Robotic process automation case study: passport and visa paperwork, from about 50 minutes to under 15

The challenge: making complex forms easier to complete correctly

The client runs a service that helps people with US government paperwork: passports, renewals, visas and other forms. The rules behind those forms are complex, and an application submitted with a missing or invalid field can be returned. Before the build, applicants often faced three sources of delay:

• Incomplete forms. Applications often went in with something missing, which meant resubmissions and delays.
• Failed payments. At busy times, roughly 15 to 20% of payments failed the first time, from a gateway error, a bank decline or a checkout that timed out. People tried again, and some checkouts never went through.
• Time. An average application took about 50 minutes.

Applicants hand over personal details, so the platform had to be secure without being any harder to use. The goal was fewer delays, less stress, and an application approved the first time it went in.

What we built: RPA for the form work, rules for the assistants

The steps in these applications are set by the government's requirements: which form applies, which fields it needs, what a valid answer looks like. So the assistants that guide each applicant run on rules, not a model, and a bot takes on form work that used to be done by hand. That's robotic process automation in its plain form: software following set steps.

Five pieces do the work:

• Plain questions, then the right forms. The applicant answers simple questions in plain words. Rule-based assistants, written from the government's requirements and the details of the application, turn the answers into the correct forms, and the platform checks for mistakes.
• A bot for the form work. Puppeteer, a library that drives a web browser from code, automates form-related tasks and cuts the manual work.
• Payments that retry on another route. When a payment fails, the platform tries it again through an alternate payment route, so the applicant doesn't type the card details in a second time. Failed transactions fell.
• Tracking and live status. Each submission is tracked, with live status updates, and customer support sits on the same platform.
• A courier when time is short. An applicant who needs documents fast can be put in touch with a government-registered courier, who takes them to a passport office by hand.

Applicants move through the forms one step at a time, in a Next.js front end built with Material UI and CSS. On the back end, NestJS runs the business logic and the long-running processes. Data and the live status updates run on PostgreSQL, MongoDB, Supabase, GraphQL and WebSockets, and hosting and file storage are on Vercel, AWS (EC2, Lambda, SQS and S3) and Supabase Edge Functions.

Where the automation stops, and who takes over

The platform asks, fills in, checks, takes payment and tracks. Two things stay with people: carrying documents to a passport office when an application is urgent, and the decision on the application, which belongs to the government.

Rules and a bot were enough here because each automated step applies a requirement that's already set. Where a step needs judgement, we put AI on that step and keep plain rules on the rest.

Results

Since launch, applications are quicker to finish and simpler to follow, and fewer of them stall at checkout or come back for a missing field.

Technologies and tools

Timeline

The build took eight months.

If your team still types the same details into web forms, our page on back office automation sets out which process we'd automate first, and where AI earns a step next to the rules. For a web platform built around your own process, see what we do as an AI app development company.