Over the period of their existence as a startup they kept stats on productivity, very agile process, lots of history tracking feature requests, time to implement, bugs found post implementation, individual code/documentation change rates, defect exposure (when bug is fixed, where did it come from, how could it have been stopped, etc) and while doing this they ended up moving twice (growing startup kind of thing) and had groups and individuals in offices/open plan/cubicle setups.

Basically this guy was really (and I mean really) analytical about characterizing as many things as possible to understand whether or not they had impact on the overall ability of his company to produce quality products and support.

What struck me though was IBM was completely unwilling to engage on the facts. And that is largely because IBM is run by their finance group.

And the CEO here wasn't trying to tell IBM that every facility should be run that way, just that they felt justified in demanding that their facility be run that way.

Anyway, IBM couldn't (or wouldn't which is expressed externally in the same way) see it and proceeded with their plan. The entire time I was there helping to integrate our startup (Blekko) I saw IBM make stupid choices like this again and again.

I see this pattern as the biggest contributing factor to IBM never getting out of the starting gate with machine learning as a product enhancer. Trapped in their own complexes and being unable to step outside of them in a controlled way left them unable to execute against what could have been a really good strategy for them.