Rework of a grant application form for a public agency

The agency ran a small-business grant and had a form that took most applicants several sittings and rejected a large share on technicalities — a wrong reference format, a missing attachment, a date entered in a way the system did not accept. Officers were spending their time on correction correspondence rather than assessment, and applications from the businesses the grant existed for were the ones most likely to fail, because those applicants had the least administrative support. The agency's own framing was that people were not reading the guidance, which was true and was not the problem.

Contract product designer for about five months, working with an agency service manager, two developers from their supplier, and a policy officer who had to approve every word. It was my first public-sector engagement and the review cycle was unlike anything commercial — every change went through an accessibility check and a policy check, both of which were slower and better than what I was used to. I ran the user testing, including three sessions in a regional office with applicants who had previously failed.

I owned the form's structure, the question sequencing, the error and validation design, and all of the on-screen guidance. That meant taking a twelve-page document-shaped form and turning it into a sequence of short pages with one idea each, designing the save-and-return behaviour, and rewriting every error message from a statement of what the system rejected into an instruction for what to do. I also produced the accessibility annotations the developers built from, and I did the checking on a screen reader myself for the first time, which I got wrong twice before the agency's specialist corrected me.

The binding constraint was that the questions themselves were fixed by policy, so I decided early to spend the whole design effort on sequence, guidance and error recovery rather than fighting for question changes I would not win. Within that, the significant call was moving every document upload to the end and letting people submit without them, with a defined window to supply them afterwards — the policy officer found a precedent that allowed it, and it removed the single biggest abandonment point, which was someone realising mid-form that a file was on another computer. I also chose plain sequential pages over an accordion on one page, against the supplier's preference, because it works better on a phone and on a screen reader and is easier to save halfway through.

Accessibility conformance was mandatory and audited, not aspirational, which shaped several decisions and improved most of them. Every piece of guidance text had to be approved by the policy officer, so I wrote in their register rather than mine and learned to submit copy in batches. There was a legislated opening date for the grant round that could not move, and the supplier's development capacity was fixed at two people, so I designed to what they could build and cut a summary-review screen I wanted. The old form also had to keep working for in-flight applications for six weeks after launch.

Abandonment fell from roughly fifty-five per cent to twenty-nine in the first round on the new form, and rejections for technical errors fell by about two thirds, which is the figure the service manager reported upward. Officer time on correction correspondence dropped enough that two people were moved back to assessment. The accessibility audit passed at the level required with two minor items. I do not want to overstate my part: the deferred-upload change came from the policy officer finding a precedent, and it probably did more than everything else combined.

The policy officer found the precedent that made deferred uploads legal and rewrote my guidance copy into language the agency could stand behind. The agency's accessibility specialist corrected my screen-reader testing twice and taught me to annotate properly. The two supplier developers built the save-and-return behaviour and told me honestly when a design would cost three weeks, which let me trade it away early rather than late.