Resumap Research · 2026-07-20 · study #4
We ran 36 resume templates through the parser behind SuccessFactors & iCIMS.
You can't get a free trial of SAP SuccessFactors or iCIMS — but you can test the engine they parse resumes with. Per SAP's own docs, SuccessFactors uses Textkernel; iCIMS uses Sovren, now part of Textkernel. So we sent the same 36-template corpus straight to Textkernel's Tx parser API. All 36 parsed: every contact field and every one of the 10 job-critical skills came through on each template, with the richest structured skills list of the whole series. The only wobble was how it split job entries on a handful of templates — a parser quirk, not the resume. Full breakdown and the corpus below.
Every figure is derived from a checked-in dataset; the corpus content is byte-identical to the Zoho, Manatal and Workable rounds.
Earlier in the series: Zoho Recruit, Manatal and Workable.
Why the engine, not the ATS
Most applicant tracking systems don't build their own resume parser — they license one. Textkernel (which absorbed Sovren, long the most widely embedded parser in the industry) is the engine under a large share of that market. The parts of it we can document from first-party sources:
- SAP SuccessFactorsuses Textkernel to parse resumes — stated in SAP's own product documentation.
- iCIMS parses with Sovren, now part of Textkernel.
Neither offers a self-serve trial you could test a resume against. The engine does — so testing it directly is the closest honest proxy for how those enterprise systems read your resume. We tested the parser, not the ATS products; we don't claim to have run SuccessFactors or iCIMS themselves.
How it parsed
All 36 templates parsed. Email, phone, location and education came through 31/36; every one of the 10 job-critical skills was retained on all 36/36 templates; and each profile carried a full structured skills list (median 68 distinct skills).
This is the richest structured extraction of the four parsers. Textkernel returned every template's skills as a discrete list — 54 to 69 skills each — so a recruiter filtering on any of them would surface the candidate. 31 of 36 templates parsed with no defect at all. The remaining 5 each hit the same single issue — work-history segmentation:
| Template | Jobs parsed | What happened (parser-side) |
|---|---|---|
| Aurora | 2 of 3 | merged jobs into fewer entries |
| Beacon | 2 of 3 | merged jobs into fewer entries |
| Bloom | 2 of 3 | merged jobs into fewer entries |
| Lumen | 1 of 3 | merged jobs into fewer entries |
| Offset | 5 of 3 | split jobs into more entries |
All 5are recorded against the parser, not the template: we dumped each source PDF's text layer with the same tooling a parser uses and confirmed all three jobs — distinct companies, distinct date ranges — are present exactly once. Other engines in this series segmented the same files correctly. We publish them anyway.
Four real parsers, one pattern
Across four production resume parsers — Zoho Recruit, Manatal, Workable, and now the OEM engine behind SuccessFactors and iCIMS — our templates' core fields parse cleanly. Zoho's first pass caught real bugs in our templates; we fixed them, and the fixes carried across every parser since, because a clean top-to-bottom text layer is what all of them want. Where a parser slips, it's its own quirk (how it splits job entries), not the resume — and we verify that against the source every time.
Data & reproduction
Everything needed to verify or re-run this test. CC BY 4.0 — cite this page.
The corpus is the fictional Victor Lassoult persona — no real personal data. Same content as the Zoho, Manatal and Workable rounds.
Limitations — read before citing
- We tested the Textkernel Tx parser engine directly, not SAP SuccessFactors or iCIMS as products. Those are named only because their own documentation states they use this engine; their full pipelines may differ.
- One fixture CV (senior software engineer, English, Latin script). Other professions, languages or scripts may parse differently.
- Parsing only. Textkernel sells a separate scoring product we didn't test; no score is measured or claimed here.
- One API call per template, 2026-07-20, EU data center. Parser behaviour changes over time; the test date is part of the dataset.
- Resumap runs this test on its own templates — an obvious interest. That's why the corpus and scored results are downloadable: don't trust us, re-run it.
Questions this data answers
What is Textkernel and why test it?
Textkernel (which now includes Sovren) is an OEM resume-parsing engine that other software licenses rather than building its own. Per SAP's own documentation, SAP SuccessFactors uses Textkernel to parse resumes, and iCIMS uses Sovren (now part of Textkernel). Those enterprise ATSs don't offer self-serve trials, so testing the shared engine directly is the closest you can get to how they read a resume — one run approximates a large slice of the enterprise ATS market.
Did every template parse?
Yes. All 36 templates parsed via the Tx REST API. Email, phone, location and education came through 31/36. Every one of the 10 job-critical skills was retained on all 36/36 templates, and each profile carried a full structured skills list (54–69 distinct skills, median 68). The only imperfection was work-history segmentation on 5 templates.
What went wrong on those few templates?
Nothing in the templates — it was the parser's job segmentation. 5 templates had their three jobs merged into fewer entries or split into more. We dumped the text layer of each of those source PDFs with the same tooling a parser uses and confirmed all three jobs, with distinct company names and date ranges, are present exactly once. A different engine (Workable, Zoho) segmented them correctly. So this is the parser's boundary detection, not a template defect — we publish it anyway.
Does this test cover the ATS match score?
No — parsing only: whether the engine reads your resume's fields correctly. Textkernel also sells a separate scoring product; we didn't test it. Parsing is the gate that matters first: a score can't help a resume the parser couldn't read.
How does this compare to the other rounds?
Same corpus, four different parsers now. Zoho Recruit caught real text-layer bugs in our templates (we fixed them and re-ran); Manatal and Workable read the fixed corpus cleanly; and Textkernel — the enterprise OEM engine — parsed all 36 with the richest structured skills of the four. Each parser has its own quirks, so a clean parse in one doesn't guarantee another, which is why we keep running the same corpus through more of them.
Can I reproduce this test?
Yes. Download the corpus below, get a Textkernel Tx trial (self-serve, free credits), and POST each PDF (base64) to the Tx resume-parser endpoint. Compare the returned ContactInformation, EmploymentHistory, Education and Skills against the source. The scored results CSV lists what we observed per template.
Use a template that parses — then check your own resume
Every template in this test is free to use. And if you want to know how your existing resume reads against a specific job description, the ATS Check parses both and gives you an honest, itemised score.