AI-First Overlays: How a Fintech Startup Balanced Automation and Control

AI · 5 min read

PennyFlow built an AI overlay that suggested categorizations, budget adjustments, and micro-savings nudges in-context across the app. Early internal demos impressed stakeholders, but beta testers reported feeling “pushed” by suggestions and became wary of opaque automation making decisions about their money.

Design and product teams ran a mixed-methods study combining diary studies with an in-app feedback prompt. Findings showed users wanted assistance but also transparent control: easy editability, clear provenance (“why this suggestion?”), and an undo path. The team rewired the overlay to be suggestive rather than prescriptive — suggestions became labeled, editable, and accompanied by concise rationale and confidence badges.

To maintain speed, PennyFlow introduced a two-tier interaction model: “Auto-suggest” default on, but with one-tap rollbacks and a persistent audit trail. They also allowed users to lock categories or set “hands-free” rules. Post-launch metrics indicated a 27% increase in accepted suggestions and a 19% reduction in negative feedback about automation.

This case illustrates a recurring design lesson for AI-enabled startups: automation must be paired with transparency and control mechanisms. When users can understand and reverse AI actions, trust and adoption follow.