Home PublicationsData Innovators 5 Q’s for Michael Haase, CEO of Plant Jammer

5 Q’s for Michael Haase, CEO of Plant Jammer

by Eline Chivot
Michael Haase

The Center for Data Innovation spoke with Michael Haase, chief executive officer and co-founder of Plant Jammer, a food tech company based in Copenhagen that developed an AI-powered cooking mobile app. Haase discussed how Plant Jammer is built to formulate recipe recommendations to its users.

This interview has been edited.

Eline Chivot: You founded Plant Jammer in late 2016. How did you come up with the idea and what did you hope to achieve or change?

Michael Haase: I came up with the idea to be able to solve my own problems. I was working 80 hours a week for a big consultancy firm, and there wasn’t time left for cooking although I really wanted to. I couldn’t find a way to make time for it. I found it frustrating to have to look for a recipe online. You go to Google where you get millions of results, and it’s difficult to quickly find something that truly works for you. My girlfriend at the time was a good chef and knew just what ingredients can be combined to figure out what to create. And then I thought that this is something a machine can do very well—learning what goes well together. So I developed an approach whereby you start with the ingredients, and cook with what you have, reducing food waste, and eating more of what you’d like to eat. It allows you to learn the fundamentals of cooking in a fast, playful, and easy way, and to cook with what you’ve got instead of depending on online search, creating a new recipe from scratch.

I worked in sustainability and studied various industries such as the agriculture sector and the mining sector from these lens. Change for a better world in those areas involves big infrastructure project, big governmental or corporate decisions. But when it comes to food, there’s something quite magical: the individual agent, the individual household, which don’t require regulatory changes, big governmental or corporate action to make a difference. This is about micro-actions happening every day. And as an individual, if I want to be more “sustainable” and respectful of the environment—consume less water and less electricity—one thing I can really do something about is food, particularly in two ways: I can reduce the food waste I generate, and I can reduce my meat consumption. Behind Plant Jammer, there is the idea to concretely achieve that change we all want through this type of effort.

Denmark is a country whose industry is based on pork and dairy. We have a meat-based culture. This approach that involves eating less meat and more vegetables has only recently started to find an echo with people. Sweden, Northern Germany, and the UK are much more ahead. There is growing awareness of food waste as an issue, particularly in France, where it’s high on the agenda. What people often don’t know is that 50 percent of food waste happens in the home kitchen, in their home, in their fridge—not in retail. We get mass offers and buy too much food we don’t get to consume in time because we live busy lives.

Chivot: How does your application generate recipe recommendations, and what data do you use to improve and test these results?

Haase: The problem in the current world of recipes is that every app and web page that provide them is based on the database of recipes that is currently available. If you want to identify a dish that includes, say, six ingredients, there could be about 1.7 billion great recipes. But there is no database with more than 2 million recipes. That means you will never be able to fit exactly what’s in your fridge into a dish using search the way you currently do online. So we need to think differently, and build a recipe from scratch. Plant Jammer isn’t based on a recipe database. We build it in real time the moment you click on “Give me a recipe with these six ingredients.” We then ask the AI how that recipe would look like.

We work with two layers of technology that enable this, both AI- and human-powered. We first collected recipes from all over the web. We included a mix of Western recipes, including in French, German, and from the United States. We taught an AI to identify patterns emerging from these resources, identifying which ingredients go well together, and which ones do not. We developed a full landscape of ingredients where you can see your olive oil with your balsamic vinegar, basil, and pine nuts, and your peanut oil with your sesame seeds and your eggplants—a whole map to “travel in.”

The second layer involved working with some of Copenhagen’s greatest chefs to gain insights about the right components required behind any good dish. We asked these chefs to look at this map of ingredients and to provide us with their own observations and ideas of patterns. Results included a “gastro-wheel” which we built to describe what you need to make any good dish. You need a base, something fresh, and what is called “umami,” which certain ingredients provide under certain cooking conditions. Umami is the feeling of depth, which is something you might be missing if you’re eating plant-based food. You need some oil to tie ingredients together, some crunch for texture complementarity, such as with almonds, pine nuts, and hazelnuts, some acidity that accentuates all flavors, and you need to play with acidity using something sweet, something spicy, and something bitter. We asked the chefs which ingredients are umami, and how umami they are. This part reflects the “human intelligence” we used to build our system, a great “engine” that has led to very interesting findings complementing the technology we used. For example, onions appear next to the fruits on our map, because they are a source of sweetness and can be caramelized.

After creating this language around the reasons why certain ingredients go well together, we needed to get to an actual recipe, and for this, we needed to materialize cooking. We asked the chefs their six steps to making a good risotto. What should we do differently if we use eggplant instead of zucchini in a risotto? What changes, what does it alter? Which steps must be adjusted? We tested various versions of risottos by replacing ingredients by others. Whenever you include risotto, eggplant, pine nuts, balsamic vinegar, soy sauce, olive oil, and fennel, you put the ingredients into the right buckets which creates the text for the actual recipe, based on the chefs’ recommended steps and narrative for the right way to create a good risotto. The logic we used this way enables to then build a risotto with any set of ingredients. We can even build an Indian risotto, or a Mexican risotto. Mixing ingredients, processes, and timing allows to build anything. This provides flexibility and develops a “jamming mindset” like a jazz band’s session: you have some rules, but can experience “freestyle.” We talked to Mark Bittman, a famous American food writer, who said that there are only 12 recipes in the world—everything else are branches of those recipes. We just need to find those. We have defined 20 now, and are still expanding, nevertheless some patterns regularly appear. A congee rice in China is actually quite similar to Italy’s risotto, process-wise.

