[sgt101]

My experience is that applying the ML to a domain is 99/100ths of the job; once you've got some traction with well known methods it's time to develop techniques that address their failings. But, any company that's done the hard yards of getting something working is definitely adding value. A good example is NLP, there's no rocket science in building embeddings, but any company that can build a specialist embedding for an important domain (medicine, engineering, finance) has an interesting and useful artefact that might be worth paying for (probably not as much as they would like though).