A no-code automation that matches customer invoices to accounts, emails them out with the right attachments, flags overdue payments, and reports back on what it did — without touching any company back
This was built for a Customer Success role managing a growing list of B2B accounts, where sending monthly invoices meant manually matching PDF files to companies and writing near-identical emails by hand. I described the constraint upfront — no access to company backend systems — and worked through the logic together: match invoices to accounts by ID, skip anything already sent that month, flag accounts with two or more unpaid months, and email a plain-language summary after each run. The AI wrote the matching logic, built the n8n workflow node by node, and generated realistic sample data to test against. I made the calls on what counted as "AI" versus plain automation, which fields mattered for the business logic, and where a human review step was worth keeping versus where full automation was safe.
A trigger button hosted on a static page kept failing with a browser security error. The real cause turned out to be that the automation platform's link wasn't actually turned on yet — the error message pointed somewhere else entirely, and it took directly inspecting the platform's internal state to find the actual problem.
Started with a plain-language description of the manual process being replaced, then answered targeted clarifying questions about data sources, edge cases, and thresholds before any building started. After that, changes came one at a time — add this field, fix this specific bug, adjust this one rule — rather than large rewrites.