SocialHub.AI
Developer Center - Data Architecture

Why Can We Provide Real-time Data Capabilities to Agency AI?

In retail consumer goods, SocialHub connects API, SDK, POS, e-commerce and third-party platform data in real time, turning consumers, products, orders, stores, members and marketing touchpoints into governed customer intelligence.

Why Can We Provide Real-time Data Capabilities to Agency AI?Business facts and behavior events converge into governed customer intelligence served from the MPP data layer.BUSINESS DATA FLOWBEHAVIOR EVENT FLOWBusiness Systemsorders, products, storesAPI Accessauth, limits, mappingOLTP Storebusiness factsCDCdatabase changesClient Eventsbrowse, search, cartEvent Gatewayvalidate, enrichEvent Busreplay, fan-out, isolationStream Processingclean, window, tagsMPP Database360, cohorts, BIlifecycle, RFM, LTVSecurity + Governance Planeaccess control / signatures / rate limitsencrypted transport / topic isolationPII encryption / masking / tokenizationtenant isolation / RBAC / field permissionsconsent / opt-out / audit
The point

Real-time architecture turns retail data into customer intelligence.

The architecture keeps system responsibilities explicit: CRM OLTP protects retail business truth, change capture and event gateways feed the event bus, stream processing prepares governed facts, and the MPP database serves analytics as the final data layer.

Data source inventory

Reliable intelligence starts with complete retail signal coverage.

The architecture ingests state data, event data, business state and behavior signals from the systems where consumer retail work actually happens.

Business facts

members, orders, points, coupons, products, stores, inventory, payments and returns

POS transactions

store purchases, receipts, refunds, basket lines, cashier/store context and return reasons

E-commerce interfaces

Shopify, Amazon, TikTok Shop, marketplace orders, fulfillment, returns and subscriptions

Product & SKU master

SKU, SPU, category, brand, price, attributes, bundles, merchandising hierarchy and availability

Behavior events

page views, product views, search, clicks, add-to-cart, login, scan, content and conversion events

Customer service

tickets, chat, complaints, refund context, satisfaction, resolution outcomes and service recovery signals

Third-party platforms

ERP, ads, audiences, external membership, attribution, ROAS, logistics and service integrations

Decision feedback

recommendations, approvals, generated actions, guardrail decisions and outcome signals

Business facts

CRM OLTP keeps retail transactions reliable.

Member, order, point, coupon, product and store records are written to the transactional store first, then exposed to the streaming backbone through change capture.

Real-time processing

Stream processing stays scoped to real-time computation.

Cleaning, deduplication, windows, real-time tags, cohorts and feature-ready facts are computed as stream processing outputs.

Serving outputs

Analytics serving is the final data layer.

The MPP database serves customer intelligence, cohorts, event detail and governed metrics as the final data-serving layer.

Operating difference

From data movement to decision movement.

The stack is intentionally split into transactional store, event bus, stream processing and OLAP serving. Each layer has a clear boundary and a clear failure domain.

Traditional ETL
Extract retail data from POS, e-commerce, campaign tools and service systems; transform later; load into a reporting database.
Real-time CRM + CDP
Write business facts to CRM OLTP, stream changes and behavior events through the event bus, process them in real time, then serve governed analytics from the MPP database.
Primary output
Tables for offline analysis.
Operational output
Customer 360, tags, cohorts, attribution, lifecycle analytics and governed BI metrics.
Runtime path

A consumer signal becomes an action before it goes cold.

1

Business facts from POS, e-commerce, ERP and CRM APIs enter API Access and Data Router, then land in the CRM OLTP database for transactional consistency.

2

Change capture publishes member, order, coupon, product and inventory changes into the event bus.

3

Behavior events enter the event gateway, where they are authenticated, validated, enriched and written to the event bus.

4

The event bus unifies business change streams and behavior events while preserving replay, fan-out, isolation and consumer decoupling.

5

Stream processing performs cleaning, deduplication, windowing, tag computation, cohort updates and feature generation.

6

MPP Database serves event detail, transaction facts, customer 360, product affinity, store performance, coupon attribution, lifecycle cohorts, RFM, LTV and repurchase analytics.

Developer takeaway

Build against governed customer intelligence, not raw pipes.

Use APIs to connect retail systems, SDKs to capture behavior, change capture to stream business facts, and the MPP database to serve governed customer intelligence.

Related: the platform loop and web tracking SDK.