01
The problem
Plenty of finance-critical systems have no API: a legacy vendor portal, a regional bank, a government tax site. The usual answer is that automation is impossible without an integration, so a person logs in every day and downloads files by hand.
Classic screen-recording bots can drive those screens, but they can break the moment a button moves or a field is restyled.
02
How it is built
- 1Schedule trigger runs the workflow overnight so the queue is ready by morning.
- 2Stagehand block logs into the portal using a persisted browser session, so credentials and cookies carry across runs.
- 3Stagehand block navigates to the invoices and downloads them, described in natural language rather than brittle selectors, so it adapts when the UI changes.
- 4Agent block extracts the vendor, amount, and dates from each downloaded file.
- 5Transform block maps the values to your ERP schema.
- 6ERP block posts the record. If the portal changes in a way the browser step cannot resolve, it falls back to a human on Slack instead of guessing.
03
Why it beats a recorded bot
The browser step uses computer vision and natural-language instructions, so a moved button is far less likely to stop the run. It drives the same screens a legacy RPA bot would, but it survives the layout changes that break them, and every action it takes is logged.
Want this workflow built and maintained for your finance stack?
Book a demo