Correct me if I’m wrong [… dear reader, if there is any one] – I seem to have missed in the brief on RPA:
- It generates human position (fte-replacing) point solutions for workflow automation for transactions that can be interpolated in the training set(s), doing very little in the full stretch of Straight-Through Processing (yes I happened to have been around in the ’90s/’00s) or at max taking it step by step;
- It does nothing for ‘outliers’ outside the training set(s) as these are doomed to be mistreated;
- Happenstance circumstances may fill the training set(s) but clarity re seasonal patterns, specific (marketing et al.) actions, etc., is missing;
- Clarity re the correctness of transactions (often screen-scraped) for the training set(s) is lacking;
- When taken in a series of sub-implementations, may help in STP;
- RPAAI (Unassisted RPA, with (only now?) an ‘AI’ component) may go just a little further but not much [as per above].
Not a problem, if one is a consultant selling this stuff. A problem,
- When not if one is to be replaced (in total, or for a major part of one’s job) by some API contraption
- OR when one’s dependent on less-than-perfect implementation and hence outcomes for one’s free life (the AI bias issue again, buried / token/scapegoat human)
- OR when one’s a concerned member of society seeing already waaay too much sh<third-person pronoun> being automated with too little reason compared to the crippling (to be) dependence on brittle automation. Because it’s brittle by design, being designed by humans to be optimal for the situation in the blip of the moment that it was [designed].