While COBOL was the first highly successful business-oriented programming language, several business-oriented languages were designed before it in the late fifties, including Honeywell-800 Business Compiler a.k.a. Fully Automated Compiling Technique (FACT).
In History of Programming Languages page 209 Jean Sammet remarks on the comparison between that language and the in-development COBOL:
since ... FACT was really a good and advanced language, any technical comparison would usually be on the side of FACT. However, the difficulties of dealing with machine independence, and the interaction of competing manufacturers certainly made the creation of a common business language orders of magnitude more difficult.
(Emphasis in the original.)
While I can certainly see that FACT didn't need to be machine-independent and COBOL did, and that machine independence would require a little more care, I am very surprised by the assertion that it was 'orders of magnitude more difficult' (even allowing for hyperbole). I would expect, for example, a business-oriented language to specify data in characters and decimal digits (as COBOL does), therefore naturally not care about things like byte size, endianness, or ones versus twos complement signed integers, that might cause trouble for a lower-level language. But perhaps there were other particular difficulties at that time, that I am not used to taking into account.
What exactly was machine-specific about FACT? What were the big difficulties in achieving machine independence?