We bought a robotic process automation tool four years ago and built forty six processes with it. Most of them drive an application's user interface, clicking and typing exactly as a person would, because for the systems involved that was the only way in. The business case claimed eleven thousand hours a year, and I still think that figure was roughly honest.
In May a supplier shipped a routine interface update. Twelve of our robots stopped, which was a nuisance. Three did not stop, and spent two days entering values into fields that had moved, which was considerably worse and took two weeks of finance effort to unpick.
The maintenance burden is real and I will not dwell on it. The part I got wrong was strategic. Automating the screen removed the reason to fix anything underneath it.
One of those forty six robots exists to copy data out of one of our own systems and into another of our own systems. Both are ours. A proper interface between them would have taken a few weeks and sat behind support contracts we already pay for, and it was never built, because the robot worked and nothing that works gets funded. We had taken a visible integration gap, the sort that appears in an architecture review and eventually attracts money, and converted it into an invisible one staffed by software pretending to be a clerk.
Automation is classified now by what it is built against. Anything driving a documented interface, where the supplier has committed to a version and a deprecation notice, is permanent and supported like any other system. Anything driving a screen is temporary by definition. It carries an expiry date, a named owner and an annual justification, and its record must state what the proper fix would cost, so that number stays in front of somebody who can approve it.
Fourteen of the forty six have been retired. Five were replaced by the interfaces we should have built originally. What remains are honest workarounds with dates against them.
Screen level automation is a loan with no paperwork attached. We took out forty six of them and recorded it nowhere a finance director would ever look.
– Serguey Shinder
