💡 왼쪽 원문을 읽으면서 오른쪽에 따라 써보세요. Tab 키로 힌트를 받을 수 있습니다.

원문 렌더가 준비되기 전까지 텍스트 가이드로 표시합니다.

> "Recruiting in 2026 is no longer human-only work. AI discovers candidates, screens them, talks to them, evaluates them, and even runs onboarding. But the final hire/no-hire call must still belong to a person." — Josh Bersin, "HR Technology 2026 Report"

HR technology has been one of the fastest-growing SaaS categories for the last five years. According to the Sapient Insights 2025-2026 HR Systems Survey, the global HR tech market is roughly $53B in 2026, and the share of solutions shipping AI as a first-class feature jumped from 12% in 2023 to 63% in 2026. ATS, HRIS, sourcing, assessment, learning, performance, and employee experience — every layer now ships LLM-based or embedding-based matching by default.

The risks have grown in lockstep. The EU AI Act (in force in 2024, fully applicable in 2026) classifies hiring, firing, and promotion decisions as "high-risk" systems. New York City's Local Law 144 (effective July 2023) makes annual bias audits mandatory for any Automated Employment Decision Tool (AEDT). Amazon's 2018 scrapping of its resume-screening AI — which had learned to penalize the word "women" — still appears in every HR tech deck as the cautionary tale.

This guide walks through the entire 2026 AI HR and recruiting stack: ATS leaders (Greenhouse, Lever, Workday Recruiting, iCIMS, SmartRecruiters, Ashby, BambooHR), AI sourcing (Eightfold, Phenom, Beamery, hireEZ, Findem, SeekOut), AI interviews and assessments (HireVue, Paradox, HackerRank, CodeSignal, Codility, TestGorilla), HRIS and EOR (Workday, ADP, UKG, Deel, Rippling, Remote.com), performance and learning, and the Korean and Japanese ecosystems.

1. The 2026 AI HR Stack Map — Nine Layers Along the Candidate and Employee Journey

The HR tech stack splits into nine boxes that follow the candidate and employee journey.

| Layer | Representative products | Role |

|---|---|---|

| Job board / media | LinkedIn, Indeed, JobKorea, Saramin, Wanted, Rikunabi | Job posting + candidate pool |

| AI sourcing / matching | Eightfold, Phenom, Beamery, hireEZ, Findem, SeekOut | Resume + public-data embeddings + matching |

| ATS (Applicant Tracking) | Greenhouse, Lever, Workday Recruiting, iCIMS, SmartRecruiters, Ashby | Pipeline management |

| AI chatbot / candidate experience | Paradox (Olivia), Mya, AllyO, Phenom Bot | 24/7 chat, scheduling, FAQ |

| Assessment / video interviews | HireVue, Modern Hire, myInterview, Spark Hire | Async video, AI scoring |

| Skills assessment | HackerRank, CodeSignal, Codility, TestGorilla, Korn Ferry Sigma | Coding, aptitude, personality |

| HRIS / employee data | Workday HCM, BambooHR, Rippling, Gusto, Justworks, Deel | Employee master |

| Workforce management / payroll | ADP, UKG, Oracle HCM, Workday Payroll | Time, payroll, tax |

| Performance / learning / EX | Lattice, 15Five, Culture Amp, Cornerstone, Docebo, Glint | 360 feedback, learning, surveys |

The key conceptual split is "ATS vs HRIS." The ATS handles pre-hire (sourcing, interviews, offers); the HRIS handles post-hire (employee record, payroll, time off, performance). They are often separate products (e.g., Greenhouse + BambooHR), though some platforms — Workday, Rippling — try to handle both.

The biggest 2026 shift is that **AI has gone from a separate category to a default feature embedded across every layer.** "Eightfold = AI sourcing" was the 2022 framing. Today Greenhouse, Lever, Workday, and BambooHR all ship native AI matching, summarization, and chat. AI is no longer a discrete module; it is the baseline.

2. The ATS Landscape — Greenhouse, Lever, Workday Recruiting, iCIMS, SmartRecruiters, Ashby

The ATS owns the workflow from applied to hired. The top six products in 2026:

|---|---|---|---|

The selection criterion is company size and hiring volume. **Pre-Series A, one or two recruiters, fewer than 50 hires per year** points to Ashby or BambooHR Hiring. **Series B through public, 5-20 recruiters, 200-1000 hires per year** is overwhelmingly Greenhouse country. **Public, multi-country, 5000+ employees** is Workday Recruiting or iCIMS, with SmartRecruiters often winning when the international footprint is large.

Greenhouse, founded in 2012, was acquired by TPG in 2024 at a valuation around $750M. Its biggest differentiator is the **structured interview** — for each role, you predefine which competencies to test, what questions to ask, and how to score them on a 1-5 scale. This productizes Google's well-known finding that structured interviews predict on-the-job performance roughly twice as well as unstructured ones.

Lever, founded in 2012, was acquired by Employ Inc. (parent of Jobvite and Talemetry) in 2022. Whereas Greenhouse is ATS-first, Lever was designed from day one as **ATS + CRM**, so passive candidates who never applied live in the same pipeline as active applicants.

Ashby, founded in 2020, raised a $30M Series C in 2024. Its differentiator is **analytics**: where most ATSs require a separate BI tool for serious reporting, Ashby ships warehouse-grade analytics natively. From 2025, Ashby AI added candidate-job matching scores, interview note summarization, and auto-draft rejection emails as first-class features, and the product has been taking share quickly.

