"Revenue-grade" is not a model size. It is a bar for ownership, escalation, and a KPI the CFO can verify on Monday morning.
Three tests before production
We use three tests before any agent touches revenue workflows.
- Ownership — a named human manager with weekly review cadence.
- Escalation — a person, not a queue, with a documented SLA.
- KPI — one metric on the P&L or pipeline report, not a vanity score.
Why demos fail the bar
Demos optimize for the happy path. Revenue-grade systems optimize for variance: wrong inputs, edge cases, angry customers, and quarter-end pressure.
If your agent cannot explain why it made a decision in plain language, it is not revenue-grade yet.
Rolling out without betting the quarter
We ship in phases: shadow mode, human-approved sends, then partial autonomy with caps. Each phase has a rollback switch and a KPI gate.
Teams that skip shadow mode usually roll back twice as hard later.
Field note
Revenue-grade is a deployment discipline, not a benchmark score.