UiPath Checks EU Training Grant Expenses for an Italian Region

The client is a regional government in Italy that acts as managing authority for EU structural funds across its territory. That includes the European Social Fund, which pays for training for companies and their workers. The region must verify each beneficiary's expense report before it reimburses a training project.

For each project, a controller opened the expense model exported from the regional system, then each attendance register, each trainer's CV and each invoice, and compared them all with one another by hand. Nothing in the process was automated.

Where the time and risk went

• Many projects per call. One funding call spans a large number of projects, and each brings registers, CVs and invoices that all need checking.
• Typing instead of judging. Controllers spent their hours copying names, hours, amounts and invoice numbers into control sheets rather than assessing them.
• No consistency. Each controller ran the same check their own way, and there was no standard control file.
• Exposure to audit. If an expense is reimbursed by mistake, it becomes an irregularity in EU funds, and the region answers for it.
• Grants paid late. Each week spent checking by hand is another week the company waits for its grant.

We did not measure time or cost savings on this project, so we describe the result as a change in how the work is done.

• Verification moved from fully manual to automated.
• Each project goes through the same control, ending in the same workbook structure.
• Controllers check evidence that is already prepared, rather than typing it in.
• Attendance registers, trainer CVs and invoices, three evidence sources, all reconciled automatically with the declared expense report.
• Delivered within a regional government on infrastructure it owns, with us leading from analysis through to testing.

Why grant robots break

Grant evidence never arrives in one shape. Every call, beneficiary and year brings registers with slightly different layouts, CVs in all kinds of formats and invoices that are scans instead of digital files. A robot designed around one sample fails on the next call. In this build each project is a separate work item. Business-rule exceptions, such as missing or mismatched evidence, go to the controller. System exceptions retry. They are handled apart, and the controller keeps the decision. A surprise document halts one project, not the whole run, and can never become a reimbursement nobody checked.

More work on controlling EU funds: an AI internal audit assistant that runs on-premise, designed for a public agency that administers EU funding.