In the exploration phase, I mapped out the financial moments where predictions would genuinely matter to a user. This included upcoming bills, recurring transfers, unusual spending patterns, paycheck expectations, and short-term forecasts.
To simulate model logic without connecting to a real system, I sketched a simple structure that mimicked how a predictive loop might behave. The inputs were basic patterns in user behavior and spending. From there, the system generated signals—things like “you usually transfer $200 on Fridays.” Those signals fed into lightweight model outputs such as forecasted balances, anomaly alerts, or reminders. The interface then responded with subtle state changes: cards shifting, highlights appearing, or new modules surfacing when needed.
One of the core features of the prototype is the predictive balance forecast, which presents a seven-day projection directly within the main balance card.
Another feature is the set of behavior-based suggestions that appear only when they’re truly relevant. The design also includes a 14-day spending graph—a simple, animated line chart that projects short-term peaks and dips. On top of these components, the interface itself shifts in real time. Cards rearrange, highlight new information, or surface modules when the model detects something useful. The small changes in layout simulate how a predictive system might respond moment-to-moment, making the product feel more alive and context-aware.
Next projects.
(2016-25©)
