Platform Overview

Many Loops. One Growth System. Every Cycle Compounds.

A composable growth system: many always-on loops running at once. Audience intelligence decides who to reach. Orchestration runs the move across Meta, Google, The Trade Desk and your owned channels. Measurement proves the lift with holdouts, then trains the next loop. All on the audience and context graph you own, self-serve, growth-engineer led, or headless.

Inside the Loop

From audience interest to compounding outcomes.

One always-learning loop across D2C and B2B teams.

Audience

B2B SaaS, 50-200 employees
Healthcare vertical, Enterprise

Interest Graph

Agents

Decisions

Learning Loop

Results feed back. Segments sharpen. Decisions compound.

Phase 1 · Foundation

Your audience, interest, and context graph.

Identify, match, enrich, and monitor your audience without moving data (Snowflake, Databricks, BigQuery). The Signals Hub reveals visitors and layers in intent signals. An ontology turns those tables into shared meaning, and a signals waterfall keeps FIRE scores, Fit, Intent, Recency, Engagement, dynamic.

Audience

Immutable IDs anchored in your data cloud.

• Immutable hashed IDs (HMAC-SHA-256), never expire, never rented
• Identity resolution across people, accounts, devices, emails
• Activity schema, canonical, time-aware event model
• External Data & Signals Graph: 50+ sources, pre-unified
• AI lookalikes generated from your best buyers

Interest Graph

Signals Hub: 1P + 2P + 3P signals unified, time-aware.

• Signals Hub: visitor reveal plus intent signals, in real time
• Keywords buyers search and the answers that cite your brand
• Psychographics and behavior, enriched with consent-based 2P/3P data
• Signals waterfall keeps FIRE scores live as interest decays

Context Graph & Ontology

The linked context your agents reason over, not raw tables.

• Context graph links four pillars: audience, brand, domain, outcome
• Linked relationships your agents traverse before they decide
• Ontology maps warehouse tables to shared business meaning
• One semantic layer, so every agent reads the same definitions

Phase 2 · Activation

Audience activation to every channel. Match-rate boosting built in.

A play is a winning move packaged once: who to reach, on which channel, inside which guardrails. A loop runs that play always-on, so every outcome retunes the next run. Role-based iWorkers do the running, and you keep approval over spend and risk.

Agents (Role-based iWorkers)

• iWorkers: digital twins of the roles your team already runs
• Tribal knowledge encoded: your naming, thresholds, exclusions
• A named teammate per role: ops, lifecycle, analytics, growth
• iHarness orchestrates them, so you talk to one, not many

Orchestration: Plays & Loops

• Plays package a winning move once, then reuse it everywhere
• Loops run always-on across paid, owned, and retail media
• 200+ integrations: Meta, Google, LinkedIn, The Trade Desk, and more
• One API: MCP Server, CLI, and iConnect, our tool router
• Bidirectional sync with a Decision Trace on every action
• Match rates from 20-40% to 70-90%, PII-safe on every payload

Phase 3 · Optimization

Every cycle smarter. This is your moat.

The Decision Fabric decides who, when, and what: causal AI + the Decisioning Waterfall (ranked rules-then-models) + FIRE scoring. Every decision is logged as a Decision Trace, auditable, explainable, defensible. Outcomes feed back. Segments sharpen. Agents improve. The platform compounds.

Decision Fabric

• Decision Engine: Decisioning Waterfall + Causal AI + propensity models
• Guardrails, evals, policies, and controls on every decision, via iHarness
• Grounded in your context graph, never in raw tables
• Decision Traces: every decision logged, signal to iWorker to FIRE to action to outcome
• SLMs run inside your data cloud: no raw data leaves, compute stays cheap

Measurement & Learning

• Outcome memory, every decision and its result, written back to your data cloud
• Causal AI: incrementality across channels, not last-touch guesswork
• Feedback loops retune FIRE scores, audiences, and agent prompts
• Compounding memory, the brand-specific learning asset competitors cannot replicate

The Decision Trace Loop

From Clicks and Sessions to Decision Traces

Analytics tracked anonymous traffic. Now every decision leaves a trace: who, why, action, outcome, human or agent.

