FocalFlow: Facilitating Decision-Making in Multi-Step Mobile Tasks for Blind and Low Vision People

Multi-step decision-making tasks, such as booking a hospital appointment, on mobile devices present a two-fold challenge for blind and low vision (BLV) users: they might struggle to gather relevant information and make informed decisions, as screen readers present task-relevant elements sequentially alongside irrelevant ones. We present FocalFlow, an LLM-powered system that (1) decomposes tasks into actionable subtasks and provides an error recovery process, (2) dynamically filters out task-irrelevant information while preserving the original interface layout, and (3) utilizes progressive disclosure to present critical information first, with decision-support details available on demand. Our evaluation with twelve BLV participants demonstrated that FocalFlow enables them to engage in more deliberate decision-making. We then discuss design implications for future non-visual systems to balance automated information curation with user agency.