Designed for the Margins, Better for Everyone: Accessibility-First Interface Design and Team-Wide Error Rates
Abstract
We compared error rates and task throughput on an internal data-entry tool between employee teams using an accessibility-first interface (larger touch targets, high-contrast text, plain-language error messages, and adjustable text sizing, developed with input from employees with visual or motor impairments and older employees) and teams using the organization's standard interface. Across all users — including the large majority with no reported impairment — the accessibility-first group logged 26% fewer entry errors (3.1 vs. 4.2 per 100 entries) and completed 18% more tasks per hour (38.4 vs. 32.6) than the standard-interface group over an eight-week rollout period.
1. Introduction
Accessible design is usually justified as an accommodation for a minority of users, with any benefit to the broader user base treated as an unplanned side effect rather than something worth measuring directly. We asked whether an accessibility-first version of a routine workplace tool would measurably improve error rates and throughput not just for employees with impairments, but for the user population as a whole.
2. Method (Summary)
We tracked 284 employees across matched teams at a single organization who used one of two versions of an internal scheduling and data-entry tool over an eight-week period, assigned by each team's existing rollout schedule rather than individual randomization. The accessibility-first version (n=142) incorporated larger touch targets, higher-contrast text, plain-language inline error messages, and user-adjustable text size, developed with a working group that included employees with visual or motor impairments and employees over 60. The standard version (n=142) retained the tool's existing dense, default layout. Entry errors per 100 records and completed tasks per hour were logged automatically by the tool; employees also self-reported whether they had any visual, motor, or age-related condition that affected their tool use.
3. Findings
The accessibility-first group logged 3.1 errors per 100 entries versus 4.2 for the standard-interface group, a 26% reduction, and completed 38.4 tasks per hour versus 32.6, an 18% increase. Restricting the comparison to the 213 employees who reported no visual, motor, or age-related condition affecting tool use, the gap held at a similar magnitude: 3.3 vs. 4.2 errors per 100 entries (21% reduction) and 37.9 vs. 32.9 tasks per hour (15% increase). The subset of employees who did report a relevant condition (n=71) showed a somewhat larger per-person error reduction than the unaffected subset, but the gap among unaffected employees was still large enough to account for most of the team-wide difference in aggregate error counts.
4. Discussion
The results suggest that design choices intended as accommodations for a subset of users can reduce ambiguity and cognitive load for the whole user population, not only the group they were designed around. One interpretation is that clearer contrast, larger targets, and plainer error language reduce the same kind of momentary confusion in unaffected users that they were built to remove for affected ones — a shared reduction in friction rather than a specialized fix. Operating costs were not directly measured here, but nothing in the design suggests the gain to unaffected users came at their expense or at any other group's expense; it accrued alongside the intended benefit to employees with impairments.
5. Limitations and Future Directions
Teams were assigned by existing rollout schedule rather than random assignment, so team-level differences unrelated to the interface cannot be fully ruled out. The tool studied was a single data-entry application; whether the pattern holds for more complex or judgment-heavy software is untested. The eight-week window may not capture longer-run adaptation effects in either direction. A planned follow-up will randomize interface assignment within teams and extend tracking to six months.
Editorial Note
This report underwent CfAHC's Rapid Review Protocol, including AI-assisted literature cross-checking and a two-reviewer human sign-off.