I'm somewhat a statistical geek myself, so his preference to statistical information over estimates make a lot of sense and really sounds good to. Estimating is hard and pretty much useless after a while, when you have some decent statistics about velocity. There is still a place and need for splitting user stories to tasks.
Maybe the biggest surprise was that he doesn't consider Kanban to be a software management process at all, it's something applied to your current process. So starting from scratch is not an option, you need to have a starting point.
Other key point was how correct use of classes of service can drive to customer satisfaction. Classes of service and predictable lead time will drive customer satisfaction without a doubt.
On a side note I got a bunch of new tips what books to read. Specially Switch Switch - How To Change Things When Change Is Hard (Chip, Dan Heath) sounds like a book that could help me fight the emotional resistance that change always triggers in people.
