- Opening
The problem that me and my team wanted to solve was that of groups who want to play dungeons and dragons (D&D) but can’t find a Dungeon Master (DM) because their existing one is burned out or nobody wants to take up the role. My group and I thus chose track B since we wanted to create a new technological solution. We ended up creating an AI DM who could help a human DM by narrating the story and also do all the things a normal DM would have to do in case there was no human DM. My role in the team was mostly to come up with the pricing on the subscription tiers and how these would compare to existing products such as virtual table top platforms (VTTs).
- My Role and Contribution
My role was mostly to figure out how our business model would even work and what the actual problem was that we would try to solve. When it comes to the problem at first I thought it would be that we would simply try to replace human DMs with our AI narrator, however after doing some research and data collection I found that many groups have DMs, but they are often just burnt out or overworked. This switches the problem we were trying to solve from “are we replacing humans DMs?” to “how can we lessen the workload on human DMs”. Because of this we focused mainly on filling in the game nights when a human DM might be too burned out to play or even assist him in prepping for the game.
- Key decisions and turning points
Originally we actually wanted an AI agent who helped supermarkets manage their inventory, however after the feedback we received from our professor we realized that this solution would not actually generate much value since software already existed which could do all the inventory management. Because of this feedback we started brainstorming again and came up with the idea to create an AI narrator for D&D. From this feedback I learned that AI creates little value where good alternative solutions already exist. When we started our Crit Happens project we were unsure about how we would actually monetize our AI narrator. The original idea was that we wanted to make our product completely free to the user by using ad revenue. However, running a purely ad-based free tier would not work since ads simply do not pay enough to run our AI narrator. Because of this I had to come up with a completely new business model which would work as a freemium model where users could test our product for a limited amount of time and would then be forced to upgrade to a paid tier which would actually be profitable to use. Another key decision we made was including pre-made campaigns and characters. Originally we thought we could just let the AI narrator come up with everything (stories and NPCs) however we quickly realized that this would be too easy for anyone to copy. If we wanted a defensible product we would have to create a unique product that couldn’t just be copied by anyone which is why I went ahead with creating custom stories and narrators. This change would also allow us to save on cost as the AI would not have to come up with completely new stories and characters and could use that computer for narrating the story and gameplay. Lastly, we had to look at how we would actually build our AI narrator. There were many options for us and many different AIs to choose from and I thought we could just build it directly on Claude. We eventually came up with three options: option1 would be to buy claude API, option 2.1 would be to buy a server and option 2.2 would be to rent a server. Out of these options we realized that we could use all three of them instead of choosing just one. So what we ended up doing is planning to launch the pilot using the Claude API, then possibly moving to a rented server once demand increases, and finally buying our own server once demand is proven and steady so that the investment is worth it.
- What surprised or failed
There were many things which surprised/failed throughout the process of designing our AI narrator. Firstly, when it came to the prototype we had some experience with vibecoding and python but had never coded an app similar to ours before. Because of this it took many failed attempts to get the voice lines to work or even basic functions on the dashboard to work. Even during a presentation many functions did not work such as chat replies, rules lookup and the DM override as well as many others. Another surprise was the cost of AI tokens and how much it would actually cost to run our product. I chose our pricing based mostly on what other platforms were charging which still left us with possible profit but only in the high scenario that I calculated. This means making our app profitable will require a lot of users and most likely fast growth because scale really matters for our product to be profitable (around 34K users to breakeven).
- Collaboration
Our team divided work very efficiently, mainly based on who preferred to do what. We had one person on our team who was far more technically knowledgeable than the rest so we decided who would be the best fit for making the prototype. We had regular meetings to discuss how we could actually make our product feasible and often discussed the actual business model we would use which changed frequently as we conducted more and more research eventually leading to the one we presented. Overall, I learned that effective task division and letting people use their strengths for specific tasks really makes a team far more effective and easier to work in.
- How my thinking about AI and strategy changed
What I learned from this project is how classic information goods have near-zero marginal cost, which is why ad-funded free models work for them but not for AI products, like ours, since they burn tokens. This forced me to completely rethink our pricing structure throughout the course of the project and pivot to a paid subscription system. Another key takeaway I had from this course is the importance of network effects. When making Crit Happens and doing research for it I saw how important networking effects were for platforms like virtual table top platforms and I realized that it would also be key to our app, since more people using our AI narrator benefits the more people use it as it gathers data from played games and allows people who may have never been in a group before (because they didn’t have a DM) to play D&D.
Overall Information Strategy has been a great opportunity for me to learn more about AI and software as a whole and has prepared me to increase my depth of knowledge when it comes to future classes on these subjects.