3. Greenhouse Deep Dive — Structured Interviews and the Integration Ecosystem

A typical Greenhouse workflow:

Example Greenhouse job stage configuration

job:

name: "Senior Backend Engineer"

stages:

- name: "Application Review"

type: "manual_review"

sla_hours: 48

- name: "Recruiter Screen"

type: "phone_screen"

duration_minutes: 30

interview_kit: "recruiter_screen_eng"

- name: "Technical Screen"

type: "interview"

duration_minutes: 60

interview_kit: "tech_screen_backend"

scorecard:

- attribute: "Coding Fundamentals"

scale: 1-5

- attribute: "System Design"

scale: 1-5

- attribute: "Communication"

scale: 1-5

- name: "Onsite Loop"

type: "interview_panel"

interviews:

- kit: "system_design_senior"

- kit: "behavioral_leadership"

- kit: "coding_pair_programming"

- kit: "hiring_manager_final"

- name: "Offer"

type: "offer"

approval_chain: ["hiring_manager", "recruiter", "vp_eng", "ceo"]

Each stage has an SLA and a scorecard, and interviewers must submit scores within 24 hours of the interview. Once enough scorecards accumulate, Greenhouse Reports' "Interviewer Calibration" view can tell you which interviewers' scores most correlate with on-the-job success.

Greenhouse's second advantage is its **integration ecosystem.** The official marketplace has 450-plus integrations: coding tests (HackerRank, CodeSignal), video interviews (HireVue, Spark Hire), background checks (Checkr, GoodHire), offer automation (DocuSign), HRIS (BambooHR, Workday, Rippling). Almost every adjacent tool is one click away.

A small Harvest API example:

Greenhouse Harvest API - list candidates and add a scorecard

API_KEY = os.environ["GREENHOUSE_API_KEY"]

BASE = "https://harvest.greenhouse.io/v1"

def list_candidates(job_id: int, status: str = "active"):

resp = requests.get(

f"{BASE}/candidates",

auth=(API_KEY, ""),

params={"job_id": job_id, "status": status, "per_page": 100},

)

resp.raise_for_status()

return resp.json()

def add_scorecard(application_id: int, scores: dict):

resp = requests.post(

f"{BASE}/applications/{application_id}/scorecards",

auth=(API_KEY, ""),

json={

"interview": "Technical Screen",

"attributes": [

{"name": "Coding", "rating": scores["coding"]},

{"name": "System Design", "rating": scores["sd"]},

"overall_recommendation": scores["overall"], # 1=strong_no .. 5=strong_yes

"interviewer_id": scores["interviewer_id"],

)

return resp.json()

The APIs split into Harvest (data sync), Onboarding, Job Board, and Partner Webhook. All support OAuth 2.0 and IP allowlists. A GraphQL beta launched in 2025 and now handles complex joined queries.

4. Lever and Ashby — How Modern ATSs Differentiate

Lever's biggest hook is the **Nurture Campaign.** A traditional ATS only manages applied candidates, but Lever puts passive candidates discovered on LinkedIn into the same pipeline and runs automated email sequences (Lever Nurture) that warm them up over six to twelve months. This makes Lever especially strong for long-cycle senior engineering and executive hires.

Ashby's differentiator is its **data model.** Most ATSs use the simple model of "candidate → application → stage → outcome." Ashby was designed from day one for multi-dimensional analytics, so any combination of channel, recruiter, interviewer, job family, and quarter slices cleanly out of the box.

-- Ashby analytics - per-channel funnel by stage

-- (Works inside Ashby Reports' built-in SQL builder)

SELECT

source_category AS channel,

stage_name,

COUNT(*) AS total,

COUNT(*) FILTER (WHERE outcome = 'advanced') AS advanced,

ROUND(100.0 * COUNT(*) FILTER (WHERE outcome = 'advanced') / COUNT(*), 1) AS pass_rate

FROM applications a

JOIN stages s ON a.current_stage_id = s.id

WHERE a.created_at >= NOW() - INTERVAL '90 days'

AND a.department = 'Engineering'

GROUP BY 1, 2

ORDER BY 1, 2;

Ashby's other strength is **AI by default.** Ashby AI, released in 2025, delivers (1) candidate-job match scores, (2) automatic interview-note summaries, (3) auto-drafted rejection emails, and (4) next-step recommendations. Greenhouse leans on marketplace apps for similar capabilities, whereas Ashby builds them straight into the UI.

5. Workday Recruiting, iCIMS, SmartRecruiters — The Enterprise Layer

Workday Recruiting is **a module of Workday HCM**, not a standalone product. That single fact drives most of its trade-offs. For global enterprises with 5000+ employees who want a single data model spanning ATS, HRIS, payroll, time, performance, and learning, Workday is effectively the only choice.

The upside is huge — at hire, a candidate is automatically populated as an employee record, and recruiting cost and quality data sit in the same BI as everything else. The downside is real: 6 to 18 months to implement, $50-150 per employee per year, and customization is gated behind Workday Studio, Workday's bespoke IDE.

iCIMS, founded in 1999 and currently owned by Bain Capital, is one of the oldest still-relevant ATSs. A large chunk of the US Fortune 500 (JPMorgan, AT&T, IBM, and similar) runs on it. Its strengths are 25 years of compounded best practices and best-in-class US compliance (EEOC, OFCCP reporting), but the UX feels heavy next to the modern alternatives.

SmartRecruiters specializes in multilingual, multi-region hiring. Customers like Microsoft, Bosch, and McDonald's, who recruit across 100+ countries simultaneously, are its sweet spot. EU and Asia labor-law support, a UI in 50+ languages, and auto-generated localized career sites are the core capabilities.

The selection question simplifies to **"which HCM are you already on?"** Workday HCM points to Workday Recruiting, SAP SuccessFactors points to SuccessFactors Recruiting, Oracle HCM points to Oracle Recruiting Cloud. With no HCM yet (or willingness to switch), Greenhouse + BambooHR/Rippling is the modern default.

