Process Intelligence Graph
The Process Intelligence Graph (PI Graph) is the foundation layer that captures how your business actually runs, across systems, across processes, and enriches that with business context and process knowledge so that you can not only see the current state, but act on opportunities. Using the PI Graph enables a shift from isolated process mining to enterprise-wide process intelligence, with scalability and the ability to embed into automation/AI.
At its core, the PI Graph is a system-agnostic, enriched digital twin of an organization’s business operations:
It pulls together process data from multiple systems (ERP, CRM, data lakes, even desktop/task data) into one unified model.
It uses an object-centric data model (objects like invoices, purchase orders, orders, etc) and captures how these objects and events relate to each other (object-to-object relationships, event-to-object relationships).
It then enriches that data with business context: KPIs, workflows, business rules, process knowledge (built from many deployments) so you get not just “what happens” but “how it happens” and “why it happens”.
The below is an example of a Process Intelligence Graph, created using the Objects & Events product area:
Using the PI Graph provides the following key benefits:
End-to-end visibility across systems: As the PI Graph is system-agnostic, you can map processes that span multiple tools, systems, departments (e.g., Order-to-Cash, Procure-to-Pay) in one unified view.
Reusable and scalable: The object-centric model means once you’ve built out objects/events for one process, you can reuse, extend, and scale across other processes without starting from scratch.
Common language for the business: By providing a unified semantic layer (objects, events, rules, KPIs) it helps business and IT align on how processes run and where value opportunities lie.
Improved insight and actionability: Because the PI Graph is enriched with process knowledge and AI and machine learning, you can more quickly identify where things are going wrong (bottlenecks, deviations) and what to do about them.
Foundation for automation and AI: The PI Graph serves as the data and context layer that enables automation, generative AI, process orchestration by exposing “how your business flows” in structure.
Here are the main building blocks of the PI Graph architecture, along with how they interact:
PI Graph component | Description | Purpose / role |
|---|---|---|
Object-Centric Data Model (OCDM) | The foundational data structure of the PI Graph. It represents real-world business objects (like Purchase Orders, Invoices, Deliveries, Customers) and events (like Create PO, Approve Invoice, Ship Goods) plus how they relate to each other. | Provides a multi-object, cross-process view of operations,beyond traditional single “case-based” process mining. |
Event data layer | Stores and connects all events (activities, status changes, transactions) from all systems. | Captures the “what happened” data that forms the backbone of the process graph. |
Object relationships layer | Defines and maintains relationships between business objects (e.g., a Purchase Order → multiple Goods Receipts → one Invoice). | Enables end-to-end process visibility and relationship tracing across systems. |
Knowledge layer | Encapsulates Celonis’ built-in process knowledge, learned from thousands of deployments: KPIs, benchmarks, best practices, known process patterns. | Enriches the graph with contextual business meaning, turning raw data into intelligent process models. |
Business context / semantic layer | Adds metadata: definitions of KPIs, business rules, roles, responsibilities, and operational logic. | Acts as a “common business language” connecting IT data to business terms and goals. |
Data integration layer | Connectors for ERP, CRM, SCM, HR, legacy systems, APIs, and streaming sources. | Feeds real-time or batch data into the PI Graph securely and consistently. |
Analytics and intelligence layer | Powers insights through process analytics, variant exploration, root-cause analysis, machine learning, and generative AI. | Enables diagnostic and predictive insights on top of the unified graph. |
Action and orchestration layer | Connects insights to action: triggering automations, alerts, or workflows directly from the PI Graph. | Turns insights into measurable operational improvements (“process optimization loop”). |
Reusability and governance framework | Includes shared definitions, templates, and governance structures for KPIs, objects, and models. | Ensures consistency, scalability, and reuse across processes and departments. |
Understanding how the key components work together
As a high-level summary, the key components in the Process Intelligence Graph work together in the following way:
Data ingestion: Data is collected from many sources (ERP systems, CRM, databases, task mining, on-premise clients). The on-prem client model allows extraction without opening firewall ports, etc.
Modeling / transformation: This raw data is mapped into the object-centric model (objects, events, relationships).
Graph construction: The PI Graph builds the unified process view, tracking how objects and events move across systems, how they interrelate, capturing the end-to-end value chain.
Enrichment with context/knowledge: Business rules, KPIs, process patterns, roles etc are layered on top. This helps turn raw process flows into insights and actionable intelligence.
Analytics and action: On top of the graph you run analytics, apply AI/ML, visualise performance, find bottlenecks, deviations, simulate changes. Then you can trigger actions or automation.
Scaling and reuse: Once you’ve built objects/events for one process you can reuse, extend, scale to other processes and systems, reducing duplication of effort.
Here’s a concrete, real-world use case example that illustrates why and when you’d use the PI Graph:
The challenge
A global manufacturing company runs its O2C process across multiple ERP systems (e.g., SAP ECC in Europe, Oracle in the U.S., and Salesforce for CRM).
They want to:
Identify bottlenecks and automation opportunities across the entire process.
Understand how customer order changes, delivery delays, and credit blocks interact to cause late payments.
Gain a unified view of the process, even though the data is fragmented across systems.
Standard process mining would require creating separate event logs for each system and process, and then manually stitching together insights. This makes it difficult to understand cross-system dependencies, like how a credit block in SAP affects invoice timing in Oracle.
The solution
The PI Graph acts as a semantic, interconnected data model, a “digital twin” of the business processes used by the global manufacturing company. It allows them to:
Connect multiple data sources dynamically (ERPs, CRMs, supply chain, etc.) into one coherent process view.
Model relationships between entities like orders, deliveries, invoices, customers, and payments.
Automatically detect cause-and-effect relationships — e.g., how often delivery delays cause overdue invoices.
Enable real-time analysis across systems without manually rebuilding event logs.
With the PI Graph, the global manufacturing company discovers:
High percentage of late payments are linked to delivery delays caused by incomplete material availability.
Orders from a specific region have higher delay rates because credit blocks in SAP are not cleared promptly after a Salesforce order modification.
Automatically recommends automation (via Celonis Action Flows) to notify credit controllers when a credit block causes invoice delay.