Where the defaults come from
| Input | Default | Source |
|---|---|---|
| Senior base, US coastal | $210,000 | Levels.fyi senior SWE median base: SF $225k, NYC $200k; ML/AI SF median $200k (accessed 12 Sep 2026) |
| Senior base, US non-coastal | $175,000 | Levels.fyi: Austin $180k, Chicago $170k senior base |
| Senior base, UK | £100,000 | Levels.fyi senior SWE £101.4k; ML/AI base P75 £155k |
| Senior base, Germany | €92,000 | Levels.fyi senior SWE €91.2k; Ravio Berlin AI Eng P3 €81.2k (May 2026) |
| Senior base, France | €72,000 | Levels.fyi senior SWE €68.3k; Paris ML/AI median €66k, P75 €94k |
| Senior base, Netherlands | €98,000 | Levels.fyi senior SWE €100.1k; Amsterdam ML/AI €99k |
| AI premium | +10% | Ravio, HRIS data 400k+ employees: +12% at P3 (Nov 2025); Berlin May 2026 shows +0.7% — the premium is collapsing |
| Employer multiplier | US 1.30 · UK 1.20 · DE 1.20 · FR 1.43 · NL 1.25 | BLS ECEC June 2026 (benefits 30.8% of comp, professional occupations); Eurostat non-wage share 2025: DE 23.5%, FR 32.3%, NL 24.7%; UK employer NI 15% + pension; Osborne Clarke 2026 German rates; fiche-paie.fr 11 Sep 2026 |
| Recruiting | 20% of base | Contingency 15–25% (EngineeringHiringCost, Apr 2026); UK AI/ML and cyber 22–25% (ON3, Aug 2026); SHRM direct cost-per-hire floor $5,475 (Oct 2025) |
| Time to hire | 75 days | Ashby Talent Trends, 54M applications, technical roles median 75 days (May 2026) — used to justify the vacancy cost inside recruiting |
| Attrition | 25% / year | Pave: AI/ML individual contributors 28% vs 17% for software engineers (Feb 2025) |
| Management overhead | 15% | Pave: 4.9 direct reports per manager (Mar 2026); 1 ÷ 4.9 of a manager's loaded cost ≈ 15% of the engineer's |
| Share of a person | 0.5 FTE | IAPP 2025 (n = 671): only 10% of organisations hire dedicated AI governance roles; Future AGI: self-built eval stack "needs half to one engineer permanently" (June 2026) |
| Setup | 240 hours | Between a 13-hour minimal suite (practitioner, July 2026) and "3–6 months with 2–4 engineers" (vendor, June 2026) |
| Tooling | $900 / month | Braintrust Pro $249; Langfuse Pro $199 + Teams $300; LangSmith Plus $39/seat; Datadog LLM Observability from $160; Confident AI Starter $200; plus ~$400 judge inference — list prices 12 Sep 2026 |
| Guardrails platform | $20,000 / year | Unverified midpoint. Lakera, Promptfoo Enterprise, Confident AI Team ($24k/yr) are quote-only above free tiers. Set to 0 if you use open source. |
Why independence is worth paying for at all
An internal function can be excellent and still fail the test that matters: a customer, insurer or auditor does not accept a team's assessment of its own work. The clearest statements of that principle, quoted:
- NIST AI RMF, MEASURE 1.3: "Internal experts who did not serve as front-line developers for the system and/or independent assessors are involved in regular assessments and updates." The Playbook adds: "Independent auditors may bring important independent perspectives for reviewing evidence of AI system safety."
- NYC Local Law 144, 6 RCNY § 5-300: an auditor is not independent if they "is or was involved in using, developing, or distributing" the system, or have an employment relationship with the deployer or the vendor.
- ISO/IEC 42001, clause 9.2.2: auditors must be selected "in a way that ensures objectivity and impartiality" (paraphrased; the standard is paywalled).
We deliberately do not cite the EU AI Act here. For most Annex III systems, Article 43(2) prescribes conformity assessment "based on internal control … which does not provide for the involvement of a notified body." The case for independent evidence rests on what buyers and auditors expect, not on a legal mandate — and we would rather say so than let you find out from your lawyer.
What this calculator does not do. It does not value the things an internal engineer does beyond evaluation — product work, incident handling, institutional memory. It does not model the cost of a bad hire, or of the six-to-twelve-month gap while you search. And it assumes a retainer covers the evaluation need; for a team shipping new AI features monthly, it may not. Use it to frame the conversation, not to end it.
- Levels.fyi salary pages (SF, NYC, Austin, Chicago, London, Berlin, Paris, Amsterdam), accessed 12 September 2026. Ravio, "Software engineer salary trends" (Oct 2025), "AI compensation and talent trends" (Nov 2025), "Berlin salary trends 2026" (May 2026).
- US Bureau of Labor Statistics, Employer Costs for Employee Compensation, June 2026 (released 9 Sep 2026). Eurostat, labour cost levels 2025 (31 Mar 2026). Osborne Clarke, Social security contributions in Germany 2026. fiche-paie.fr, charges sociales employeur (11 Sep 2026). EmployerCalculator UK, 2026/27.
- Ashby, Talent Trends: recruiting operations benchmarks (7 May 2026). Pave, AI/ML talent insights (11 Feb 2025); span of control benchmark (3 Mar 2026). SHRM Benchmarking 2025 via InterviewCost.
- IAPP, AI Governance Profession Report 2025 (16 Apr 2025). Future AGI, "When is building your LLM evals actually worth it" (8 Jun 2026). Hamel Husain, Evals FAQ (updated 1 Sep 2026).
- Vendor pricing pages: LangSmith, Langfuse, Braintrust, Arize, Patronus, Confident AI, Promptfoo, Datadog, accessed 12 September 2026.
- NIST AI RMF Playbook, MEASURE 1.3 and 2.6. NYC Rules 6 RCNY § 5-300. Atoro, ISO 42001 internal audit requirements (28 Aug 2026). EU AI Act Article 43.