RPA · 4 min
How an RPA bot fails, and why that is the lesson
The impressive demo is the perfect file. The useful bot is the one that halts when the file is not perfect.

Bots get shown on clean data. Offices are not clean data. A date is missing, a column moved, a download did not finish. If your bot continues anyway, it is inventing a result.
Write the stop before the shortcut
List the ways the task already goes wrong when a person does it. Each of those is a branch. The branch should end in a log line another person can read: which file, which field, what you expected.
Not every task wants a bot
If the next step depends on a judgment call, leave it with a person. Automation is for the part you can describe without shrugging. We would rather ship a short bot with a clean stop than a long one that guesses.
- Describe the clicks before you record them
- Keep one sample that is supposed to fail
- Read the log as if you did not build the bot
