Disclosure: I'm an IBM employee working on legacy enterprise architecture absolutely nowhere near Watson. If modern stuff is on the left side of spectrum (AI/ML, CI/CD, Javascript and web frameworks etc), I'm somewhere at the edge of solar system on the right side of the spectrum :)
That being said, my personal impression from the market, and completely unauthoritative, is that Watson is a brand-name, NOT a single product; it's a collection of products, some built, some purchased, some integrated, some stand-alone, some simple, some complex, under the AI/ML/"Cognitive" umbrella. Most companies have branding/pillars - in IBM, Infosphere used to be information management, Lotus was collaboration and communication, Rational for project management / testing / modelling / etc, and so on.
So you have "Watson Predictive Analyzer" and "Watson Chat Service API" and "Watson AI Doodad #37", etc where you can seamlessly take out Watson, replace it with IBM or not at all. There is no single Watson product that I'm aware of. Some stuff was likely even re-branded from previous analytics/BI into Watson branding as deemed appropriate.
Globally-intentionally or not (opinions differ wildly), market perceived "Watson" as "that big machine that won Jeopardy" - a single drop-in product, as opposed a suite of independent products that can certainly do many things, but need hard work at data cleansing, formatting, modelling, testing, etc to do useful things.
I mean, cognitive/watson/AI implementation projects seem to be order of magnitude faster than my experience with Enterprise Architecture projects; but still not a "plug in box, push button".
Word from the inside is that to successfully utilize Watson a lot of careful work has to be done with managing and formatting data... and to some extent experimentation for any given solution. This requires a lot of time from IBM, and a lot of time / resources from the company who buys it.
However, Watson was sold and advertised as more of a drop in solution with a lot promised of results that Watson has never produced.
Accordingly the massive costs incurred by customers were unexpected and the processes to do it right rushed, customer expectations were not met (quite the opposite) and the results were often quite poor.
IBM seems to have backed away from all those grand claims.
Most healthcare professionals I've spoken to consider it an absolute joke, it's apparently a bit of a laughingstock compared to its grand claims of diagnosing all the cancers
It's still around, but now it's generally being sold as what it actually is - an NLP framework for building chatbots and search engines - as opposed to magic sentient AGI that can solve any problem.
