We are always looking for new ways to improve people’s lives and periodically meet in some location to brainstorm new ideas. During one such meeting in Silicon Valley in 2015, we were discussing the obesity and overweight epidemic and looking at various sites and apps that were addressing the fitness / wellness space. We wanted to make a difference to people with weight-related health issues.
We were struck by the level of effort required to use apps like MyfitnessPal, Fitbit, etc., and felt that these apps appealed more to ‘power users’, who used these apps for record-keeping and validation from other users. Most likely, they had already established ‘healthy’ diet and exercise habits and were using fitness apps to further hone their habits. Power users were deeply engaged with the quantitative aspects of the app and reveled in data entry, analysis and feedback. In contrast, ‘average users’ would consider these requirements to be very onerous and were likely to stop using the apps in a short while. We felt that we should come up with a much easier approach, one that would appeal to the average user, and require far less effort.
We consulted with our friends and others about what they would want in an ideal weight-loss app, and what they have experienced with other apps. They told us that what they needed was not just succinct information and actionable instructions, they also needed some hand-holding. They felt they were left to their own devices at meal times, especially in certain situations like a buffet, eating out with a group, someone pushing unhealthy food, etc.
We analyzed a large number of papers and journal articles as well, and some key findings stood out to us. One is that people lost some weight on almost every ‘diet’, but were unable to sustain them; the lost weight invariably came right back!
This is a widely recognized phenomenon, aptly summarized by American humorist Erma Bombeck: “In two decades I've lost a total of 789 pounds. I should be hanging from a charm bracelet.”
Mark Twain, another American humorist, once said: “Giving up smoking is the easiest thing in the world. I know because I've done it thousands of times.” The same can be said for weight-loss!
Another very surprising finding that stood out was that exercise, no matter how intense or sustained, did not play an important role in weight-loss! Diet played a far greater role, and therein lay the rub, because people find it very difficult to sustain drastic changes to their diets over the long term.
We also observed that while existing apps placed a lot of emphasis on recording and analyzing every morsel eaten by the user, informing the user to the nth degree of detail, but not really motivating any changes in behavior.
Also, these apps coached users ‘after-the-fact’, i.e. the app would tell them what all they ate and how they could do better next time - after they had already eaten their meals. Users would have to remember the instructions and make sure they followed them – all on their own! We felt we could do a whole lot better by hand-holding our users as and when they were actually eating their meals. We would give them actionable instructions and useful information just before they ate and that would make all the difference! They wouldn’t need to remember their coaching instructions from their last meal, and they wouldn’t need to enter any data. In effect, Tweak & Eat would be like having a nutritionist looking over the user’s shoulder as they were eating!
We came up with an approach that would require very little user effort (most of the work would be done by our nutritionists) – it would allow people to eat what they normally did, but with a ‘Tweak’ under the guidance of our nutritionists. All the user would have to do is to take a picture of their meal and send it to us, and within a minute, our nutritionists would send back a tweak. No data entry, no remembering anything,... it couldn’t be made any easier than this!
As all this backend activity would be manned by qualified nutritionists, we realized that we would eventually run into scalability issues, and we began looking at AI for solutions. The state of AI at the time, especially for image recognition left a lot to be desired, but things have improved dramatically in the last couple of years and we are now in the testing stages of an AI nutritionist that can reliably (more than 90%) identify a large number of foods. Our AI nutritionist will get better over time, as we continue to train it with ‘real-world’ images of meals sent by our users. This is an important point, because training an AI nutritionist using stock images is easier but not very reliable. When we are satisfied that the AI nutritionist is reliable enough, we will put in in production and thus not only alleviate the scalability issues but also position ourselves for long-term growth into the billions of tweaks.
We then laid out the technical design and requirements, applied for US and International patents (still pending), and began development with a small team of very talented individuals.
The ‘non-fitness enthusiast’, by which we assume you mean the average user, will really resonate with the fact that he or she is getting a real-time ‘Tweak’ to the current meal. Having a nutritionist ‘looking over their shoulder’ and ‘hand-holding’ them to healthy eating has enormous appeal.
We capture certain body and health parameters as part of the registration process, and as we interact with the user, we learn more about his/her preferences. The AI has been trained to recognize a number of foods and will eventually perform more and more of the nutritionist’s activities, and human nutritionists will be there to approve the AI’s results, and to handle exceptions and deviations.
Both are possible. Our AI generates certain types of personalized diet plans, and our nutritionists do so as well - it depends on the services desired by the user. In all cases, in order to get a nutritionist’s hand-holding, the user will have to send a picture of whichever meal they want tweaked.
India, Indonesia, Singapore, Malaysia and Philippines, right now. India is the leader – we launched the app & service here first.
We believe females are using the app more than males. It is very difficult to characterize the ‘average’ Indian fitness enthusiast, but regardless, we have seen positive changes in eating behavior after people use Tweak & Eat for just a few weeks.
The recommendations are initially preconceived to provide a starting point for each user, and as the user sends tweaks, the database builds and the nutritionist (as well as the AI) learns more about the user’s preferences and makes more context-and-history-based recommendations.
Non-perishables can be shipped anywhere, whereas perishables will be limited to specific locations.
This information is company confidential. Nutritionists will initially cover morning to night (typical meal times); eventually we may provide 24X7 coverage based on user demand.
We will expand our repertoire of user-engagement vehicles (Tweak Wall, Recipe Wall, etc.) and we will try to keep users engaged and coming back for more. They will be able to manage all their health & wellness activities - we want to become the one-stop-shop for all things ‘wellness’. Also, we will be expanding our premium services and provide more specific premium services for various health conditions and health objectives.
How AI can help manage diabetes and other food related issues with ‘Just-in-time’ nutrition assistanceRead more...