To outsource something you have to be able to unambiguously specify requirements (otherwise costs blow up as you go back and forth). Once you’ve made it truly unambiguous, the next logical step is automation.
Not all automation is economically efficient - i.e. it's cheaper to hire outsourced janitors than building state of the art cleaning robots to automate their jobs.
> Not all automation is economically efficient - i.e. it's cheaper to hire outsourced janitors than building state of the art cleaning robots to automate their jobs.
And a trap many companies large enough to afford full time janitors don't realize: with outsourcing janitors they're relinquishing a lot of control towards the service provider and the work quality. The result more often than not is disgruntled employees waiting for days for stuff such as replacing a cracked toilet seat to be done.
This does not follow, since there are things that machines cannot yet do but that humans can. Producing new humans is one such example. It is not at all clear whether "CEO-ing" is another example.
It might be different now, but prior to ML approaches working for captchas people absolutely paid for outsourced labor to solve captchas all day. And that was an intentionally un-automatable problem.
You can see it even more clearly with "consultancy." The reason consultancies can get away with using so many newly minted MBAs, is that what most of them do is just reflect back the to the company what their employees already knew, but in a form that is consumable and justifies the business case for that action.
There is some internal reason why the organization can't just make the decision, even though plenty of employees know what needs to get done. It could be weak leadership, inability to take risks, analysis paralysis, fear of unfamiliar territory, unwillingness to own the risk (CYA) and even more dysfunctional characteristics (like different teams unwilling to talk to each other). Brining in someone external can cut through much of that mess.