For what it's worth, I've got both PnP machines and manual gear.
For a run of 10 of those boards, I'd go for PnP - but I don't have access to minimum-wage labour to do the hand assembly - that might swing things back to manual. Perhaps all the easy components (passives, easy silicon) by machine, and the tricky chips and connectors by hand (well, silicon on the BGA / CSP placement rig, connectors by hand). That volume and complexity does put the job safely into 'there's no ideal way, so it's going to be a bit expensive' category.
Component sourcing for small batches is a pain, too - stuff tends to turn up in utterly machine hostile forms - short strips, single chips in bags, and just plain wrong, all of which pushes the balance towards manual assembly. Having the machines toss away components because they're wrong is more annoying than having hand assembly stall while people scratch their heads for a while.
Going to 'real' production methods too early (and this really does feel like there are a few spins to go) can just be a distraction. Still, it'll give the hardware team something to do while the softies work their way through that enormous and ever-growing 'spec'.
Edit: Obviously, this is just me. The Soap guys are obviously much closer to the project and know much more detail.