6. Eightfold AI — The Ambition of a Career Intelligence Platform

Eightfold AI, founded in 2016 in Santa Clara, has been one of the most heavily backed companies in AI HR. A $220M Series E in 2021 took it to unicorn status, and follow-on funding in 2024 pushed the valuation to roughly $2B. CEO Ashutosh Garg, a former Google Search engineer, started with the thesis that search algorithms could be applied to careers.

The pitch is simple: **skills-based matching.** Instead of using surface attributes like degrees and titles, Eightfold builds an embedding vector representing what a candidate can actually do and matches it against role embeddings. As of 2026, Eightfold says its **Talent Intelligence Graph** indexes about 1 billion public profiles, 1M+ role definitions, and 600K+ skill taxonomies.

The four flagship modules:

| Module | Function |

|---|---|

| Talent Acquisition | External sourcing and matching |

| Talent Management | Internal mobility and promotion |

| Talent Flex | Freelancer and contractor pools |

| Talent Diversity | Bias detection and diversity reporting |

Inside Talent Acquisition, the headline feature is **Career Site Personalization.** When a candidate uploads a resume to a company career site, Eightfold instantly recommends the five to ten roles most aligned with their profile. Typical career-site conversion runs 2-3%; case studies show that lift to 8-15% once Eightfold is wired in.

Inside Talent Management, the headline feature is **internal mobility.** For employees stuck in the same role for 1-2 years, Eightfold surfaces three candidate internal moves based on their skills. Vodafone, BNP Paribas, and Conduent have published case studies showing they redeployed talent internally at roughly half the cost of equivalent external hiring.

Eightfold-style API calls (conceptual)

def get_match_score(candidate_id: str, job_id: str) -> dict:

resp = requests.get(

"https://api.eightfold.ai/v1/match",

headers={"Authorization": f"Bearer {API_KEY}"},

params={"candidate_id": candidate_id, "job_id": job_id},

)

Returns: { "score": 0.87, "matched_skills": [...], "gap_skills": [...] }

return resp.json()

def recommend_internal_roles(employee_id: str, top_k: int = 5) -> list[dict]:

resp = requests.get(

"https://api.eightfold.ai/v1/internal-mobility/recommend",

headers={"Authorization": f"Bearer {API_KEY}"},

params={"employee_id": employee_id, "top_k": top_k},

)

return resp.json()["recommendations"]

Eightfold's two main weaknesses are **price** and the **black-box concern.** Annual cost runs $50K-$150K for a 1000-employee company and climbs into the high six figures past 5000. And the system's ability to explain why a particular candidate was recommended still falls short of the "explainability" the EU AI Act expects. Eightfold added "Reason Codes" in 2025 to address this, but it remains a work in progress.

7. Phenom and Beamery — Candidate Experience and Talent CRM

Phenom, founded in 2010 in Ambler PA, essentially defined the **Talent Experience Management (TXM)** category. Where Eightfold focuses on the matching engine, Phenom focuses on the entire journey from "candidate lands on the career site" to "employee gets onboarded."

The four persona-based modules:

- **Phenom for Candidates**: career-site personalization, chatbot, video interview scheduling

- **Phenom for Recruiters**: sourcing, AI matching, campaign automation

- **Phenom for Employees**: internal mobility, learning, mentoring recommendations

- **Phenom for Managers**: team talent analytics, succession planning

In particular **Phenom Bot** (think Olivia for the Phenom stack) handles candidate inquiries on career sites 24/7 — questions like "Do you have a senior backend role open?" or "Can I reschedule my interview?" A large fraction of US enterprise career sites (General Motors, Mercedes-Benz, JPMorgan) run Phenom Bot under the hood.

Beamery, founded in 2014 in London, is the leader in **talent CRM.** ATSs manage applicants; CRMs manage non-applicants — passive senior engineers and executives you want to warm up over six to twenty-four months. That nurture cycle is the core skill.

Beamery's hook is its proprietary **Talent Graph**: a graph database of candidates, companies, skills, roles, conferences, and publications. That lets you write compound queries like "ML engineers who spent five years at Google and have a former colleague at Anthropic" — far beyond traditional ATS search.

The CRM category beyond Phenom and Beamery includes **Gem**, **Avature**, and **SmashFly** (acquired by Symphony Talent). Greenhouse and Lever also ship their own CRM modules.

8. hireEZ, Findem, Fetcher, SeekOut — AI Sourcing Tools

These tools automate candidate discovery itself. They crawl and index LinkedIn, GitHub, conference talks, papers, and patents to return the top 100 candidates for a role on demand.

| Tool | Specialty | Core feature |

|---|---|---|

| hireEZ (formerly Hiretual) | Engineering and technical roles | GitHub activity, open-source contributions |

| Findem | Diversity and complex criteria | "Attribute Search" across 50+ data sources |

| Fetcher | Automated cold outreach | AI-drafted + auto-sent email sequences |

| SeekOut | Diversity, veterans, healthcare | Indexed protected-class pools |

hireEZ specializes in **GitHub depth.** Beyond matching a GitHub username, it analyzes commit counts, languages used, repo star counts on contributed projects, and issue response times. A senior ML engineer search via LinkedIn gives you "worked at company X." The same search via hireEZ can return "merged 5+ PRs to PyTorch."

Findem leads with **Attribute Search.** Filters go well beyond keyword + company + level: think "3+ years at current company AND 2+ promotions AND member of a diversity group AND English/Spanish AND healthcare domain." Findem composites profiles from 80+ data sources behind the scenes.

hireEZ-style search API (conceptual)

def search_engineers(skills: list[str], min_years: int = 3, location: str = "Seoul"):

payload = {

"filters": {

"skills": {"any_of": skills},

"experience_years": {"min": min_years},

"location": {"city": location, "radius_km": 50},

"github": {"min_stars": 100, "languages": skills},

"size": 50,

}

resp = requests.post(

"https://api.hireez.com/v1/search",

headers={"Authorization": f"Bearer {TOKEN}"},

json=payload,

)

return resp.json()["candidates"]

The ethical question for all of these tools is **collection without consent.** LinkedIn explicitly forbids crawling outside its API; the hiQ Labs v. LinkedIn case made the law a moving target between 2017 and 2022. GDPR and the EU AI Act both require either legitimate interest or explicit consent for candidate data processing — and whether auto-indexing a public LinkedIn profile clears either bar is still actively debated.

