Good post David. One thing you mention is the cost of ML talent which drives the cost of using ML in an organisation. There is also the fact that these projects take a lot of time to implement and some organisations are frustrated to hear talking about algorithms when they just want to “get on with it”. I often hear CMO, COO and others telling me: “We’ve been working on this model for weeks but I don’t see much coming out”. Or “We have this ML based model and it works but we really struggle to scale”. Or even “There is a disconnect between my DS and my analysts. The former know machine learning but not my business. The latter know my business but go blank when talking about algo. I’d love to have my data analyst be able to produce models (i.e. remove skill gap) and do so in minutes rather than months (speed of efficiency)”. Or “Our DS suffer from cognitive biases and it’s tough to extract interesting features and exploring smart/dumb datasets is lengthy”. These are all operational challenges slowing the adoption of ML. Which is where auto-ML will actually help (auto ML is a solution to build supervised models quickly, explore smart data rapidly and introduce agility between DS team and stakeholders). As I take PredicSis.ai to market (an auto ML solutions used by big companies like AMEX, Orange, EDF and by a range of small startups around the world), these are the pains I hear being expressed on a daily basis. As you rightfully mentioned, the potential of ML is huge. As more and more organisations start to think auto ML (big ones and tiny ones thanks to a cloud based, pay per use modelling for $3ph) we will, I believe, see a widespread adoption of ML.
Sorry, that was a long note, as Blaise Pascal said, I didn’t have much time to write a short one but hope it makes sense and adds a little bit of value.