RevenuePilotby Foxes TechnologyRevenuePilot's engine is deterministic. It reads reconciled source data, applies your explicit guardrails, and returns a short ranked queue of moves an operator can act on—each one carrying the arithmetic and the boundary that produced it.
Every step is inspectable. If a step cannot be evidenced, the recommendation is withheld rather than downgraded to a guess.
Products, options, departures, bookings, cancellations, costs and commissions are reconciled from your connected system. Unreconciled or stale inputs are flagged, not silently filled in.
Booking pace and remaining capacity are derived from signed source data and compared against historical lead-time cohorts for the same product and departure window.
The engine persists its own hindcast error and coverage, then calibrates a confidence score. Low-confidence signals are excluded from eligibility instead of being presented with false precision.
Floor, ceiling, maximum movement, minimum confidence and cooldown are applied as hard constraints. A proposal that violates any constraint is rejected, not clipped and shipped.
Surviving moves are ranked by contribution impact, urgency and operational pressure, and the workspace queue is deliberately bounded so a revenue review stays finishable.
The signal, economics, uncertainty, policy boundary and required operator action are recorded together, so the decision can be defended after the fact.
Guardrails are evaluated in code before a recommendation is ever shown. They are the same constraints re-checked at execution time, so what you review is what could run.
| Guardrail | What it constrains | When it is enforced |
|---|---|---|
| Minimum confidence | Excludes weakly-evidenced signals from eligibility | Generation and execution |
| Maximum movement | Caps how far a single decision may move a price | Generation and execution |
| Cooldown window | Prevents repeated changes to the same product | Generation and execution |
| Price floor and ceiling | Holds every proposal inside your commercial corridor | Generation and execution |
| Daily write limit | Bounds how many changes can occur in one day | Execution |
| Expected-version check | Refuses to overwrite a price that changed underneath us | Execution |
Language models are excellent at explaining a decision and unreliable as the decision itself. RevenuePilot draws that line in the architecture.
Revenue Copilot can describe why a recommendation exists, compare priorities and clarify a boundary. It cannot create an executable signal, alter a guardrail or trigger a write—those paths simply do not accept model output.
A move that lifts revenue while eroding contribution is not a win. Where verified variable cost and channel commission are available, the engine ranks on contribution impact and states plainly when cost coverage is incomplete.
Recommendation generation is keyed by organization, request key, request hash, source snapshot, policy and cycle. Replaying the same request returns the existing recommendation; reusing a key with different evidence fails loudly instead of quietly forking the truth.
A revenue system earns trust by being explicit about its limits. These are ours.
No. Live price execution is disabled by default. Enabling it for a product requires a certified connector, a completed shadow qualification period, reviewed policy settings and explicit operator approval—and even then movement stays inside your configured caps.
The engine distinguishes between 'empty' and 'unknown'. Missing variable cost, stale source data or unreconciled bookings produce a flagged gap and a withheld recommendation rather than a confident-looking guess built on a default value.
Yes. Every recommendation carries a decision receipt containing the source snapshot, the economics used, the uncertainty, the policy boundary that applied and the operator action required.
No. The deterministic engine produces and bounds every recommendation. The AI surface explains, compares and investigates; it has no route to create a signal, change a guardrail or execute a write.
Start with a scoped portfolio, agreed guardrails and a measurable review cadence. No live price writes are required to evaluate the engine.