9. HireVue and the Rise (and Retreat) of Video-Interview AI

HireVue, founded in 2004 in South Jordan UT, was the face of video-interview AI for a decade. It raised a $90M Series G in 2019, was acquired by The Carlyle Group in 2020, and consolidated the category by acquiring Modern Hire in 2022.

The original pitch was **scoring candidates from face, voice, and language signals.** Computer vision measured facial expression, eye contact, and smile frequency; speech analysis tracked tone, pace, and pause length; NLP scored answer content. The composite was an "Employability Score."

In 2019 EPIC filed an FTC complaint, and by 2021 academic research had documented racial and age bias in facial analysis. In January 2021, HireVue publicly **dropped facial analysis entirely.** Today only speech and answer content are analyzed, and scores feed interviewer tooling rather than automatic accept/reject decisions.

HireVue in 2026 looks like this:

- **Async video interviews**: candidates record on their own schedule; recruiters review later

- **AI answer summaries**: a 30-minute interview condensed to a 3-minute summary

- **Coding assessment hooks**: HackerRank and CodeSignal callouts

- **Game-based assessments**: Modern Hire's SHL-derived behavioral games

- **Fairness monitoring**: automatic per-group pass-rate reporting

Competing products include **myInterview** (UK), **Spark Hire** (US SMB), and **Willo** (founded 2020, growing fast globally). In Korea, **inAIR by Midas IT** powers many large-enterprise AI interviews end-to-end.

10. Paradox / Olivia — The Champion of Conversational AI

Paradox, founded in 2016 in Scottsdale AZ, owns the recruiting-chatbot category with its assistant **Olivia.** Customers skew toward global high-volume hiring brands like McDonald's, Lowe's, CVS Health, and Unilever.

The problem Olivia solves is sharp: **the funnel for hourly roles is huge and fast.** McDonald's may field dozens or hundreds of applications per store per week, and a single store manager cannot respond by hand. Olivia handles candidate questions, simple pre-screen (over 18? driver's license? available hours?), and interview scheduling over SMS and WhatsApp, 24/7.

Paradox Olivia conversational flow (sketch)

intent: "schedule_interview"

required_slots:

- candidate_id

- job_id

- location_id

flow:

- bot: "Hi! I am Olivia from McDonald's. I see you applied for a Crew Member role at our 5th Ave location. Are you still interested?"

- candidate: "Yes!"

- bot: "Great. Before we schedule, are you at least 16 years old?"

- candidate: "Yes, I am 19."

- bot: "Perfect. What is the best time to chat with our manager this week? I have Tue 2-4pm, Wed 10am-12pm, or Thu 5-7pm."

- candidate: "Wed at 10am works"

- bot: "Confirmed for Wed Mar 18 at 10am at 5th Ave. I will text you a reminder the morning of. Thank you!"

post_actions:

- calendar_event_created

- manager_notified

- ats_status_updated: "Interview Scheduled"

Paradox's impact is measurable. McDonald's case studies report average time-to-hire dropping from 21 days to less than 1 day, and candidate response rates rising from 30% to 80%. In high-volume hiring it is effectively the default.

Competitors include **Mya Systems** (founded 2012, acquired by StepStone in 2020), **AllyO** (founded 2015, acquired by HireVue in 2020), and **Phenom Bot.** In Korea, Kakao's recruiting chatbot and JobKorea's chatbot fill an analogous role.

11. Coding Assessments — HackerRank, CodeSignal, Codility

Technical hiring almost always includes a coding assessment. The big three products in 2026 are HackerRank, CodeSignal, and Codility.

| Product | Strength | Pricing |

|---|---|---|

| HackerRank | Biggest community (23M+), broad interview + assignment formats | $250-$1,000/mo |

| CodeSignal | Standardized General Coding Assessment (GCA) score | Quote |

| Codility | Strong in Europe, strict cheating detection | $300-$1,500/mo |

HackerRank, founded in 2009 in Mountain View, leans on its **interview pool.** Instead of authoring your own questions, you pick from 4000+ vetted problems and get automatic percentile normalization across candidates. HackerRank Interview also provides a pair-programming environment that ports the whiteboard interview cleanly to remote.

CodeSignal, founded in 2014, differentiates via its **GCA (General Coding Assessment)** score — a 70-minute, four-problem test scored on a 1-850 scale that companies like Capital One, Brex, and Robinhood use as an "SAT-equivalent" first filter, e.g., "GCA 600+ to advance."

Codility leads Europe, and its **plagiarism detection** is considered the strictest. Code-similarity analysis, tab-switch counts, paste frequency, and webcam face recognition combine into a fraud score. Since the ChatGPT and Claude wave in 2024 sent cheating concerns through the roof, every coding-assessment vendor has heavily invested in this surface.

HackerRank-style results retrieval

API_KEY = os.environ["HACKERRANK_API_KEY"]

BASE = "https://www.hackerrank.com/x/api/v3"

def get_test_result(test_id: int, candidate_email: str):

resp = requests.get(

f"{BASE}/tests/{test_id}/candidates",

headers={"Authorization": f"Bearer {API_KEY}"},

params={"email": candidate_email},

)

candidate = resp.json()["data"][0]

return {

"score": candidate["score"],

"max_score": candidate["max_score"],

"percentile": candidate["percentile"],

"plagiarism_score": candidate["plagiarism_score"],

"questions": [

{

"title": q["name"],

"score": q["score"],

"language": q["language"],

"tab_switches": q.get("tab_switch_count", 0),

}

for q in candidate["questions"]

}

Beyond coding, skills assessment is filled in by **TestGorilla** (multi-job aptitude + personality batteries), **Korn Ferry Sigma** (executive assessment), and **SHL/Aspiring Minds** (cognitive ability + personality).

