{
  "project_id": "AI-03",
  "title": "Account reconciliation",
  "source_model": "finance-hours-v1-2026-10-22/model.json",
  "source_category": "Accounting and close",
  "scope": "One reconciliation workflow: two datasets, one entity, three deterministic matching rules and up to ten exception types; use two historical periods for development and validation, with up to five initial users. GenAI explains differences and drafts investigation notes with source links; deterministic checks verify amounts; no automatic journal posting.",
  "basis": "Existing working judgment estimates; case studies do not calibrate these hour ranges.",
  "work": [
    {
      "title": "Define matching rules and review requirements",
      "phase": "process",
      "phase_share": 1,
      "hours": [
        16,
        32,
        56
      ],
      "basis": "Agree matching rules, exception definitions, materiality and evidence required to clear a difference; assign preparer and reviewer duties."
    },
    {
      "title": "Prepare reconciliation data and reference periods",
      "phase": "data",
      "phase_share": 1,
      "hours": [
        20,
        40,
        80
      ],
      "basis": "Map the two exports and identifiers, prepare period-specific examples and check completeness, permissions and deterministic control totals."
    },
    {
      "title": "Configure matching and exception explanations",
      "phase": "build",
      "phase_share": 1,
      "hours": [
        28,
        56,
        112
      ],
      "basis": "Configure matching and exception routing, grounded explanations, evidence references and reviewable reconciliation output."
    },
    {
      "title": "Validate balances and investigation notes",
      "phase": "validation",
      "phase_share": 1,
      "hours": [
        32,
        64,
        128
      ],
      "basis": "Compare against completed periods, test missing/duplicated records and misleading explanations, independently verify totals and document remaining differences."
    },
    {
      "title": "Train preparers and pilot the reconciliation",
      "phase": "rollout",
      "phase_share": 1,
      "hours": [
        12,
        24,
        48
      ],
      "basis": "Document preparer/reviewer steps, train initial users, shadow one production run and agree fallback and release criteria."
    }
  ],
  "total_hours": [
    108,
    216,
    424
  ],
  "annual_upkeep_hours": [
    32,
    64,
    128
  ],
  "upkeep_basis": "Judgment: monthly mapping/error review at 2/4/8 hours, plus quarterly regressions after rule, model or export changes at 2/4/8 hours. Performing and reviewing ordinary period reconciliations is separate operating work.",
  "boundary": "Need reconciliation-level implementation and validation hours, the number of records and exception classes, and the human review effort that remains after deployment.",
  "organization_assumptions": "Central task-level settings in organization-model.json; reference hours remain unchanged in this file.",
  "difficulty": {
    "label": "Moderate",
    "reasons": [
      "Identifiers and amounts must reconcile across both datasets.",
      "Matching rules must distinguish genuine differences from data issues.",
      "Explanations must cite evidence without inventing a cause.",
      "Finance reviewers must validate exceptions and control totals."
    ]
  }
}