Cartbreak
3-phase inclusion of groups, built in public and iterated after their response.
01. At a glance
A fulfillment center needs to know the optimal headcount to run at, balancing cost, delivery promises, cross-utilization, and overtime. I built a system dynamics simulation in Stella Architect to find those numbers, and found 3 structural reasons that made it impossible: the model can't track individual orders, can't enforce priority, and measures work in a unit the warehouse doesn't actually use. Paused the project before it produced a false answer, documented exactly where and why the method breaks, and scoped the alternative that would actually work.
02. Cartbreak, live
Our case competition problem statement said "AI personal shopping companion for fashion", but exploring it as a whole revealed that the footwear space was underserved and an afterthought on shopping websites.
Eg: Myntra has 6 filters for sleeve type alone, but only 7 for all of footwear.
03. Usage Data
I was building a digital twin: a simulation of the warehouse in Stella Architect that could be run to optimize for these numbers. The first step is to build the warehouse inside this software, and accuracy was the priority.
04. Pre-purchase awareness
Talon isn't a better shopping filter. It's a shoe-life companion.
The wedge is fit prediction — a foot scan and a quick style read that gets someone a real answer to "will this actually fit me" in their first session, because that's the pain acute enough to make someone download a new app.
But the reason to keep it open is what happens after: care reminders, repair-shop referrals, resale when a pair's done. That's what turns a single purchase into an ongoing relationship, and it's also what makes the sustainability angle real instead of decorative — Talon can only make money from extending a shoe's life if it's still in the picture after the sale.
05. Questions chosen
In order:
Measurement onboarding
2-step AI search
Product choice confidence
and Post-purchase.
The first 3 phases address the pain point and bring you in, the last phase makes you stay.
Phase 1: Measurement onboarding
Data: The avg search term pre-AI was 3-4 words. AI search is an avg 23 words.
The insight is clear: Users are ready for a conversational experience, and willing to put in a greater effort upfront, for precise results and reduced scrolling later.
Phase 2: 2-step search
Data: The avg search term pre-AI was 3-4 words. AI search is an avg 23 words.
The insight is clear: Users are ready for a conversational experience, and willing to put in a greater effort upfront, for precise results and reduced scrolling later.
In this additional step, the user is asked clarifying questions to better understand their request, and deliver fewer but more personalized results.
Phase 3: Product choice confidence
06. User's choice
The next step was to find another solution. System dynamics didn't work, but there is another type of simulation that would: Discrete Event Simulation (DES).
Where this model only ever sees totals, DES tracks every order, worker, and invoice as its own entity, with its own timestamp and path through the warehouse. It solves all 4 problems above by design, not by workaround. It is also significantly more expensive to build, and usually means depending on an outside vendor rather than owning the model in-house.
07. Shareable cards
… was not the optimal headcount schedule. It was 2 things that are less clear, but still valuable:
08. Insights
Case competitions are famous for their short deadlines, life is famous for shorter. What I would do if I were building this further:
09. Next Steps
As the product person in the team, I experienced what it was like to make something that others would rely on downstream, defensibly enough to present to judges. My teammates had to build data architectures, financial models, integration roadmaps, GTM strategies and more based on my work.
Thank you to Pranita Potghan and Mishthi Narang for being the best teammates I could've asked for.
09. What I learnt
As the product person in the team, I experienced what it was like to make something that others would rely on downstream, defensibly enough to present to judges. My teammates had to build data architectures, financial models, integration roadmaps, GTM strategies and more based on my work.
Thank you to Pranita Potghan and Mishthi Narang for being the best teammates I could've asked for.