Compliance (primary law sources)
- GDPR fines up to 4% of annual global turnover, or about $22 million, whichever is higher — GDPR Article 83. [PRIMARY]
- EU AI Act — recruitment classified as a high-risk system — Annex III. [PRIMARY]
- NYC Local Law 144 — bias audits required for automated employment decision tools — NYC Department of Consumer and Worker Protection. [PRIMARY]
- Amazon scrapped its internal recruiting tool after it "taught itself to penalize resumes that contained women-associated words" — Reuters, 2018. [PRIMARY]
Three of these are law, not advice. The fourth is the cautionary tale that explains why visible reasoning and a human on the leash are non-negotiable.
Build-vs-buy cost
Managed-automation market range [MARKET RANGE — present as a range, never as our price]
- $2,000–$5,000/month managed retainer tier — Digital Agency Network, 2026 — https://digitalagencynetwork.com/ai-agency-pricing/
- $500–$5,000/month done-for-you — Arsum, 2025 — https://arsum.com/blog/posts/ai-automation-agency-pricing/
- $500–$5,000/month for SMBs — Latenode, 2025 — https://latenode.com/blog/industry-use-cases-solutions/enterprise-automation/17-top-ai-automation-agencies-in-2025-complete-service-comparison-pricing-guide ; SalemWise, 2025 — https://www.salemwise.com/insights/how-much-does-ai-automation-really-cost-for-smbs-and-how-to-budget-for-it-without-wasting-money
⇒ The ~$2,500/month figure used in the build-vs-buy maths is the midpoint of this prevailing SMB range. It is a market assumption the reader can adopt — not a quote.
DIY labour — the part that actually bites
- Benefits = 29.9% of total compensation (≈1.43× wage multiplier) — US Bureau of Labor Statistics, Employer Costs for Employee Compensation — https://www.bls.gov/news.release/ecec.nr0.htm [PRIMARY]
- US software-engineer salary — Glassdoor — https://www.glassdoor.com/Salaries/software-engineer-salary-SRCH_KO0,17.htm [MARKET RANGE]
- UK software-engineer salary — Levels.fyi — https://www.levels.fyi/t/software-engineer/locations/united-kingdom [MARKET RANGE]
- ⇒ Fully-loaded mid-level engineer ≈ $10,000–$15,000/month (US) — our calculation; inputs cited above. ≈ 4–6× a $2,500 managed retainer.
- Fractional CTO ≈ $3,000–$15,000/month — TLVTech — https://www.tlvtech.io/post/understanding-fractional-cto-rates-a-guide-for-entrepreneurs-and-business-leaders [MARKET RANGE]
DIY infrastructure & tooling — the cheap parts [PRIMARY — official list prices]
- **Cloud Run: free tier covers 2M requests/month, scales to zero (an architecture property, not the production cost)** — https://cloud.google.com/run/pricing ; https://cloudchipr.com/blog/cloud-run-pricing ; Cloud SQL — https://cloud.google.com/sql/pricing ⇒ a production deployment is not the near-zero hobby case. With a warm instance (min-instances ≥ 1 to kill cold starts), a managed audit store (Cloud SQL), queueing (Pub/Sub) and log ingestion, reckon ~$120/month at the low end, ~$300–$350 typical for a mid-size agency, and $500–$750+ under heavier load / HA (our calculation; GCP list rates). Scale-to-zero only reaches near-$0 with negligible traffic.
- Grafana Cloud: $0 free / $19+ Pro — https://grafana.com/pricing/ ; Datadog cost comparison — https://www.vantage.sh/blog/datadog-vs-grafana-cost
- PagerDuty: $21–$41/user/month — https://www.pagerduty.com/pricing/incident-management/ ; on-call pay $500–$1,200/engineer/month [MARKET RANGE] — https://rootly.com/on-call-software/pay
- LLM API list prices — OpenAI — https://openai.com/api/pricing/ (GPT-5 ≈ $1.25 in / $10 out per 1M tokens; GPT-4o-mini $0.15 in / $0.60 out per 1M) ⇒ on a GPT-5-class model running an agentic screening loop (~2–4 model calls per CV, ~3,000 input + ~700 output tokens each) reckon ~$20–$75 per 1,000 CVs — a few cents per CV. A budget model (GPT-4o-mini) is ~10–15× cheaper (~$2–$4 per 1,000). At agency volume (~10,000 CVs/month) that's a few hundred dollars/month on a frontier model (~$250–$700) vs ~$25/month on a mini model. (Our calculation; OpenAI list prices, batch mode ~halves it.)
- Failure/cost anchors: Gartner (>40% cancelled by 2027; ≥50% over budget, see why DIY and off-the-shelf fail) and Sculley et al., NeurIPS 2015 (ongoing ML maintenance cost).
Further reading (YS): "Subscription engineering vs hiring vs marketplace" — https://you-source.com/blogs/subscription-engineering-hire-or-marketplace
YS Managed AI Automation (the offering)
- "Designed, deployed and run by us," 98.4% renewal rate, "live in 30 days," 24/7 monitoring, dedicated Slack channel, sub-2-hour response; SOC 2 Type II ready, TLS 1.3 in transit / AES-256 at rest, data never used for model training — YS — https://you-source.com/ai-automation ; https://you-source.com/recruitment [PRIMARY — YS]
Further reading (YS): "How to Implement AI in a Recruitment Agency" — https://you-source.com/blogs/implement-ai-recruitment-agency · "What Agentic AI Actually Is" — https://you-source.com/blogs/what-is-agentic-ai
A word on the numbers you'll quote next
Use the [PRIMARY] figures with confidence — they're regulators, government statisticians, peer review and official price pages. Treat the [MARKET RANGE] ones as ranges, because that's what they are; the moment you collapse a $500–$5,000 spread into a single confident figure, you've stopped doing maths and started doing marketing. And the [VERIFY] flags are there on purpose: better an honest asterisk than a clean lie.
The technology is the easy part. So is quoting a statistic. Standing behind it for years is the work — which is the whole book, in one line.