There's an entire spectrum of options here from complete outsourcing (banks and gov) all the way to writing everything yourself (Google).
I've lately found a middle ground that seems to work well - licensing source code or libraries instead of entire closed web solutions. It enables a small team of good engineers to be really productive and ship something to production rather quickly.
You pay for maintenance and bugfixes for what you license and can still retain control of your data and interfaces.
I read something great to this effect recently, but I can't remember where. The gist of it is that adding people increases overhead, so the absolute smartest play is to keep the tiniest team physically possible and give them an absurd amount of leverage - which frequently takes the form of licensing the things that you really -can- commodify well.
The fundamentals of data engineering (https://www.oreilly.com/library/view/fundamentals-of-data/97...) is adamant on this. If you're a new data engineering team, just buy things that download the data you need, because fetching data from APIs is (usually) pretty simple to do with automated tooling. Then your team can focus on the part that we haven't nailed, like making sane models.
I've lately found a middle ground that seems to work well - licensing source code or libraries instead of entire closed web solutions. It enables a small team of good engineers to be really productive and ship something to production rather quickly.
You pay for maintenance and bugfixes for what you license and can still retain control of your data and interfaces.