12. HRIS — BambooHR, Workday HCM, Rippling, Gusto, Deel

After hiring, the employee record itself lives in the HRIS. The 2026 leaders, by size segment:

| Product | Fit | Differentiator |

|---|---|---|

| BambooHR | 50-500 employees SMB | Friendliest UX, includes Hiring module |

| Gusto | 5-100 US startups | Payroll + benefits + HR in one |

| Justworks | 5-200 US | PEO (co-employer) model |

| Rippling | 50-1000 multi-domain | IT + HR + Finance unified |

| Workday HCM | 1000+ global | Enterprise standard |

| ADP Workforce Now | 50-5000 US-centric | 50+ years, strong payroll |

| UKG Pro | 1000+ US | Ultimate Kronos Group |

| Oracle HCM Cloud | 1000+ global | Oracle stack integration |

| Deel | Global remote teams | EOR + Payroll |

| Remote.com | Global remote teams | EOR + Equity |

BambooHR, founded in 2008 in Lindon UT, is the SMB HRIS standard. The UX is friendliest in class. Self-service (PTO, profile edits), onboarding workflows, and a light performance review tool all live in the same product. For 50-500-person companies adopting their first real HRIS, it is the most common first pick.

Rippling, founded in 2016 in SF, is the fastest-growing in the segment, valued at $13.5B in 2024. Its trick is an **Identity Graph.** Hire someone, and their identity automatically propagates to Slack, Google Workspace, GitHub, 1Password, the laptop, VPN, payroll, and benefits. It was the first product to unify HR + IT + Finance into a single data model.

Workday HCM (founded 2005, IPO 2012) is the de facto standard for 1000+-employee globals. FY2024 revenue around $7.3B, market cap north of $60B. HCM, Payroll, Recruiting, Learning, Performance, and Adaptive Planning all run on the same platform.

13. EOR (Employer of Record) — Deel, Remote.com, Oyster, Multiplier

EORs have been the fastest-growing HR category post-2020. As "remote + global hiring" normalized, EORs let, say, a Korean company hire engineers in Japan, India, or Brazil without standing up a local entity — the EOR becomes the on-paper local employer and seconds the worker to the customer.

| EOR | HQ | Countries | Notes |

|---|---|---|---|

| Deel | SF | 150+ | Largest player, $12B valuation in 2024 |

| Remote.com | SF | 60+ | Operates its own entities (direct-hire model) |

Deel (founded 2019) combines EOR, contractor management, and global payroll. By 2024 it crossed $12B valuation and $1B annual revenue. Its UI is modern and the API is strong, which is why Korean tech companies like Toss and Hyperconnect routinely use it for cross-border hiring.

Deel-style global onboarding (conceptual)

def create_eor_employee(payload: dict):

resp = requests.post(

"https://api.letsdeel.com/rest/v2/eor",

headers={"Authorization": f"Bearer {DEEL_TOKEN}"},

json={

"first_name": payload["first_name"],

"last_name": payload["last_name"],

"email": payload["email"],

"country": payload["country"], # ISO 3166-1 alpha-2

"job_title": payload["job_title"],

"start_date": payload["start_date"],

"annual_gross_salary": payload["salary_usd"],

"currency": "USD",

"client_legal_entity": payload["entity_id"],

"benefits_package": "standard",

"equity": {

"type": "RSU",

"amount": payload.get("equity_units", 0),

)

return resp.json()

EOR fees run roughly $400-$700 per employee per month on top of the employee's salary. Standing up your own entity costs $30K-$100K in the first year and $20K-$50K annually after that, so EORs win economically until you have around ten employees in a country.

14. AI Risk and Regulation — EU AI Act, NYC Local Law 144, the Amazon Case

The biggest shadow over AI HR is **bias.** The 2018 cancellation of Amazon's five-year resume-screening AI project is the canonical cautionary tale. The model was trained on a decade of resumes that skewed male, so it learned to penalize the word "women" outright. Amazon eventually killed the project, but only after years of internal use.

Regulation has been catching up.

**The EU AI Act** (in force 2024, fully applicable in 2026) classifies hiring, firing, and promotion decisions as **high-risk** AI systems. High-risk systems must:

- Document data governance

- Provide human oversight mechanisms

- Meet technical standards for accuracy, robustness, and cybersecurity

- Notify users (employers) that AI is in use

- Give candidates a right to explanation

Penalties top out at 7% of global revenue or 35M euros, whichever is higher. As of May 2026 it is the strongest AI regulation in the world, and every global HR tech vendor has an EU AI Act program underway.

**NYC Local Law 144** (effective July 2023) mandates annual bias audits for any Automated Employment Decision Tool (AEDT) used in New York City. The audit must be conducted by an independent third party, and a summary — including pass-rate gaps by race and gender — must be posted publicly. Candidates can sue directly for non-compliance.

NYC Local Law 144 - 4/5 rule (EEOC) calculation

def four_fifths_rule(passing_rates: dict[str, float]) -> dict:

"""

EEOC Uniform Guidelines four-fifths rule:

if any protected group's pass rate is below 80 percent of the

highest-passing group's rate, adverse impact is presumed.

