Falcom seems to have a problem with being overly ambitious. Wasn't Sky supposed to be one game that was later split into two? Then they were going to make a sequel in Erebonia but decided it wasn't time yet so went to Crossbell instead? Then decided they needed another Sky game to bridge the gap? And Cold Steel I and II were supposed to be one game as well.
I mean, kudos to them for actually delivering on such ambitious projects, but you'd think they'd plan a little more accurately by now, haha.
I'm pretty sure they have a planned direction on things. Kondo himself has said that they know they're about 60% of the way through the story, and that they can finish it in approx. 6 games, give or take a game or two.
I think in some cases, like the going to Erebonia after Sky, he said that they just felt like doing Crossbell next instead. I think that the Crossbell stuff started, then realized that maybe it would be easier to do something else instead of loading Zero with a TON more exposition for things they didn't realize they needed to put into place for people to understand better... thus making a new game in between the two.
Which probably played out in their favor a bit, too. Because it also meant that Zero and Ao wouldn't be split over consoles. Had they gone through straight to Zero, I bet it would be likely they have released Zero just before getting dropped by Softbank, thus releasing Ao to PSP only.
I think the Softbank issue that happened may have also been a catalyst in changing a few of the games and the way they developed them, on top of everything as well. (There's a lot about that re: Ys Seven and I think it was the big victim of the situation, but that's for another time/thread.)
I think Cold Steel was riddled with a bunch of issues, because not only was Falcom changing the graphics engine they had been using for the series for ten years already, but they hadn't done something to that scale in PhyreEngine before, too. (They had stated that Sony had to help them with Celceta, as an example.) I feel that Cold Steel II improved a LOT on issues they had with it. I'm also pretty sure Falcom's basically accepted that a Kiseki arc will consist of 2 or more games (and thus the approx 6 games means 2~3 more arcs). So I imagine it was more of a matter of finding where to split the story and keep going.
However, when they were working on Cold Steel, they probably didn't realize they would have the issues they had with some of Cold Steel II's content, as Kondo did say it was capacity/performance related. Thus Cold Steel has grown into a third game. Before Cold Steel II came out, Kondo had said that the 'next game' would be in Calvard. Thus the CSIII wasn't exactly 100% in the books quite yet.
I think it's not entirely a planning thing- because you can plan for years in advance, but still run into issues. Or find something that you need to tweak story-wise, or think that something else needs to be added, or heaven forbid re-order some of the reveals of the story. You will see that this even happens with long planned out TV series.
As an example: Carlton Cuse and Damon Lindelof were pretty open on the development, planning, and writing behind Lost during its airing, as an example. They knew how the story would end from the get-go, but started making changes based on the availability of actors and how the viewers reacted to characters. Roles were switched around during planning, and situations were changed because certain things didn't work out well with the audience. That sort of thing.
I imagine that creating a story to the scale of Kiseki is probably not that different. Especially since it seems like a 15-20 year endeavor for the company.