How can AI filter attention without removing user judgment?
Bottleneck is a product and service-design case about AI-mediated attention filtering. It reframes notification management around classification, privacy, cognitive load, transparency, and user control rather than autonomous optimization.
Bottleneck is a product and service-design case about AI-mediated attention filtering. It asks how classification can reduce cognitive load without quietly removing user judgment, especially when notifications carry social obligations, work pressure, privacy concerns, and different degrees of urgency.
The project reframes notification filtering as a responsible interface problem rather than a generic AI service concept. It uses research, system mapping, and prototype flows to define what the system may classify, what it should explain, and where the user must remain able to override or inspect decisions.
Problem
Notification overload is not only a volume problem. It is a judgment problem: people need to decide what deserves attention, what can wait, what crosses personal boundaries, and what context the system does not understand.
An AI layer can help classify signals, but it can also make attention decisions feel opaque or paternalistic. The design challenge was to reduce noise while preserving agency.
Classification System
The concept explores notification categories, urgency states, sender relationships, timing, and work-life context as inputs to an attention filter. These classifications are treated as inspectable service logic, not invisible automation.
Privacy is part of the system boundary. The service needs enough context to be useful, but the interface must communicate what it uses, what it infers, and how users can tune or disable its behavior.
A series of personas were drawn to highlight the different positions regarding the same digital overload issue
Design Intervention
The interface work includes layered cards, status modes, AI controls, and explanation patterns. The goal is to let users understand why something was delayed, surfaced, grouped, or escalated.
The design system avoids presenting AI as a personality. It frames the system as a configurable filter whose value depends on transparency, reversibility, and user control.
Outcome
Bottleneck produced a research deck, service map, storyboards, interface prototypes, naming work, and a design system for adaptive notification filtering.
As a secondary case, it supports the larger portfolio argument around AI-mediated judgment: classification can be useful only when users retain the ability to inspect, correct, and decide.
