The original document has the hard facts: Until 2011 they spent 11.7 Mio Euros on the migration and another 2.08 Mio Euros on workflow optimizations that they tackled at the same time. The calculated cost for a migration of the Windows NT environment to a newer windows version including support and the same workflow optimizations are at 15.52 Mio euros. That's a saving of 1.74 Mio Euros or roughly 11.2%. This is taking into account that most migration costs now have been paid, and savings will subsequently will rise. They also have anecdotical evidence that error rates and ongoing support costs have dropped. It's certainly anything but a canonical example for a botched project.
That's a saving of 11.2% comparing an actual cost to an estimated cost; if the estimated cost blew out (which seems to happen a lot in migrations) then the savings could be even bigger.