"""

max_rate = max(passing_rates.values())

threshold = max_rate * 0.8

return {

group: {

"rate": rate,

"impact_ratio": rate / max_rate,

"adverse_impact": rate < threshold,

}

for group, rate in passing_rates.items()

}

Example - per-group pass rates

result = four_fifths_rule({

"white_male": 0.50,

"white_female": 0.42,

"black_male": 0.35, # 0.35/0.50 = 0.70 < 0.80 - adverse impact presumed

"black_female": 0.32,

"asian": 0.48,

"hispanic": 0.41,

})

After NYC Local Law 144, similar duties have followed under **Illinois's AI Video Interview Act**, **Maryland HB 1202**, and **Colorado SB 21-169.** At the federal level the proposed ENFORCE the AI Act of 2025 would extend many of these obligations nationwide.

15. Performance, Learning, EX — Lattice, 15Five, Culture Amp, Cornerstone, Docebo

After ATS and HRIS comes the **post-hire journey.** The 2026 categories:

**Performance management + 360 feedback**:

- **Lattice** (founded 2015, valued at $3B in 2024): OKRs + reviews + 1:1s

- **15Five**: weekly check-ins + pulse surveys

- **Lattice competitors**: Leapsome, Engagedly, Culture Amp, Betterworks

**Employee experience and surveys**:

- **Culture Amp**: the standard for employee engagement surveys

- **Glint** (LinkedIn): pulse surveys

- **Officevibe**: strong with SMB

- **Peakon** (acquired by Workday, 2021): continuous surveys

**Learning / LMS**:

- **Cornerstone OnDemand**: enterprise LMS leader

- **Docebo**: AI-driven recommendation LMS, $1B+ market cap in 2024

- **360Learning**: collaborative learning

- **Workday Learning**: default for Workday customers

- **LinkedIn Learning**: tightly integrated content library

The shared 2026 theme across these categories is **AI coaching** plus **data unification.** Lattice released "Lattice AI" in 2025 — write a 1:1 note and you get coaching prompts back. Culture Amp's "Skills Coach" turns 360 feedback into specific action steps powered by an OpenAI backend.

16. The Korean HR Market — JobKorea, Saramin, Wanted, Rocketpunch, JAVIS, flex

The Korean HR tech market has followed an evolution distinct from the US. The main players:

| Category | Korean leaders |

|---|---|

| Job board (general) | JobKorea, Saramin |

| Job board (tech-focused) | Wanted, Jumpit (Saramin subsidiary), Rocketpunch |

| ATS | Greeting HR, Plapla, JOBDA (Midas IT) |

| HRIS | flex, Shiftee, JAVIS (Jarvis and Villains) |

| Performance / 360 | CLAP, NOVEL |

| AI interview | Midas IT inAIR, MIDAS Cognitive Test |

**JobKorea**, founded in 1996, is Korea's oldest general job board. Affinity Equity Partners acquired it for roughly 900B KRW in 2021. It covers entry-level through experienced hires across all job functions, with Albamon (part-time roles) as a subsidiary.

**Saramin**, founded in 2005 and listed on KOSDAQ, is JobKorea's main competitor. FY2024 revenue was around 150B KRW with operating margins in the 30s. Its subsidiary Jumpit is tech-recruiting focused.

**Wanted (Wanted Lab)**, founded in 2015 and KOSDAQ-listed in 2021, is widely seen as Korea's most modern HR tech company. Its core is **recommendation matching**: candidates upload resumes, companies proactively message them, and successful hires pay a referral bonus (typically 1-5M KRW) to whoever made the introduction. Cumulative matches surpassed 1M by 2026.

**flex**, founded in 2018, is the Korean HRIS leader. It exploits the gap that BambooHR cannot perfectly handle Korean labor law and year-end tax reconciliation. PTO, time, payroll, year-end tax, and hiring all live in one product. flex raised a Series D in 2024.

**Jarvis and Villains (Jabis)**, founded in 2017, sells an SMB HRIS that bundles accounting and tax with HR. Investors include KakaoVentures and KB Investment.

The Korea-specific wrinkles are **year-end tax reconciliation** and **the four major social insurances.** Global HRISs do not handle these natively, so Korean employers usually run either a Korea-native HRIS or pair a global HRIS with a Korean payroll outsourcer (ADP Korea, a labor law firm, etc.).

17. The Japanese HR Market — Rikunabi, Mynavi, Doda, HRMOS, TeamSpirit, jinjer

Japan's HR tech market is more conservative than Korea's but has been modernizing fast since 2020.

| Category | Japanese leaders |

|---|---|

| Shinsotsu (new grad) job board | Rikunabi, Mynavi |

| Mid-career recruiting | En Tenshoku, Doda, Bizreach, Green |

| ATS | HRMOS Recruiting (Bizreach), HERP, Talentio |

| HRIS | SmartHR, HRMOS Core, freee HR, Jobcan |

| Time and attendance | TeamSpirit, KING OF TIME, Jobcan Kintai |

| Performance | Ashita no Team, SmaRevi, HRBrain |

| Employee survey | wevox, Motivation Cloud |

| AI interview | HARUTAKA, Interview Maker |

**Recruit Holdings** is the giant of Japanese HR tech. It owns Rikunabi (new grad), Indeed (acquired in 2012, the world's largest job board), and Glassdoor (acquired in 2018). FY2024 revenue was roughly 3.5T yen, making it the world's largest HR tech company by revenue.

**Mynavi** is Rikunabi's main competitor. It splits the shinsotsu market roughly 50/50 with Rikunabi and runs vertical sites for mid-career, part-time, medical, and nursing.

**Bizreach**, founded in 2009, brought recommendation matching to Japan in a way similar to Wanted in Korea. Its subsidiary HRMOS combines ATS and HRIS in a single modern product, and **HRMOS Recruiting** is one of the fastest-growing Japanese ATSs.

