Look at it from the other perspective.
You're looking for a new senior developer, starting in the next few weeks. That means with your 4 rounds of interviews, you want to start the first round within the next week or 2. Your engineers have capacity to do the 4 rounds for ~10 applicants in that time.
You receive 1000 applications.
You don't have time to read every resume and start calling people in the next few days, so you build a tool to automate it. The tool parses text from PDFs, tries to categorise it into "experience", "projects", "location", so you can exclude people with <4 years of experience and not in your country, and find those whose projects showed skills you're looking for.
The tool correctly parses 500 of the resumes, and finds 50 who match your experience, location and skill requirements. Now you manually read those 50 properly, and reach out to the ones who stand out.
That tool is the ATS. You don't care about the 500 who couldn't be parsed, or whether some of the 470 were parsed incorrectly. It's not a test of the user's ability to follow rules. It's just a necessity to get hiring done in a reasonable time frame.