A Velros AI ops team versus RPA, and where each one fits

A macro memorises where things are on the screen

Screen automation does not understand what it is doing. It memorises the order: click the third field, read the fourth line of the next window. For as long as the order holds, it is faster and more accurate than a person.

So when the screen changes, it stops that morning

The store moves a button, the bank redesigns its login, someone inserts a column. The order no longer matches. The tool reads the wrong cell without knowing it is wrong, or halts. Usually the person who notices is a customer.

The cost lands on whoever repairs it, not on the licence

In the first quote the licence is the visible number. What accumulates is the time spent realigning the steps every time a screen moves. The more work you automate, the more there is to repair, and often exactly one person in the company knows how.

What happens at the first exception decides everything

"I lost the order number, can I still get a refund?" A memorised sequence stops there. Something that reads the sentence works out what is missing, drafts the question that would fill the gap, and sends only the judgment up to a person. This is why McKinsey placed the impact of generative AI on activities that require understanding natural language, which it put at roughly a quarter of working hours.

McKinsey, The economic potential of generative AI, 2023

It leaves a standard behind, not a script

When a member of staff corrects a draft, the wording and the reason for it stay with the company, and next week there is less to correct in the same situation. A memorised sequence does not learn from being fixed. The same correction gets made by hand, again.

There are still places where a macro is the right answer

A fixed form, a very high volume, effectively no exceptions. Moving a named file into a named folder; pasting the same table into the same cell every morning. Where there is no judgment, there is no reason to attach something that judges.