I've been part of two efforts, one at a very large company, one at a startup. From where I stand, your biggest issue is going to be getting a sufficient data s

I've been part of two efforts, one at a very large company, one at a startup. From where I stand, your biggest issue is going to be getting a sufficient data set, and having sufficient traffic in whatever you're recommending to be able to actually test your models. Technical aspects in how you train your models and such are fun, but way, way down the list of things that are likely to matter in the short to medium term. Like, data scientists are nice to have, but you're not really going to be able to fully utilize them until you have the capability to build, deploy, and test a model at scale. If going third party helps you do this, you probably should.