A resume arrives as a file. A recruiter needs it as data: a name, a phone number, the last three employers, a list of skills to match against a job. The step in between is resume parsing, and it happens so quickly in a modern applicant tracking system that most people never think about it. It is worth understanding, though, because parsing quality decides whether your search, ranking, and reporting are built on solid ground or on guesswork.
Step one: getting text out of the file
Resumes show up in every format: PDF, Word, plain text, the occasional image someone photographed with a phone. The first job is to extract raw text in reading order. For a straightforward single-column PDF that is easy. For a two-column design template with icons and a sidebar, reading order gets tricky, and a naive extractor will interleave the columns into nonsense.
Scanned or image-based resumes need optical character recognition before anything else can happen. OCR is good now, but it still trips on unusual fonts, low-resolution scans, and decorative layouts. If a candidate record looks oddly empty, the original file being an image is the first thing to check.
Step two: finding the sections
Once there is text, the parser looks for structure. Most resumes follow loose conventions: a header with contact details, then experience, education, and skills in some order. The parser identifies these blocks using headings, formatting cues, and patterns like date ranges. From there it works inside each block.
- Contact information: name, email, phone, location, and links. Emails and phone numbers follow tight patterns and are usually the most reliable fields.
- Work history: employer, title, start and end dates, and the description under each role. Dates are the anchor; the parser uses them to split one job from the next.
- Education: institution, degree, field, and graduation year.
- Skills: sometimes an explicit list, sometimes phrases pulled from job descriptions, sometimes both.
The output is a structured record. In Apex ATS this becomes the candidate profile you search and filter on, with the original file still attached so nothing is lost if the parser misses something.
Step three: turning fields into something useful
Raw fields are not the end goal. The reason parsing matters is what you can do once the data is structured.
- 01Search and filter. Find every applicant with a forklift certification, or anyone who worked at a competitor in the last five years, across every job you have ever posted.
- 02Keyword scanning against the job. Each role has requirements. The parser's output lets the system check which of those appear in the resume and where, so a shortlist of 12 out of 150 takes seconds instead of an afternoon.
- 03Ranking. Combine keyword matches with things like recency and years in relevant roles to order the list. Ranking is a starting point for a human, not a verdict.
- 04Duplicate detection. The same person applying to three roles under slightly different email addresses is a common headache. Parsed phone numbers and names make it possible to merge them.
- 05Reporting. Source, location, and experience level only show up in analytics if they were captured as fields in the first place.
Where parsing goes wrong
No parser is perfect, and it helps to know the usual failure modes so you can spot them.
- Creative layouts. Two-column templates, text inside tables, and skills shown as star ratings are hard to read in order.
- Ambiguous dates. A resume that lists only years, or uses Present in one place and Current in another, can produce odd tenure calculations.
- Headers and footers. A name repeated on every page can get pulled into the work history.
- Title versus employer. When both are on the same line with no separator, the parser has to guess which is which.
- Skill inflation. A skills section that lists 60 technologies matches everything and tells you nothing. Keyword ranking should weight skills that also appear in the experience section.
What good parsing does not do
Parsing extracts. It does not judge. A high keyword match means the resume mentions the words in your job description, which is not the same as the person being good at the job. Candidates who write plainly and honestly can rank below those who pasted the posting back into their skills section. Use the ranked list to decide who to read first, then actually read them. The parser's job is to save you the sorting, not the thinking.
If you treat parsed data as a first draft of the truth, verify it where it matters, and fix the layouts that trip it up, you get most of the benefit: fast search, sensible shortlists, and reports you can trust. That is a good deal for a step that takes under a second per file.