**SmartHR**, founded in 2015, is Japan's HRIS leader. It fully automates Japanese-specific admin like onboarding paperwork, year-end tax adjustment, maternity/parental leave, and My Number registration. The 2024 valuation came in around 180B yen. SmartHR is to Japan what flex is to Korea.

**TeamSpirit**, founded in 2011 and TSE-listed, focuses on time and attendance. It is optimized for Japanese labor law specifics like the 36 agreement, overtime caps, and variable-hour systems.

**jinjer**, founded in 2016, sits alongside Jobcan as one of the two pillars of Japanese SMB HRIS, sold as modules across attendance, HR, payroll, and recruiting.

The Japan-specific wrinkle is the **shinsotsu ikkatsu saiyō (new-grad simultaneous hiring) culture.** Each year corporate recruiting opens in March of college junior year for an April-the-following-year start date, and the process runs over six months. Rikunabi and Mynavi are deeply optimized for that pattern, which makes the Japanese new-grad market structurally hard for foreign HR tech vendors to crack.

18. Compensation Transparency — Glassdoor, Levels.fyi, Blind, Comparably

The other major axis is **compensation transparency.** The main 2026 sites:

- **Glassdoor** (founded 2008, acquired by Recruit Holdings in 2018): the largest pool of company reviews, interview write-ups, and salary data

- **Levels.fyi** (founded 2017): the FAANG-grade reference for tech-company total comp packages (base + bonus + RSU) by level

- **Blind**: anonymous workplace community with per-company channels — usually the fastest leak channel for any internal company news

- **Comparably**: focused on culture and diversity scores

- **JobPlanet** (Korea): the Glassdoor of Korea

- **OpenWork** (Japan): the Glassdoor of Japan

The impact is two-sided. For candidates, information asymmetry has shrunk and negotiating leverage has gone up. For employers, comp packages are now competed in public. Pay-transparency laws that took effect in 2024 in California, New York, and Colorado — all requiring salary ranges on every posting — accelerated the shift.

A fast-growing trend is scraping and normalizing Levels.fyi data to use as an internal compensation benchmark, and companies like Carta, Pave, and Figures are productizing exactly that.

19. Backlash Against AI Interviewers — When Candidates Opt Out

Through 2024 and 2025 a small but visible movement against AI interviewers grew on LinkedIn and Reddit. Candidates increasingly state "I won't apply unless a human runs the first interview," and some companies (Stripe, Anthropic, etc.) explicitly tell candidates "we don't auto-reject with AI" on their career sites.

Common reasons candidates opt out:

- **Fairness concerns**: no way to know why the algorithm rejected you

- **Dignity**: 60 minutes alone in front of a camera feels degrading

- **Preparation gap**: you can't prepare when you don't know what is being scored

- **Technical bias**: penalties for non-native pronunciation or webcam glitches

In response, since 2025, AI interview vendors emphasize a **"Hybrid Mode"** — AI is used only for pre-screen, humans make the final call, and candidates can opt out of AI scoring in favor of a human-only interview. HireVue, Modern Hire, and Paradox all now expose this option.

20. End-to-End Workflow — A Modern 8-Product Recruiting Stack

A representative 2026 stack for a Series B global SaaS company (100-300 employees, 3-5 recruiters, 100-200 hires per year):

| Stage | Tool |

|---|---|

| Job posting | LinkedIn + Indeed + Wanted (Korea) + Wellfound |

| Sourcing | hireEZ (engineering), Gem (general) |

| ATS | Greenhouse or Ashby |

| Coding assessment | HackerRank or CodeSignal |

| Video interview | HireVue (optional) |

| Chatbot | Paradox (high-volume only) |

| Background check | Checkr |

| Offer + signing | DocuSign |

| HRIS | Rippling or BambooHR |

| Global EOR | Deel |

| Performance / 360 | Lattice |

| Learning / LMS | Docebo or LinkedIn Learning |

| Employee survey | Culture Amp |

Data flows roughly like this:

[Job Board] --> [Greenhouse ATS] --> [DocuSign Offer] --> [Rippling HRIS] --> [Deel Payroll]

^ |

| v

[hireEZ Sourcing] [Lattice / Culture Amp]

The integrations between stages typically come from one of three sources:

1. **Official marketplace apps**: 90% of cases — one-click OAuth + automatic sync.

2. **Zapier / Workato / Tray.io**: used when there is no first-party integration, or when a little custom logic is needed.

3. **Custom (direct API)**: high-security or high-volume cases.

The single biggest trap when assembling a stack is **running two products in the same category at once.** Greenhouse + Lever splits the candidate database and breaks SLA tracking. Pick one product per layer and treat its data model as the single source of truth.

21. Recruiting Analytics — Four Funnel Metrics

To run recruiting like a data team, track these four metrics:

| Metric | Definition | 2026 industry average |

|---|---|---|

| Time-to-fill | Days from posting to accepted offer | 35-50 days for tech roles |

| Cost-per-hire | Average cost per hire (sourcing + agency + job-site spend) | $4,000-$15,000 |

| Quality-of-hire | Performance score 1 year after hire | Hard to measure; company-defined |

| Source effectiveness | Hire conversion by channel | Direct apply 35%, referrals 25% |

**Source effectiveness** is the basis for recruiting budget allocation. Modern ATSs like Greenhouse and Ashby slice this automatically — "100 candidates from LinkedIn ads, 5 interviews, 1 hire, $25K cost-per-hire" is one click.

