Skip to main content

Overview

Decision flows are the core execution engine of Kaireon. They are visual pipelines built with a drag-and-drop editor that define how offers are selected, scored, and ranked for each customer.

Flow Stages

A decision flow consists of ordered stages:

Enrich

Loads customer data from schema tables at decision time. Data is cached in Redis for performance. Enriched fields are available as customer.* variables in subsequent stages.

Compute

Evaluates formula-based computed fields per candidate offer. Uses the formula engine to calculate personalized values like base_rate * (1 + customer.loyalty_tier * 0.05).

Filter

Applies qualification rules to narrow the candidate set. Removes offers the customer isn’t eligible for.

Score

Applies scoring models to rank offers. Supports multiple scoring engines:
  • Scorecard — Weighted attribute scoring
  • Bayesian — Bayesian probability model
  • Gradient Boosted — ML-based scoring

Rank

Final ranking using arbitration weights. Produces the ordered list of recommendations.

Flow Configuration

Each flow includes:
  • Entry conditions — When this flow should execute
  • Candidate selection — Which offers to consider
  • Stage ordering — The sequence of processing stages
  • Output config — How many offers to return, response format

Decision Traces

Kaireon can record forensic traces of flow execution for debugging and auditing. Traces capture:
  • Input context (customer ID, attributes)
  • Each stage’s input/output
  • Scores and rankings at each step
  • Final recommendations with reasoning
Tracing is tenant-configurable with a sample rate setting.