AI cannot stand by itself, it can provide you patterns, but you need to train it really well to give it some straitjacket recommendations. AI is great when you know that you are looking for and missing “something umami” for a dish, and need to ask the AI to give you this information. If you ask it to give you a recipe that makes sense and is good with four ingredients however, AI is less good at it. The combination of human and artificial intelligence is what makes our application meaningful.

We improve our algorithm based on our database of 45,000 people using Plant Jammer on a weekly basis. We learn from their daily behaviors—for instance, people don’t seem to include peanut oil in their “box.” We also improve our service through kitchen experiments. For example, we invite people into the experimental kitchens of Miele, one of our investors, we give them a table of ingredients and the app, and then we observe their behaviors. When are they excited, confused, or angry? We apply these results to the next version of the app, to the database, and the algorithm. Through such experiments, we learned that people get really nervous when cooking something different or new—until they start smelling the food. So we modified the app so that the first thing you do will give you that smell early on in the process, such as by cooking onions or spices in the frying pan straight away. It’s a small hack that leads you to a more relaxed, playful attitude, which makes it more fun to cook.

Chivot: Who is your target user, and how would they use the application?

Haase: We realized that the people who are continuously using our app and for which it works best are 18- to 24-year-old female users. They tend to be urban, students, and they don’t have much money so they need to cook themselves. Eating less meat and reducing food waste is, for them, an economical lifestyle, it fits their budget. In addition, they are concerned with being and feeling healthier. If you cook yourself, you eat “real food,” and if your food habits are plant-based rather than meat-based, you feel better. This nutritional aspect is a parameter we particularly focus on: helping users to identify what they want to eat more or less of, and enabling that in their journey.

There are two sides to the user experience: one of them is shopping, the other is cooking. Ours users first fill up the app with their pantry of ingredients: what they have in their fridge, what types of spices, oils and vinegar they already have, etc. They receive a weekly recommendation notifying them of what they seem to be missing, such as an umami ingredient, and suggesting six umami ingredients they could add to their pantry if they want to make good recipes more easily. Users do not have to act on those recommendations. When they go shopping, they can open the app to be guided through the various aisles depending on the location of these missing ingredients. Currently we are not focusing so much on guiding them to particular supermarkets, but that is one iteration we will work on later on. We plan on developing the app so that it could inform users of local offers from supermarkets depending on the supplies available. At this point though, we are focusing on providing a good, playful shopping and feeling experience, through which users are not constrained by the rigidity of a list, but can pick just one of these ten ingredients that would be good for them.

When getting to the cooking phase back home, as a user, I just need to open the app, where I can see the pictures of recipes that fit with what I have. I can pick the one that I find looks interesting, and can be told through the app’s recommendation that I might want to add something crunchy to my dish, or something to spice it up—an ingredient I already have—and then I can use the recipe, cook with it, and then save it in my profile. Finally, I can update my data based on what I have left in my pantry, so that I can receive better recommendations next time I use the app. I can share my recipes through Plant Jammer’s community with the other users, which can use and modify each other’s recipes.

Chivot: How would the average home cook benefit from your application?

Haase: The major benefit that Plant Jammer provides is the reduction of the distance from “What do I have?” coming home for dinner to “A delicious dish” created easily, in a playful mode. That distance tends to be large and cumbersome, so people may go for a takeaway option, something quick, as they come back from work, tired, and do not want to make any more decisions. Also, with our app, people start using what they have instead of shopping for more while they actually have enough. That triggers an economic motive and leads to cost-efficiency as an additional benefit. They start buying food based on what they need rather than because they think it might be interesting, making it more of a “thrifty” approach to grocery shopping. And as mentioned before, the health element is a key benefit as well. The app shows people whether they are eating enough protein or magnesium, and enables them to boost this by suggesting more of those ingredients they would need to achieve a more diversified nutrition. Users can also list in their profile the ingredients they are allergic to or prefer not to eat.

Chivot: The microwave and frozen food were groundbreaking innovations in the world of home cooking and changed food habits. How will cooking apps, robot chefs, and 3D printed food change how we cook or what we eat?

Haase: The way we will cook and eat will become more and more personalized. There are certain aspects to food and preferences that we share which can be integrated in technologies using AI and machine learning. But as there are many personal aspects involved as well, such as nutrition, taste, your microbiome, your stomach, what you can or should eat. This will lead us to new, upgraded experiences. I can imagine how we will soon have our own flavor footprint, which we will be able to send to restaurants before going out for lunch or dinner. The restaurant would then be able to see which of today’s customers must eat gluten-free, love broccoli, hate cilantro, or prefer spicy food.

Further, I believe we need to rethink the concept of a recipe, to understand it not as an entity but as involving a mix of components and other elements such as nutrition and flavors. Being able to cook will remain a survival skill. We’re currently in a middle zone, where our cookbooks have become a piece of furniture, but at the same time, our kitchen is not very digital yet. We’re no longer just using books to cook, but are not entirely “online” and connected in the kitchen either. Most people still go and cook by their habits, cooking less than 10 different dishes a year, and then maybe use a bit of Instagram to post a picture of their dishes. Going by habit in the kitchen is not good for our health, food waste, and the human experience, so we really need to find something better.

Mobile phones have not yet entered our kitchen the way they have entered the entertainment industry. We don’t really want to constantly have our phones near our faces, we also want to sense things. So what I think will work for the kitchen of the future will be the types of technology that enable you to taste more, to feel more, and to be more present when you are cooking and eating—not necessarily through an app you have to use, but more through a feeling: using this technology, I don’t have to be using my phone too much while cooking. I can cook with the help of just a few clicks, and I feel good because I am present in the moment and in touch with my physical environment. Google Home devices and Amazon’s Alexa reflect that companies have understood that. They are investing in this kind of kitchen experience which is a way to use speech to be present and not use technology as a substitute to real life.

You may also like

Show Buttons
Hide Buttons