Product · Private CDP · AI Operational Layer

Private customer data platform (CDP) for regulated industries

A customer data platform (CDP) brings customer records and events together. PrivyCDP runs on infrastructure your team controls. Explore how data enters the system, how records connect and how approved actions reach your tools.

Walk Through the Product ArchitectureReview Deployment Architecture →
Product design · single-tenant · MCP-ready · customer-owned
Product Architecture

Single-tenant · Customer-owned environment

Isolated

Identity Graph

Event Stream

Decision Engine

AI Agent Layer

Connectors

Audit & Governance

Outside boundary — blocked egress
AdTech CloudMarTech SaaSPublic Analytics
Designed for
Financial servicesInsuranceTelecommunicationsHealthcareGamingCommerce

Section 01 · Architecture Layers

Three Architectural Layers

The data flow has three parts: collect events, apply rules and send approved actions. Each part runs in your environment. Your team plans capacity and checks the logs.

Ingestion Layer

First-party event, batch, identity, and consent intake under customer-owned credentials

Event Collector

SDK · HTTP · Kafka

Batch Loader

Warehouse · Lakehouse

Identity Ingest

Deterministic · probabilistic

Consent Ingest

Policy · residency

Decision Layer

Identity resolution · real-time decisioning · AI agent operations · segmentation

Identity Graph

First-party resolution

Decision Engine

Real-time rules + ML

AI Agent Layer

Private LLM · MCP

Segmentation

Streaming + batch

Activation Layer

Journey orchestration · open activation interface · MCP-ready connectors · immutable audit

Journey Runtime

Workflow orchestrator

Activation Gateway

Reverse-ETL · Webhook

Connector Bus

MCP-ready interface

Audit Ledger

Hash-chained, immutable

Section 02 · Core Capabilities

Core Capabilities

Review four parts of the product. For each one, test what data it uses, who can change it and what gets logged.

UnifyAvailable

Unify First-Party Data

Connect customer records from your own data stores and event streams. Keep the work in the environment you control.

Customer-owned identity graph
DecideAvailable

Segment and Decide in Real Time

Group records and apply rules to incoming events or batches. Review rule versions and audit records in your own systems.

Real-time orchestration
OrchestrateAvailable

Orchestrate Customer Journeys

Build steps that send approved actions to your existing tools. Check each connector against the release you plan to use.

Open integrations
OperateAvailable

Operate with AI Agents

Use AI to help review groups, workflows, unusual events and reports. Set limits on data access and require review of the results.

MCP-ready · agent-assisted

Section 03 · Integration Interface

Integration Interface

Nine integration categories cover the operational interface area of an enterprise customer data platform. Every connector runs under customer-owned credentials, inside the customer environment.

Interface

Warehouses

Snowflake, BigQuery, Redshift, Azure Synapse — bidirectional sync under customer-owned credentials.

Interface

Lakehouses

Databricks, Iceberg, Delta — federated queries against the customer's existing lakehouse layer.

Interface

Identity Resolution

Deterministic and probabilistic graph stitching across systems, retained inside the customer boundary.

Interface

Event Streams

Kafka, Kinesis, Pub/Sub, Event Hubs — high-volume ingest with no metered per-event billing.

Interface

Activation Channels

Email, SMS, push, in-app, programmatic — customer-controlled credentials, no shared activation tax.

Interface

Reverse ETL

Push curated profiles and decisions to downstream systems of action — CRM, support, billing, ad platforms.

Interface

AI Models

Bring-your-own-model interface for private LLMs, embeddings, and scoring models inside the customer VPC.

Interface

Vector Stores

Pluggable vector backends for embeddings, retrieval, and feature stores — customer-owned persistence.

Interface

Audit & Logging

Immutable, hash-chained ledger plus structured logs into the customer SIEM and observability stack.

Section 04 · Open & MCP-Ready

Open APIs, Webhooks, and an MCP-Ready Agent Interface

PrivyCDP exposes open APIs, webhooks, and an MCP-ready interface for agent connectivity. Customer logic — segmentation rules, decision policies, journey graphs, and activation contracts — remains controlled, versioned, and auditable inside the customer environment. There is no shared decisioning plane, no opaque vendor model, and no forced public endpoint.

The AI operational layer is positioned as agent-assisted operations for the customer's own teams — not as autonomous marketing. Every AI-assisted action is recorded in the customer's immutable audit ledger alongside the human or system actor that authorized it.

Product compliance designFramework evaluation: SOC 2 · ISO 27001 · GDPR · HIPAA · PCI-DSS · FedRAMP

Planning your evaluation

Follow one customer event from source to activation

Use this proposed acceptance workflow to validate the published architecture against the integrations you actually need.

Ingest with an explicit contract

Define an event ID, source timestamp, customer identifier and permitted fields. Test a duplicate, a late event and a failed delivery. Agree who owns schema changes and how failed events are replayed.

Resolve identity without losing permission state

Document which identifiers may join profiles and which systems are authoritative. Test an identifier change and an opt-out together; verify that a merge preserves the intended suppression rules.

Activate to one approved destination

Choose a real destination and confirm the connector, authentication method, rate limits, retry policy and delivery receipt. External email, SMS and advertising destinations create egress even when the CDP is privately hosted.

Decide whether self-hosting fits

A private CDP needs an operating owner for infrastructure, access, backups and releases. If your team needs a fully managed service, price that scope explicitly. Category-level integration support is not proof of a ready-made connector for every named product.

Related engineering context: MailOps and activation delivery evidence. Use it with the page-specific evaluation questions above.

Discuss your deployment with engineers.

Bring your data flow, operating owners and evaluation questions.

Request an architecture reviewReview Enterprise TCO Model →Request permissioned references and release-specific evidence during evaluation.

Further reading

Published: · Updated:

Editorial dates describe content changes. Confirm current sources before making legal, security or purchasing decisions.