Prevenue Docs

Intelligence Learning Evidence

Understand learned timing, confidence, preventability, suppression, coverage, and instrumentation guidance used in account and Funnel context.

Intelligence learning is Prevenue's read on which product-event patterns tend to appear before revenue movements. It refreshes as product events, account snapshots, billing outcomes, and reviewed attribution arrive.

These statistics can inform Account investigation, Signal explanations, and bounded Funnel defaults. They summarize evidence and do not expose raw event payloads.

Core definitions

StatDefinition
Lead-time distributionObserved time between the first usable learned signal and a later revenue movement, shown as percentiles rather than one outlier.
Signal half-lifeHow long a learned signal usually remains actionable after it first appears.
Opportunity windowRecommended action window derived from lead time and half-life.
Preventability scoreA 0-100 prioritization score for Save motions. It is not proof that an intervention will save an account.
Suppression intelligenceGuidance for when Save or friction evidence should suppress Grow/Convert messaging, select a support path, or remain internal.
Prediction coverage frontierWhich outcome groups have decision-ready, directional, exploratory, or missing predictive evidence.
Instrumentation ROIData gaps most likely to improve future model coverage, such as identity, mappings, value metrics, or mature outcomes.
Evidence strengthQualitative amount and consistency of evidence behind a recommendation.
Model confidenceConfidence in the learned recommendation, separate from account score and outcome probability.
CaveatsSample-size, freshness, missing-input, excluded-outlier, or non-causal interpretation limits.

How Funnels use learning

Funnels may use learned evidence to propose reviewable timing, cooldown, suppression, path, or follow-up defaults. A prediction is not billing, entitlement, identity, or outcome authority.

Keep learned inputs explainable and bounded:

  • Preserve the confidence score even when the interface uses Low, Medium, or High tolerance.
  • Fall back to deterministic eligibility or no render when evidence is stale or below tolerance.
  • Keep Save suppression visible when it changes a Grow or Convert path.
  • Record the evidence version used for a decision.
  • Require authoritative evidence before reporting a conversion.

Evidence states

  • Insufficient — not enough mature movement or matched-account evidence.
  • Directional — useful for investigation or ranking, but not for controlling eligibility.
  • Decision-ready — validated evidence, confidence, freshness, fallbacks, and reporting support bounded Funnel use.

Thresholds can evolve by model and workspace. The state and caveats matter more than memorizing a global number.

Interpreting results

Treat preventability as prioritization, not proof. Treat lead time and stale-after as timing guidance, not guaranteed outcome dates. Treat instrumentation ROI as a setup recommendation, not a precise uplift forecast.