-- Recruiting funnel - by channel and quarter

WITH funnel AS (

SELECT

DATE_TRUNC('quarter', a.applied_at) AS quarter,

a.source_category,

COUNT(*) AS applied,

COUNT(*) FILTER (WHERE a.reached_stage >= 'phone_screen') AS screened,

COUNT(*) FILTER (WHERE a.reached_stage >= 'onsite') AS onsite,

COUNT(*) FILTER (WHERE a.outcome = 'hired') AS hired

FROM applications a

WHERE a.applied_at >= NOW() - INTERVAL '12 months'

GROUP BY 1, 2

)

SELECT

quarter,

source_category,

applied,

ROUND(100.0 * screened / NULLIF(applied, 0), 1) AS apply_to_screen_pct,

ROUND(100.0 * onsite / NULLIF(screened, 0), 1) AS screen_to_onsite_pct,

ROUND(100.0 * hired / NULLIF(onsite, 0), 1) AS onsite_to_hire_pct,

hired

FROM funnel

ORDER BY quarter DESC, hired DESC;

Twelve months of funnel data usually surface a familiar pattern: companies spend the most on LinkedIn ads but the most actual hires come from employee referrals.

22. The Limits of Recruiting Automation — What Must Stay Human

Some parts of HR should not — or cannot — be automated.

1. **Final hire decision**: as the EU AI Act spells out, hiring and firing decisions must be made by humans. AI is a sort and filter helper, not the decision-maker.

2. **Culture fit**: hard to quantify, and "culture fit" too often becomes an excuse to undermine diversity. Leaving this deliberately un-automated is safer.

3. **Offer negotiation**: it depends on personal context, comp expectations, and competing offers, and rapport matters more than scoring.

4. **Reference checks**: informal conversations with the candidate's references cannot be automated.

5. **First day of onboarding**: meeting peers and the manager on day one is not replaceable by software.

The golden rule of recruiting automation is **"AI discovers, humans decide."** The fuzzier that line gets, the more bias risk and candidate backlash you accumulate at the same time.

23. Looking Ahead — HR Tech After 2027

A few visible threads as of May 2026:

- **Skills-based organizations**: redesigning teams around skills rather than job titles. Eightfold and Workday Skills Cloud lead.

- **AI agents inside HRIS**: BambooHR, Rippling, and Workday all ship AI assistants today. "Submit my time off," "Download my W-2" — natural-language requests.

- **Talent marketplaces internalized**: internal mobility happens before external hiring becomes the standard motion.

- **Continuous performance**: yearly reviews give way to weekly check-ins and live feedback. Lattice and 15Five standardize this.

- **HR + Finance unified**: Rippling's IT + HR + Finance vision spreads. Hire → onboard → IT auto-provision → payroll runs in a single flow.

- **Regulation tightens**: full enforcement of the EU AI Act, more US state AEDT laws, and Korea and Japan beginning to draft guidance.

The bigger trend is that **AI stops being a separate HR tech category and becomes a default capability across every category.** "AI recruiting tool" was a differentiator in 2023; by 2026 every ATS ships AI; by 2028 most users will not consciously notice it.

24. References

- Sapient Insights HR Systems Survey 2025-2026 — `https://www.sapientinsights.com/hr-systems-survey/`

- Josh Bersin HR Technology Report 2026 — `https://joshbersin.com/`

- Greenhouse Software developer documentation — `https://developers.greenhouse.io/`

- Lever Developer Center — `https://hire.lever.co/developer`

- Ashby API documentation — `https://developers.ashbyhq.com/`

- Workday Recruiting overview — `https://www.workday.com/en-us/products/talent-management/recruiting.html`

- iCIMS overview — `https://www.icims.com/`

- SmartRecruiters developer docs — `https://developers.smartrecruiters.com/`

- BambooHR API — `https://documentation.bamboohr.com/`

- Eightfold AI — `https://eightfold.ai/`

- Phenom — `https://www.phenom.com/`

- Beamery — `https://beamery.com/`

- hireEZ — `https://hireez.com/`

- Findem — `https://www.findem.ai/`

- SeekOut — `https://seekout.com/`

- HireVue — `https://www.hirevue.com/`

- Paradox — `https://www.paradox.ai/`

- HackerRank for Work — `https://www.hackerrank.com/work`

- CodeSignal — `https://codesignal.com/`

- Codility — `https://www.codility.com/`

- TestGorilla — `https://www.testgorilla.com/`

- Workday HCM — `https://www.workday.com/`

- ADP Workforce Now — `https://www.adp.com/`

- UKG (Ultimate Kronos Group) — `https://www.ukg.com/`

- Rippling — `https://www.rippling.com/`

- Gusto — `https://gusto.com/`

- Deel — `https://www.deel.com/`

- Remote.com — `https://remote.com/`

- Oyster — `https://www.oysterhr.com/`

- Lattice — `https://lattice.com/`

- Culture Amp — `https://www.cultureamp.com/`

- Cornerstone OnDemand — `https://www.cornerstoneondemand.com/`

- Docebo — `https://www.docebo.com/`

- EU AI Act overview — `https://artificialintelligenceact.eu/`

- NYC Local Law 144 (AEDT) — `https://www.nyc.gov/site/dca/about/automated-employment-decision-tools.page`

- EEOC AI guidance — `https://www.eeoc.gov/ai`

- Amazon scrapped recruiting AI (Reuters, 2018) — `https://www.reuters.com/article/us-amazon-com-jobs-automation-insight-idUSKCN1MK08G`

- JobKorea — `https://www.jobkorea.co.kr/`

- Saramin — `https://www.saramin.co.kr/`

- Wanted — `https://www.wanted.co.kr/`

- flex — `https://flex.team/`

- Midas IT inAIR — `https://www.midasitc.com/`

- Recruit Holdings — `https://recruit-holdings.com/`

- Mynavi — `https://www.mynavi.jp/`

- Bizreach / HRMOS — `https://hrmos.co/`

- SmartHR — `https://smarthr.jp/`

- TeamSpirit — `https://www.teamspirit.com/`

- Glassdoor — `https://www.glassdoor.com/`

- Levels.fyi — `https://www.levels.fyi/`

- Blind — `https://www.teamblind.com/`