A lot of the time the client would be better off buying actual custom software. It's very easy to end up with a SAP (say) "configuration" so complex that you'll spend more on maintenance than you would if it were a standalone program.
I agree in theory, but I thoroughly disagree in practice.
If a client could spec/develop good software and cost it realistically, then sure. But I have rarely had clients like that - instead they think the problem-space is simple and end up with a half-baked solution that can't be maintained.
Developing greenfield software is a high risk strategy. (Edit: with a high reward if core function, and not much reward if not core).
This happens because management is doing cover-your-ass methodology of software procurement. If they backed a greenfield development and it failed, then get blamed. But if they choose SAP/salesforce/bigCo, the same failure can be blamed on the vendor, or if it ran over budget, they can claim the vendor is expensive.
'Custom' is scary in a way that 'customizable' isn't.