Almost Ended Up Looking Ugly — A Look at the Making of Toss Front 2

Why Read This from an HCI Perspective

This article frames a payment terminal not just as hardware, but as an interaction problem—how people actually touch, look, and experience failures in the real world. In particular, it keeps asking whether things are truly ‘easy to use on-site,’ not merely whether they are ‘technically possible,’ examining details like NFC placement, the replacement mechanism, and preventing malfunctions. That makes it a valuable reference for HCI and UX practitioners. It also clearly shows how product design finds a balance among usability, aesthetics, and operational maintainability.

CIT's Commentary

The core of this case isn’t about adding features—it’s about imagining inconvenient scenarios first and then redesigning the structure accordingly. Problems like NFC not being detected on a narrow counter, or having to physically replace the reader when it breaks, can’t be solved by thinking only about the user journey; they require considering the operational journey as well. What’s especially interesting is that the team didn’t just increase convenience—they intentionally constrained intervention paths to reduce malfunctions and visual confusion. In safety-critical systems, a good interface isn’t about having more buttons; it’s about a structure that clearly tells users what to do and when. This approach can carry over directly to LLM-based AI agents or kiosks, and it expands into research questions about designing from the start where users can intervene and what recovery paths exist when things fail.

Questions to Consider While Reading

• Q.I’m curious how discomfort for on-site users and operators was measured differently during the process of changing the NFC placement and the replacement structure.
• Q.For functions that require user intervention—like swapping the reader—what design principles were used to reduce malfunctions while still giving users a sense of control?
• Q.How can the problems discovered in this kind of hardware interaction design be applied to failure modes and recovery design for LLM-based AI agents?