{
  "schema_version": "1.0",
  "surface_type": "guide_answer",
  "guide_id": "model-switch-margin-impact",
  "question": "How do I quantify margin impact when switching from model A to model B?",
  "canonical_url": "https://www.unitcostai.com/guides/model-switch-margin-impact",
  "related_tool_url": "https://www.unitcostai.com/tools/model-switch-impact",
  "formula": "margin_delta_pct = margin_candidate_pct - margin_baseline_pct",
  "assumptions": [
    "Usage, retrieval, and caching assumptions stay identical across runs",
    "Both runs use the same daily pricing snapshot context",
    "A stronger model can be the wrong default when unit-price increase exceeds retry or quality gain on routine work",
    "Quality risk is evaluated separately from pure margin math"
  ],
  "example": "Baseline margin 74.0% and candidate margin 79.5% gives +5.5 margin points. If a premium candidate only helps a small share of sessions but raises every default request, route it to exceptions instead of making it the default."
}