{
  "source": "RevenueLab",
  "source_url": "https://www.revenuelab.fyi",
  "dataset": "ai_economics_benchmarks",
  "updated": "2026-07-29",
  "license": "CC BY 4.0 with attribution to RevenueLab.fyi",
  "citation": "RevenueLab — AI economics benchmarks (https://www.revenuelab.fyi/api/public/ai-benchmarks.json)",
  "notes": "Planning benchmarks compiled from public vendor disclosures, published pricing pages, and RevenueLab's calculator model assumptions. These are ranges for modelling, not measurements of any single company. Always validate against your own instrumentation before making a pricing or headcount decision.",
  "benchmarks": [
    {
      "metric": "AI-native gross margin",
      "typical": "50-70%",
      "range": "35-80%",
      "note": "Versus 75-85% for classic SaaS. Model COGS is per-action, not near-zero."
    },
    {
      "metric": "Enterprise AI licence adoption",
      "typical": "30-60%",
      "range": "10-85%",
      "note": "Share of purchased seats used weekly. The single biggest driver of realised ROI."
    },
    {
      "metric": "Hours saved per active user per week",
      "typical": "1-4",
      "range": "0.5-8",
      "note": "Knowledge work, before subtracting verification time. Self-reported figures run 2-3x higher than measured."
    },
    {
      "metric": "Output requiring human verification",
      "typical": "20-40%",
      "range": "10-60%",
      "note": "Rework tax on AI output. Rises with task complexity and regulatory exposure."
    },
    {
      "metric": "Time-saving realisation rate",
      "typical": "50-70%",
      "range": "20-100%",
      "note": "Share of saved hours that convert into extra output or avoided hires rather than slack."
    },
    {
      "metric": "Workflow automation coverage",
      "typical": "60-85%",
      "range": "30-95%",
      "note": "Share of task volume an AI system fully handles. 100% coverage is almost never achieved."
    },
    {
      "metric": "Automation maintenance cost",
      "typical": "15-25% of build cost / yr",
      "range": "10-40%",
      "note": "Prompt drift, model updates, upstream format changes, accumulating edge cases."
    },
    {
      "metric": "Heavy-user usage multiple",
      "typical": "5-10x median",
      "range": "3-20x",
      "note": "Top decile of users. Drives 40-60% of total consumption."
    },
    {
      "metric": "Credit pack redemption rate",
      "typical": "60-85%",
      "range": "40-95%",
      "note": "Unredeemed credits are breakage. Enterprise prepay redeems higher than consumer packs."
    },
    {
      "metric": "Credit markup over raw cost",
      "typical": "2-5x",
      "range": "1.5-10x",
      "note": "Must cover payment fees, support, infrastructure and the heavy tail."
    },
    {
      "metric": "Included allowance in hybrid pricing",
      "typical": "70-90% of median usage",
      "range": "50-100%",
      "note": "Set so 80-90% of seats never see an overage line."
    },
    {
      "metric": "Overage price vs list rate",
      "typical": "20-40% below list",
      "range": "0-60% below",
      "note": "Overage above list reads as a penalty and drives renegotiation."
    },
    {
      "metric": "Good AI tool payback period",
      "typical": "3-6 months",
      "range": "1-18 months",
      "note": "Beyond 12 months, tooling churn usually outruns the business case."
    },
    {
      "metric": "RAG retrieval share of request cost",
      "typical": "10-25%",
      "range": "5-50%",
      "note": "Embedding plus vector search. Generation dominates unless context is very large."
    },
    {
      "metric": "Prompt caching saving",
      "typical": "40-70%",
      "range": "20-85%",
      "note": "For workloads with a stable system prompt and repeated context."
    },
    {
      "metric": "Self-host break-even GPU utilisation",
      "typical": "~40-60%",
      "range": "30-80%",
      "note": "Below this, hosted API pricing is generally cheaper all-in including ops."
    }
  ],
  "pricing_models": [
    {
      "model": "Per seat",
      "pros": "Simple to buy, predictable revenue, familiar to procurement",
      "cons": "Fully exposed to the heavy-user tail",
      "best_for": "Usage naturally bounded by human working hours"
    },
    {
      "model": "Pure usage",
      "pros": "Cost and revenue perfectly aligned, scales with value",
      "cons": "Bill shock, hard to budget, often blocked by procurement",
      "best_for": "Developer tools and infrastructure"
    },
    {
      "model": "Hybrid seat + overage",
      "pros": "Predictable floor for the buyer, capped downside for the vendor",
      "cons": "More complex pricing page, needs usage visibility",
      "best_for": "Most AI SaaS products — the current market default"
    },
    {
      "model": "Credit packs",
      "pros": "Prepaid cash, capped exposure, cost made visible",
      "cons": "Purchase friction, users ration usage, breakage policy risk",
      "best_for": "Consumer and prosumer AI tools"
    },
    {
      "model": "Outcome pricing",
      "pros": "Strongest value alignment, high willingness to pay",
      "cons": "Attribution disputes, hard to forecast, needs measurable outcomes",
      "best_for": "Agents with a clean, countable result"
    }
  ],
  "calculators": [
    "https://www.revenuelab.fyi/ai-feature-pricing-calculator",
    "https://www.revenuelab.fyi/ai-credits-pricing-calculator",
    "https://www.revenuelab.fyi/ai-seat-vs-usage-pricing-calculator",
    "https://www.revenuelab.fyi/ai-tool-roi-calculator",
    "https://www.revenuelab.fyi/ai-automation-payback-calculator",
    "https://www.revenuelab.fyi/ai-vs-headcount-calculator",
    "https://www.revenuelab.fyi/ai-freelance-rate-calculator",
    "https://www.revenuelab.fyi/ai-calculators"
  ]
}