Skip to main content
Every customer interaction is an opportunity. The challenge is picking the right offer from dozens of options, personalizing it for the individual, and delivering it through the channel most likely to convert — all in milliseconds. KaireonAI is a Next-Best-Action decisioning platform that does exactly this, combining data ingestion, offer management, ML scoring, and real-time arbitration into a single open-source system.
KaireonAI is Apache-2.0 licensed and fully self-hostable. Run it on your infrastructure with complete control over your data and models.

​
The Problem KaireonAI Solves

Most companies have offers, campaigns, and messages spread across disconnected tools. Email marketing knows about promotions. The app team has banners. The call center has scripts. Nobody is coordinating which customer should see which offer when. The result: customers get irrelevant recommendations, miss time-sensitive offers, or get bombarded on every channel at once. KaireonAI replaces this fragmented approach with a single decision engine that evaluates every eligible offer for each customer, scores them using ML models, and returns the best options ranked by a combination of propensity, business value, relevance, and priority.

​
See It in Action: Starbucks Rewards

Imagine you operate the Starbucks Rewards program. You have 10 offers — BOGOs, discounts, and informational messages — and 6 channels: web, email, mobile push, social, batch email, and manual outreach. When a customer opens the app, which offer should they see? Without KaireonAI: You show the same “20% off Frappuccino” banner to everyone, regardless of whether they prefer lattes, already redeemed a discount today, or typically only respond to BOGO offers. With KaireonAI: The Recommend API evaluates all 10 offers for this specific customer, filters out anything they have already seen this week (contact policies), scores the rest using a trained model that considers their purchase history and income level, and returns the top 3 — personalized, ranked, and ready to display. 
curl -X POST http://localhost:3000/api/v1/recommend \
  -H "Content-Type: application/json" \
  -d '{
    "customerId": "cust-001",
    "channel": "Mobile",
    "attributes": { "age": 30, "income": 75000 },
    "count": 3
  }'
The response includes scored, ranked offers with creative content ready for rendering: 
{
  "decisions": [
    { "offerName": "BOGO Frappuccino", "score": 0.85, "rank": 1, "channel": "Mobile" },
    { "offerName": "25% Off Merchandise", "score": 0.72, "rank": 2, "channel": "Mobile" },
    { "offerName": "Earn 3x Stars This Week", "score": 0.68, "rank": 3, "channel": "Mobile" }
  ],
  "meta": { "totalCandidates": 60, "qualifiedOffers": 8, "durationMs": 45 }
}
The Starbucks dataset ships as built-in sample data. Load it in Settings > Sample Data and follow the full tutorial to build this from scratch.

​
Key Capabilities

Real-Time Decisioning

The Recommend API evaluates, scores, and ranks offers in under 200ms. Every request runs through enrichment, qualification, contact policies, ML scoring, and arbitration — no pre-computation required.

PRIE Arbitration

The multiplicative PRIE formula (Propensity x Relevance x Impact x Emphasis) produces a single score from four dimensions. A zero in any dimension eliminates the candidate — no irrelevant high-value offers sneak through.

8 Scoring Engines

From transparent scorecards (no training data needed) to neural collaborative filtering (learns latent preferences from interaction data). Start simple, upgrade without changing your Decision Flows.

Omnichannel Delivery

Email, push, SMS, in-app, web, WhatsApp, webhook, and direct mail — all managed from one platform. Contact policies enforce frequency caps per channel so you never over-contact.

Visual Pipeline Editor

Decision Flows are built on a drag-and-drop canvas with 16 composable node types across 3 phases: Narrow, Score & Rank, and Output. No code required.

Built-In Experimentation

Champion/challenger testing with holdout groups, deterministic traffic splitting, uplift calculation (z-test), and optional auto-promotion. Measure real impact before rolling out changes.

24 Data Connectors

Ingest customer data from S3, Snowflake, BigQuery, PostgreSQL, Kafka (batch polling), and more — 17 connectors are production-ready and 7 are coming soon. Build visual ETL pipelines with 15 transform types.

Five Monitoring Dashboards

Operations, Business, Data Health, Model Health, and Attribution dashboards with real-time metrics, Prometheus integration, and actionable alerts.

LLM Explanations

Turn any decision trace into a written explanation in three modes — regulator, agent, or customer. PII-redacted, cached per tenant, audit-logged for regulator mode.

​
Platform Architecture

ModuleWhat It DoesKey Features
Data PlatformIngests and transforms customer data25+ connectors, entity schemas with real PostgreSQL tables, visual ETL pipelines
Decisioning StudioConfigures what you can recommend and howOffers, categories, creatives, qualification rules, contact policies, Decision Flows
Algorithms & ModelsScores and ranks candidates using ML8 scoring engines, experiments, champion/challenger, auto-learning
DashboardsMonitors everything in real timeOperations, business KPIs, data health, model health, attribution

Architecture Deep Dive

Decision engine internals, data pipeline design, infrastructure adapters, and security model.

​
Start Here

Try the Playground

Create a free account and explore the platform instantly — no setup required.

Self-Host in 5 Minutes

Clone, install, and run your first recommendation locally.

Starbucks Tutorial

Build a complete decisioning pipeline step by step using real Starbucks Rewards data.