Anecdotally, the main business value I've seen from ML/AI tech has been in cases where 1. A basic solution shipped and made a ton of money e.g. Ads, Search, re

Anecdotally, the main business value I've seen from ML/AI tech has been in cases where 1. A basic solution shipped and made a ton of money e.g. Ads, Search, recommendations etc. 2. It is financially feasible to have a dedicated team(s) make small incremental progress on these solutions. Even very small gains are beneficial. 3. The business perceives a threat if they fall behind in this area. The thing is that the gains on the basic solution (heuristics, off the shelf pre-trained CV model, open voice recognition) are pretty small, and if the threat of others making progress goes away - the inferred value of further investment will probably vanish as well. Other applications which put the AI in the driver's seat (sometimes literally) seem far from production - or if they do work, then they work reasonably well using an alternate approach from what you might expect.