Home

/

Build It in a Weekend. Run It for Years.

/

Sources & Further Reading

Sources & Further Reading

Appendix D
Appendix
5
min read

Every statistic cited across this guide to building AI recruitment automation is anchored to something you can check. This is that audit trail: grouped by theme, with the source, the year, and a link.

A note on how to read it. We mark each entry as one of two kinds:

  • [PRIMARY] — a study, a regulator, a government statistics office, an official price page, or a peer-reviewed paper. Trust these as numbers.
  • [MARKET RANGE] — vendor or industry figures that vary by source. We never lean on one of these as a single authoritative number; we present them as a range and tell you the spread.

A handful of widely-quoted recruitment-industry figures still need primary verification before a hard print run. We've flagged those [VERIFY] rather than dress them up as settled fact. That's the honest version.

If a claim isn't traceable below, treat it as opinion, not evidence.

The time problem (admin, screening, the cost of the status quo)

  • 257 applications per role in 2025 (up from 207); ~80% rejected at first screen — Employ, Hiring Benchmarks Report, 2025 (257.6 in 2025, up from 207.2 in 2024) — https://www.hrdive.com/news/hiring-benchmarks-report-employ-2025-more-applicants/809604/ [PRIMARY] (~80% first-screen rejection remains an industry estimate — [VERIFY])
  • ~12 hours/week lost to admin per recruiter (~20 for top firms); ~4 hours/week on CV formatting; 10–45 minutes to reformat one CV by hand — industry and YS estimates, framed conservatively. [VERIFY]
  • 7.4 seconds — average human time spent per resume — Ladders eye-tracking study, 2018 — https://www.theladders.com/career-advice/you-only-get-6-seconds-of-fame-make-it-count [PRIMARY]
  • Average time-to-fill ~42 days — SHRM, 2025 Recruiting Benchmarking Report — https://www.shrm.org/content/dam/en/shrm/research/2025-recruiting-benchmarking-report.pdf [PRIMARY] (SHRM reports this as time-to-fill; some sources loosely call it time-to-hire)
  • 61% of staffing firms using AI in 2025 (up from 48%) — Bullhorn GRID 2025 Talent Trends report — https://www.bullhorn.com/grid/grid-2025-talent-trends-report/ [PRIMARY]
  • 45 CVs screened in ~52 seconds → 20 shortlisted / 15 rejected / 10 flagged — YS demo proof point. [PRIMARY — YS]

Further reading (YS): "AI Resume Screening: 45 CVs in Under a Minute" — https://you-source.com/blogs/ai-resume-screening-recruiters

Why DIY and off-the-shelf fail (the part that earns the book's thesis)

  • ~95% of enterprise GenAI pilots deliver no measurable P&L impact; ~67% success for partner/vendor-built vs ~33% for internal DIY — MIT Project NANDA, The GenAI Divide: State of AI in Business 2025 (released July 2025) — https://mlq.ai/media/quarterly_decks/v0.1_State_of_AI_in_Business_2025_Report.pdf [PRIMARY]
  • >40% of agentic AI projects cancelled by end of 2027 — Gartner, 2025 — https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027 [PRIMARY]
  • "Agent washing" and ~130 genuinely agentic vendors out of thousands of claims — Gartner, 2025. [PRIMARY]
  • ≥50% of GenAI projects over budget — Gartner, 2025, reported via — https://itdaily.com/news/cloud/half-of-ai-projects-over-budget/ [PRIMARY — Gartner]
  • ≥30% of GenAI abandoned after proof-of-concept by end of 2025 — Gartner — https://www.gartner.com/en/articles/genai-project-failure [PRIMARY]
  • 42% of companies scrapped most AI initiatives in 2025 (up from 17% in 2024) — S&P Global Market Intelligence, Voice of the Enterprise: AI & Machine Learning, Use Cases 2025 — https://www.ciodive.com/news/AI-project-fail-data-SPGlobal/742590/ [PRIMARY — S&P Global]
  • Ongoing maintenance dominates the lifetime cost of ML systems — Sculley et al., Hidden Technical Debt in Machine Learning Systems, NeurIPS 2015 — https://papers.nips.cc/paper/5656-hidden-technical-debt-in-machine-learning-systems [PRIMARY — peer-reviewed]

Further reading (YS): "Why Off-the-Shelf AI Recruitment Tools Break" — https://you-source.com/blogs/managed-ai-automation-recruitment · "Your AI Coworker Didn't Fail. The Rollout Did." — https://you-source.com/blogs/agentic-ai-coworker-rollout

Reliability & security

  • Error compounding: 95% per-step accuracy → ~60% over 10 steps, ~36% over 20 steps — arithmetic illustration (0.95ⁿ). Presented as maths, not as a survey. [PRIMARY — calculation]
  • ~15% (14.78%) of library/API changes break backwards compatibility — Xavier et al., Historical and Impact Analysis of API Breaking Changes: A Large-Scale Study (317 Java libraries, 9K releases), SANER 2017 — https://homepages.dcc.ufmg.br/~mtov/pub/2017-saner-breaking-apis.pdf [PRIMARY — peer-reviewed]
  • ~23.8M secrets leaked on public GitHub in 2024 (up ~25% YoY) — GitGuardian, The State of Secrets Sprawl 2025 — https://blog.gitguardian.com/the-state-of-secrets-sprawl-2025/ [PRIMARY]
  • 73% of security professionals admit to using unapproved (shadow) SaaS — Next DLP survey of 250+ security pros at RSAC and Infosecurity Europe 2024 — https://www.helpnetsecurity.com/2024/07/10/shadow-saas-security-risks/ [PRIMARY — survey]
  • 41% of US job seekers attempted hidden-text "prompt injection" to game screeners — job-seeker survey, 2025 — https://blog.theinterviewguys.com/job-seekers-are-hiding-secret-text-in-their-resumes/ [MARKET RANGE — single survey]
  • Toptal supply-chain breach (July 2025): 73 repos exposed, 10 malicious npm packages published — The Hacker News / BleepingComputer, 2025 — https://thehackernews.com/2025/07/hackers-breach-toptal-github-publish-10.html [PRIMARY — security reporting]

Reference: the Semantic Kernel function-calling and guardrail-gateway patterns shown across this guide are illustrative excerpts built on Microsoft.SemanticKernel — https://learn.microsoft.com/semantic-kernel/ [PRIMARY — vendor docs]

the-math-no-recruiter-can-win-by-hand
what-an-ai-agent-actually-is
the-leash
the-toolkit
the-model-small-capable-swappable
talking-to-your-ats
use-case-1-resume-screening-against-a-job
the-shape-of-the-loop
running-it-thought-action-observation
use-case-2-cv-formatting-redacting-for-clients
reformatting-into-your-branded-template
resume-shortlisting
that-was-easy
security-compliance
keeping-pii-out-of-the-llm
exceptions-reliability
silent-api-drift-the-ats-changes-under-you
when-it-fails-anyway-dead-letter-and-the-leash
monitoring-observability
maintenance-the-lifecycle
the-scorecard-success-metrics-kpis
build-vs-buy-vs-managed
what-an-engineer-actually-costs
what-the-wider-data-says-happens-next
conclusion-how-this-gets-run-for-you
the-promises-behind-the-service
fuller-code-listings
one-full-screening-react-loop-semantic-kernel
env-deployment-reference
secrets-in-dev-vs-production
bullhorn-jobadder-endpoint-cheat-sheets
sources-further-reading
compliance-primary-law-sources

Download the full PDF for free?

Download full PDF
build-it-in-a-weekend.pdf
Oops! Something went wrong while submitting the form.
Related Chapters