The AI Billing Data Model You Actually Need

The minimal tables

• Customers, Accounts, Subscriptions
• Agents & Signals (events you meter)
• UsageEvents (immutable, versioned)
• Costs (vendor, rate, allocation)
• PriceVersions (effective_at, rules)
• Invoices & LineItems (ties to usage + outcomes)

Event example (JSON)

{
  "customer_id": "acme",
  "agent_id": "outbound_v2",
  "signal": "meeting_held",
  "quantity": 1,
  "metadata": {"duration_min": 27, "attendee_role": "VP Sales"},
  "occurred_at": "2025-08-28T09:15:00Z",
  "price_version": "v1.3"
}

Cost example (allocation)

• LLM tokens → workflow step → customer.
• RAG queries/minutes → feature → customer.
• Storage/egress → monthly proration → customer.

Rev-rec pointers

• Usage lines: recognize on consumption.
• Outcome lines: recognize on completion/acceptance.
• Commitments: recognize over term; overages as consumed.

Why this model future-proofs you

You can switch value metrics, add outcomes, or re-price without data migrations—just add a new PriceVersion and keep the event log immutable.

Want the full schema (SQL) as a downloadable? Ping us—we'll share the UnitPay starter pack.