Seven years

Principal Research Scientist & Director of Decision Sciences. Seven years at Epsilon this month.

Phase one: build the foundation. DiME started as a visual analytics framework. The goal was to make something genuinely capable. Analysts could wire any view to any other, follow threads wherever they went, and never feel like the tool was the bottleneck. That took a while. The framework had to be right before anything else could be right.

Phase two: make it usable. A capable system that nobody can onboard is just a science project. The next phase was packaging: simplifying the interface, building better defaults, making sure new users could get value without reading a manual. Power users want depth; new users want speed. You end up designing the same feature three times before you get the abstraction right.

Phase three: add a voice. The natural language interface (Ada) came last. Once the framework was solid and the UX worked, we layered on a conversational layer that understood what users were trying to do. Not a chatbot. A collaborator that knows the analytical context, remembers the session, and scaffolds the next step instead of just answering words.

The hardest part I didn’t expect. It’s not the first sprint. Any team can go fast for three months. The hard part is keeping the horizon in view: building for today in a way that doesn’t paint you into a corner tomorrow, while still shipping things that matter now. Maintaining velocity over years means making bets about what the system needs to become, not just what it needs to do right now.

That part I didn’t learn in school.