Ideally, serious games are just normal games with a 'serious' topic like pollution, recycling, etc.
Serious games do quite well when the topic is to teach people how to drive a boat, how to fly a plane, etc. Note that these type of games work particularly well when the simulation is realistic (crashing and sinking for example) because the user needs to think that the simulation is realistic for them to believe they can, and actually, benefit from it.
The biggest problem in serious games at the moment is for more 'cognitive' tasks where the it is more difficult to tell whether the user has learned anything or not. Coupled with the fact that the clients don't fundamentally understand what a game is, and in what ways people can learn from it, we get these 'gamified' textbooks.
A client can start off wanting a serious game, but will also want everything he's used to; pie charts, QA's and generally things that don't go well with games at all. This is probably due to the fact that the money that is put into a serious game is considered an investment, and people will want figures to justify that investment.
This is where we end-up with this bastardized serious game that is neither one thing or another and is pretty no more useful than a text-book.
I believe that the solution is for developers to find new and innovative game mechanics that will enable them to properly test the users' knowledge and give the client this information without having to actually ask a single question.