IBM automated 94% of its HR routine work. CNBC's headline for this: "Companies replaced workers with AI, and now they regret it."
How can both things be true at once?
Imagine a restaurant that buys an industrial dishwasher and lets go of two human dishwashers. A month later it turns out the machine can't get burnt-on grime off pots - those still need soaking by hand. They hire one dishwasher back.
Is the dishwasher a failure? Quite the opposite - it's still washing 90% of the dishes. The owner just didn't think ahead about what it can't do, and who covers that. AI didn't fail here - they drew the line in the wrong place.
After a similar experience, CBA bank said it plainly: "we misjudged the roles" - not that the AI didn't work. In other words, they didn't fully understand what their call center employees actually did. And once the AI bot couldn't keep up, they realized the people were already gone.
In my experience, this is the typical case. Most owners don't know what their people actually do all day. They know the work gets done, more or less. And that invisible 6% of activities (often more) is exactly what AI can't handle.
Where the line gets drawn
IBM automated employee Q&A, standard procedures, repeat requests - everything with one clear-cut answer. What's left is the 6% of situations with no single right answer.