I doubt it would cost nearly that much. I could build something similar in a couple of days, and I'm not even a developer by trade.

Just use something like pHash (e.g. https://github.com/JohannesBuchner/imagehash), screen cap the centre of the screen (say, a quarter of the screen in total) every minute, hash and then send the 8 byte hex string back to home base.

Ah, the "I'm no developer, but this is easy" person. Glad to have your valuable input here :)

"I could build a twitter clone over the weekend!"

The problem you are describing is one that many large companies spend huge resources on today, so guarantee you that it is not something you do in a couple of days.

Even if you had the system up and running, you would still need to create a large database system managing 100 billion data points each day and integrate the stream of information with your customers APIs.

$2m worth of developer time going into this project is very likely. Thinking you could do this in a couple of days is simply ignorant.

Ok. Now get the database of content to match it against and build a program that automatically determines what the user is engaged in.

Ok, I'll just go and buy a major media content publisher. Sony, perhaps...