Our project had a rough start. We had a plethora of ideas and starkly differentiating points of importance on what idea was worth pursuing. Our main point of discussion/contention was the core value. We did not want to do something that already existed and were very wary of solutions that were already covered or did not seem worthy/defensible. I finally compromised on our adopted project: SWYM. An AI analysis and “coach” mainly built for triathletes to get correct swimming techniques and strokes. I very much doubted this idea for a long time, as it did not seem to me at first that we could properly defend this idea in the market, nor that it was a worthy project. I did not feel that it was enough to generate real revenue and eliminate a pain point, making it a nice to have for triathletes, not essential, and that it would be hard to create any meaningful sustainable competitive advantage. This was when I was just thinking of the business model itself and AI as a model, which would change throughout the course. Ironically, I ended up covering the role of creating the value capture logic and really showcasing the sustainable competitive advantage in our model.
At first, it was hard for me to visualize how we actually would create value. To me we were just another AI for some sort of sport. But after discussing with my team and pinpointing just the triathletes as customer group and understanding how they think, and what they were paying for simple coaching, I could start to see some hope. I’m most proud of my 3-column value logic because after 2 iterations, it became a clear and solid case for our solution, with an undeniable competitive advantage which was made clearly sustainable later by identifying the scarce asset which can be amplified by AI, and was clearly explained in the AI canvas. The first version came to be a simple realization that it was cheap enough to create with general AI models, and demand was clearly visible due to the massive price difference in AI video analysis compared to an actual coach (simple pricing arbitrage). Knowing this was not enough to make a compelling case I decided to go directly to the AI canvas, where the real moat started becoming clearer. When I combined the AI canvas with the insights from Barney & Reeves about AI only amplifying what is scarce, I knew we struck gold. Our asset ended up becoming the coach validated footage and the swimmer’s records of what worked. This meant the learning loop was our moat and I could put together a clear value capture logic with a defensible moat to an idea that at first I was starkly against, which was not what I expected at all. On the other hand, when researching competitors clear ceilings became visible. While we did have a good value capture logic and moat at the end of this iteration process, looking at the general sports tech industry and possible competitors like Strava and Garmin, it became clear that there was envelopment happening and that an exit strategy is a highly likely outcome for SWYM, as a case similar has already happened to Runna and sets a good precedent for our projects future. Toward the end of the project my teammate was able to successfully integrate the pose model, and we got to see firsthand how we would train the model. This experience significantly highlighted the importance of coaches still identifying the joints and bone structure in videos and led me to reiterate the feedback section on the AI canvas, giving more significance towards labeled videos as actual asset, as I previously had envisioned that the model would train itself rather than rely on human expertise.
I was not directly involved in building the prototype, but I learnt a lot from observing my teammate in terms of capabilities, workflow and tools. Seeing firsthand a working demo created in such a short time was inspiring and furthered me to be inquisitive. He even found an open-source working pose estimation model with decent accuracy, which really consolidated the fact that the model does not make the business. I learnt a lot from him in terms of general setup, development frameworks, tools like Xcode and react native to create a great mobile showcase too. I tried helping with the 3d model to showcase techniques for the swimmers which ended up a failure, but being part of the iterative development of part of the prototype was eye opening to the possibilities now, with AI’s and tools really showing that the barrier to entry for many businesses lowered significantly. Conversely, it also showed how while barriers may be lowered, the dependency on suppliers also shifted, as we used googles Gemini Api to develop quick but then were restricted to its volume capabilities and other factors.
Regarding teamwork I found myself in a good environment with lots of good energy and ideas floating around. We picked up each other’s slack and helped each other out. We split work equally and simultaneously developed the project. This turned out to be less than ideal as simultaneous work led to strategic misalignments and lots of rework instead of iterative and incremental buildup. This was evident when my realization about the importance of having coaches still in the loop was made clear and my team members were forced to adjust to this strategic shift. I can say that any friction with the project came from how we structured it and not the people within.
Looking back, I was able to identify the sustainable competitive advantage potential after going through the steps learnt in the course and applied them correctly. Now, I still have reservations because we did want it to work. I learned that I would like harder and more pessimistic testing and even more pushbacks. However, for this project I’m quite happy as I already started it pessimistic and I was able to be convinced after applying most theories and insights gained throughout the course. Additionally, I truly noticed how a customer centric approach is the most important, no matter the great tech, or trends. Our teammate was a triathlete, and he was convinced this was great because he saw it first-hand, which is why I was willing to try. Apart from this, my biggest generalist takeaways are that AI models don’t matter anymore as they are open sourced and copied fast, and that proprietary data and a good feedback loop system is now a new and very lucrative durable moat. Lastly, it is always good to look at the bigger picture and not just the business/idea in the market, but the market itself too and its behavior and impact. This came from realizing that there was envelopment happening and a clear ceiling becoming apparent, which is not to scare off from the possible value creation but a reality check that must be addressed in the strategy.
