The figure will vary somewhat depending on things like dev experience, previous coding practices, automated testing etc, but there is a corpus of research that suggests that a ballpark figure is somwhere around that 40% is about right. Here's what Perplexity says about it https://www.perplexity.ai/search/what-evidence-is-there-that....

Seems plausible here, particularly if allow for the often considerable additional preemptive Maintenance effort required trying to avoid breaking the existing product while adding new features.