way of us getting billions of data points of information,” says Lee. “The scale is so, so big that it just can’t be done by human trawling.”

And while many of us post pictures on social media without much agency, AI spots patterns. “Instagram is really interesting because people tend not to take photos of food they don’t like,” says Lee.

“Then, often, people will put hashtags about the texture. For example, they might say, #fried, crispy or barbecued, so we get a picture of the cooking method. If somebody pins a recipe on Pinterest, that gives you a great sense of an intention to cook.”

These insights, alongside Waitrose’s own customer and sales data, help it to get a more definitive idea of whether a product will fly, which Lee says helps “de-risk” a process where the “potential of a product failing is quite high”.

This explosion of social media use has left the food industry, with its bureaucratic product development processes, trailing in its wake, says Andy Upton, who co-founded Panku, the Japanese and Korean street food brand.

“Consumers are moving quicker than retailers on trend prediction,” says Upton, whose latest project is the AI-based app Sooggi, which aims to solve the age-old problem of what to have for dinner. “Vegan and plant-based was a wake-up call because they were all behind the curve.”

When Upton used machine learning to research food trends before the 2020 launch of Panku, which today has more than 140 kiosks in Asda stores, several other trends, including bubble tea and CBD (cannabidiol), also jumped out.

“The interesting thing about using AI for trend prediction is you see the trends a long way off,” he says. “Five years ago I tried to talk to retailers about bubble tea and they were like: ‘What are you talking about?’. It only becomes important when it becomes relevant, and to become relevant, it needs a critical mass.”

Illuminated combined John Lewis and Waitrose logos at the entrance to a store
Waitrose says using AI helps reduce the significant risk of a new product failing. Photograph: Peter Lane/Alamy

But what about when AI goes wrong – as it did for New Zealand supermarket Pak’nSave when an app designed to generate helpful meal ideas served up recipes for deadly chlorine gas and “poison bread sandwiches”? “This is where the human element is really important,” says Lee.

“You have to treat it as a conversation because sometimes it will throw out a concept you know is not going to work. You have to go through the process of having a conversation with the platform to refine it further, constantly asking questions, which will get it closer and closer to what you want.”

Waitrose is also testing Tastewise’s new AI chatbot TasteGPT. At a recent conference in London, Lee cooked “Indian/Mexican fusion tacos” – a concoction that pooled trending cuisines, proteins and flavours that it had identified.

But rather than feeling threatened by a chatbot that can generate recipes in seconds, Lee sees the potential to develop products faster than the current 12-month average.

“The process you go through to write a recipe is quite extensive because it has to be so precise,” he says. “If you’re a gram out in a 100g recipe, it’s going to be wildly out by the time you make a tonne of it.

“What this enables me to do is get recipes and photographs tailored to food trends that I can then test with customers before any of my team put ingredients to pan.”

The end result of this will, he believes, be happy shoppers: “You’ll see things on shelves closer to [the time] when people want them.”



READ SOURCE

This website uses cookies. By continuing to use this site, you accept our use of cookies.