The story
I went to open a digital account at a very non-digital bank. The kind with marble floors, polite staff, and where you bring a stack of papers to be admitted.
The woman behind the desk was lovely — patient, efficient, clearly doing her best. But her real job wasn’t customer service. It was competitive data entry.
First, she scanned the papers. Then she began typing. My ID number, my name, my company address, my home address, my date of birth — all fed dutifully into one system. Then, the same data again into another. And again into a terminal window that looked like it had been designed by someone who hated joy. Watching it felt less like banking and more like observing a Victorian factory: clack, clack, clack — the machinery never stops.
By the tenth time she re-typed my ID number, I was half-expecting a cheerleader to appear and shout, “Go team!” This wasn’t fraud prevention or compliance. This was just machines that didn’t speak to each other and outsourced their stubbornness to a human being.
Forty-five minutes later, account open, papers signed, handshake done.
The very next day I got a notification from the bank’s shiny app. Unless I uploaded all the same documents again and digitally signed the same forms, my account would be suspended in 45 days. It was like déjà vu, but with paperwork.
The solution
- 10 years ago RPA was all the rage.
- Now we have LLM-driven agents that will plan the actions for the process, ask for the required information upfront then proceed to enter the data in the right places.
- The plan can be saved as a recipe for later usage.
- If the agent encounter errors, the user can take over, correct the input, and let the agent resume its flow
- Always start small, and build trust on your way up!