IBM have built several, individually successful pieces of technology, all branded as "Watson", and used the impression that they are a single entity to market it as far smarter than it actually is.
Under the hood "Watson" technologies seem to fall into 3 categories:
- The very successful publicity stunt on Jeopardy. The technology here was good, although from what I understood from tech talks given by IBM, it was a bit more brute force than you might think.
- Custom built software, often for medical and legal purposes, that is effective, if limited in scope, and comes with a hefty consulting and support contract. You're essentially buying a data science team at IBM.
- Productised machine learning APIs, that aren't particularly great, and are very isolated in their abilities.
(1) is fun but not useful to anyone really, (2) is useful, but expensive and no better than what you'd get from most other consultancies, or might be worse knowing IBM's consulting practices, and (3) is too limited to build the core of a business on, so will likely only ever be used to provide a step up until companies move technologies in house.