Unit economics
Digital SaaS · ICP 1 — AI-embedded SaaS — per-feature & per-customer margin
May 2026 · month-to-dateSynthetic data
Attribution coverage is 73.7% — some cost has no
customer_id. The unattributed residual is shown, never dropped. Review coverage →Cost per unitCOGS ÷ metered units
Revenue per unitRecognized ÷ metered units
Margin per unitRevenue − cost, per unit
Margin by customer
51 customersOne point per customer: x = revenue, y = cost. Points in red sit below break-even (negative margin) — the customers being served at a loss.
Losing money on 12 customers
- 1Unattributed
- 2Yellowstone Outdoors
- 3Atlas Logistics 2
- 4Beacon Analytics
- 5Cobalt Finance
Margin per unit · last 6 months
Tokens breakdown
in 8.3B · out 4.1B · cached 1.7B
Usage input for the per-unit cost denominator (input / output / cached split).
The cost↔revenue join (stamped customer_id / feature_id) is what turns infra-keyed cost into customer-keyed margin. Every figure is governed by @radicas/core — click any number for its two-layer provenance; the same metric keys back the MCP surface.