Citizen & Employee Experience

• Natural-language intake. The applicant describes what they need in their own words instead of completing a form, and the assistant assembles the structured request.
• Arabic-first bilingual design. Arabic treated as a primary experience with correct right-to-left layout and terminology, not a translation layer added at the end.
• Accessibility to WCAG AA. Built in from the first screen, which is both a public-sector obligation and a measurable improvement for every user.
• Continuity over restart. A journey that can be paused, resumed and recovered when information is missing.
• Designed for exceptions. The unusual case is handled gracefully rather than dumped on a call centre.
• Proactive status. The applicant is told what happened and what comes next, which removes most of the chasing.

We run usability testing with actual users, not internal reviewers, before a service goes live. Task completion, time on task and the points where people give up are measured and designed against.

This is also the cheapest stage at which to find a problem. A journey flaw found in testing costs a day; the same flaw found after launch costs a public service incident.

Most AI delivery treats the interface as the last five percent. We treat adoption as the deliverable, because an entity is judged on whether citizens used the service, not on whether the model was accurate.

Why does an AI assistant need accessibility work if it is conversational?
Because conversation is only one of the channels. Users still read status, upload documents, review summaries and confirm decisions on screen. Those surfaces carry the same accessibility obligations as any other public service, and for many users they are the primary interaction.

Is Arabic support just translation?
No, and treating it as translation is the most common reason Arabic services underperform. Arabic requires right-to-left layout, correct administrative terminology, bilingual document handling and testing with Arabic-speaking users. We design it as a first-class experience rather than a localisation pass.

How do you measure whether the assistant is working?
By completion rate, time to outcome, exception rate and the proportion of journeys that need human intervention. Satisfaction scores are useful context, but completion is the measure that tells you whether the service replaced the old way of working.