The old world:

• Anonymous sessions stitched by cookies
• Clicks count traffic, not people
• Last-touch guesses at what worked
• AI agents acting on your behalf are invisible to analytics

The new world:

• Humans and agents are first-class, identified actors
• Every action ties to an immutable ID
• Each decision links to its outcome in one trace
• Causal measurement replaces guesswork

EVENTS: Every human and agent action, captured server-side

IDENTITY: Resolved to immutable IDs: person, account, or agent

SIGNALS: Scored, consented, FIRE-ranked in real time

DECISIONS: Who, when, what, where, inside policy gates

OUTCOMES: Revenue, pipeline, and lift tied back to each decision

every result becomes the next signal

loops back to EVENTS

Decision Trace Examples

Decision Trace #8412 - Sample WIN-BACK LOOP

• Who: 2,314 lapsed buyers, segment: category fatigue
• Why: repurchase window exceeded 1.6x · FIRE ≥ 68 · consent verified
• Action: Klaviyo flow B + Meta exclusion refresh, inside iHarness gates
• Outcome: $41k revenue vs. holdout, 14 days

Decision Trace #9107 - Sample COVERAGE LOOP

• Who: 41 unengaged decision makers across 12 target accounts
• Why: account FIRE ≥ 72 · three buying roles uncovered · no contact in 45 days
• Action: LinkedIn + The Trade Desk account audience, sequenced to sales
• Outcome: 9 more accounts engaged vs. holdout, 21 days

Same loop, same trace, B2B pipeline instead of repeat purchase.

Customers & Partners

Trusted by customers and partners

Features FAQs

What growth and data teams ask us.

• Why is iCustomer so opinionated?
  Blank canvases fail: iCustomer ships with domain expertise built in: proven plays, an opinionated data model, and decision science refined across D2C and B2B growth. Your team does not design a system from scratch; you set goals and guardrails, and a fully AI-native stack, built for agents rather than retrofitted with them, does the heavy lifting. Your taste stays in charge.

• Why an MCP (Model Context Protocol) Server?
  Your AI tools should be able to ask your data questions: the iCustomer MCP Server exposes audiences, FIRE scores, and decisions as governed tools any agent can call, with the iHarness enforcing guardrails on every action. One API instead of tool bloat: your copilots, agents, and workflows all act on the same trusted context, with PII protected throughout.

• Can I use iCustomer with Claude or Codex?
  Yes. If you can describe it in Claude Code or Codex, you can ship it: connect to the MCP Server, and your assistant can query audiences, draft decision loops, and deploy them through the CLI into the warehouse and tools you already run. Claude-native and headless teams run the entire loop as code, versioned in git, without ever opening a UI.

• Is iCustomer multiplayer?
  Yes, by design. A brand runs many decision loops at once, and many people run them together: growth sets goals and targets, data governs the foundation, ops approves budgets and policies, and AI agents execute inside those guardrails. Everyone, human or agent, works from one shared context, ontology, and audit trail, so there are no conflicting segments, no double-sends, and no wondering who changed what. One system, one login, every team in the loop.

• Is Audience Loop a separate product?
  No. There is one platform and three ways in. Audience Loop is the self-serve way in: product-led, credit-based, one-prompt launch, fast install, every module, Audience through Measurement & Learning. Growth Engineer led is the enterprise way in: a platform and a partner, our engineers embed with your team until your loop compounds, audience-tiered, full causal AI and iWorker depth. And for code-led teams, Claude Code, Codex, or headless CLI, the same loop ships as code into the data cloud you already run. One platform underneath, always.

• What about PII and compliance?
  Locked down by design. Nothing moves and nothing gets replaced: iCustomer is not a CDP, it runs warehouse-native on Snowflake, Databricks, and BigQuery with zero data copies and zero raw-data egress, plus PII and leak protection on every activation payload. SOC 2 Type II certified, consent enforced at decision time, and every action leaves an auditable Decision Trace.

Turn your data into outcomes

Start free with Audience Loop. Or talk to us about a Platform pilot on your own data warehouse.