What is Process Mining?
Process mining extracts knowledge from event logs in information systems, enabling fact-based process analysis and improvement.
Global Market Value
$3.8B
2024 estimate
Fortune 500 Adoption
68%
+12% YoY
Avg. Process Efficiency Gain
30%
After deployment
Celonis Market Share
~27%
Leader in Gartner MQ
The Core Idea
Every time a business transaction happens — a purchase order is created, an invoice is approved, a ticket is resolved — the system logs a timestamped event. Process mining reads these digital footprints and reconstructs the actual process, not the assumed one. It answers: What really happens in our business, and why?
Three Pillars of Process Mining
Discovery
Build the process model automatically from raw event data — no manual mapping required.
Conformance
Compare actual process execution against the target model. Identify where deviations occur.
Enhancement
Extend or improve existing models using additional event data (performance, decisions, resources).
Traditional Analysis vs Process Mining
Organizations have long used interviews, workshops, and periodic reports to understand their processes. Process mining replaces assumptions with evidence.
📊 Traditional Analysis
⏱ Manual workshops take weeks or months to complete
🧍 Relies on expert opinions and interviews — often biased
📸 Snapshot-in-time; doesn't capture variation over time
📉 Sample-based — only a fraction of cases analyzed
🔁 Loop detection and rework are frequently missed
💸 High consulting costs for periodic reviews
⬡ Process Mining
⚡ Automated analysis in minutes from existing system logs
📡 100% objective — based on actual data, not perception
🔄 Continuous monitoring — captures process drift over time
📊 Entire population of cases, not just samples
🔍 Rework, loops, and bottlenecks precisely quantified
💡 Self-service insights for operational teams
When to use which?
Traditional analysis is still valuable for new process design (where no historical data exists) and qualitative context gathering. Process mining excels when the process is already running in a system like SAP, Salesforce, or ServiceNow — which is the case in virtually every modern enterprise.
What is Celonis?
Celonis is the global leader in process mining software, founded in Munich in 2011 by three TU Munich students. It turns process data into execution capacity.
Founded
2011
Munich, Germany
Valuation
$13B
As of 2021 funding
Customers
3,000+
Across 60+ countries
Employees
3,500+
Globally
Celonis Execution Management System (EMS)
PROCESS DATA LAYER
Connects to SAP, Salesforce, ServiceNow, Oracle — extracting full event histories from any ERP or CRM system.
PROCESS ANALYTICS
Real-time dashboards, OLAP filters, custom KPIs, and AI-powered root cause analysis.
ACTION ENGINE
Pushes recommendations directly into SAP or other systems. Closes the loop from insight to execution.
PROCESS APPS
Pre-built apps for P2P, O2C, Logistics, Finance — with industry benchmarks out of the box.
Key Differentiators
| Feature | Celonis | Generic BI Tools |
|---|---|---|
| Process-native | Yes | No |
| ERP connectors (SAP, Oracle) | Pre-built | Custom |
| AI-driven root cause | Yes | No |
| Action automation | Yes | No |
| Industry benchmarks | Built-in | Unavailable |
Data Mining vs Process Mining
Both extract insights from data — but they ask fundamentally different questions.
🗃️ Data Mining
🔎 Goal: Discover patterns in structured datasets
📦 Input: Static tables, databases, feature vectors
🔢 Output: Classification models, clusters, association rules
📌 Focus: What is in the data?
🧪 Techniques: Regression, clustering, decision trees, neural nets
⬡ Process Mining
🔎 Goal: Reconstruct and analyze business processes
📦 Input: Event logs with Case ID, Activity, Timestamp
🔢 Output: Process models, flow maps, conformance reports
📌 Focus: How does the process actually flow?
🧪 Techniques: Alpha algorithm, Heuristics Miner, Inductive Miner
The Minimal Event Log Schema
| Case ID | Activity | Timestamp | Resource |
|---|---|---|---|
PO-1001 | Create PO | 2024-01-10 09:00 | Alice |
PO-1001 | Approve PO | 2024-01-10 14:30 | Bob |
PO-1001 | Goods Receipt | 2024-01-13 11:00 | Carol |
PO-1002 | Create PO | 2024-01-10 10:15 | Alice |
PO-1002 | Goods Receipt | 2024-01-11 08:45 | Dave |
PO-1002 | Approve PO | 2024-01-11 16:00 | Bob |
PO-1002 has a deviation: Goods were received before PO approval — a control violation. Data mining would never catch this; process mining flags it immediately.
Benefits of Process Mining
From cost reduction to compliance — process mining creates measurable business value across every function.
Cost Reduction (P2P)
25%
Typical savings
Touchless Invoice Rate
80%
After automation
Compliance Deviation Rate
-60%
Post-implementation
Process Analysis Time
-90%
vs traditional methods
Core Benefits by Category
OPERATIONAL EFFICIENCY
• Eliminate bottlenecks with real throughput data
• Identify rework loops and redundant steps
• Prioritize automation targets using data
• Identify rework loops and redundant steps
• Prioritize automation targets using data
COMPLIANCE & RISK
• Detect SoD violations and policy breaches
• Continuous audit trail from system logs
• Regulatory reporting backed by evidence
• Continuous audit trail from system logs
• Regulatory reporting backed by evidence
COST REDUCTION
• Reduce manual effort through targeted automation
• Cut late payment penalties in P2P
• Lower DSO in O2C processes
• Cut late payment penalties in P2P
• Lower DSO in O2C processes
CUSTOMER EXPERIENCE
• Reduce order-to-delivery cycle time
• Identify failure points in customer journeys
• Benchmark against industry peers
• Identify failure points in customer journeys
• Benchmark against industry peers
Business Challenges in Modern Companies
Large organizations face structural complexity that makes traditional management nearly impossible. Process mining was built to address this directly.
Common Pain Points — and PM Solutions
| Challenge | Impact | How PM Solves It |
|---|---|---|
| Process fragmentation across ERP modules | High | Unified view across SAP MM, SD, FI modules |
| Invisible rework and manual workarounds | High | Loop detection in process graphs |
| Missed SLAs and late deliveries | Medium | Throughput time analysis per case variant |
| Lack of automation ROI clarity | Medium | Automation candidate scoring with frequency data |
| Audit preparation is manual and slow | Medium | Automated compliance reporting from event logs |
| Digital transformation without baseline | Strategic | As-is process capture before redesign |
The Visibility Gap
Organizations typically operate with only 12–18% visibility into their end-to-end processes. The rest is hidden in system logs, email threads, and tribal knowledge. Process mining closes this gap — not with consultants and workshops, but with the data that already exists in every system of record.
Process Mining Market Growth
From a niche academic concept to a multi-billion dollar enterprise software category in under a decade.
Market Size 2019–2028 (USD Billion)
CAGR 2023–2028
33%
Compound annual growth
Key Drivers
AI + ERP
modernization
modernization
Top Vendors
Celonis
UiPath · SAP
UiPath · SAP
Digital Footprint
Every action in an enterprise system creates a timestamped record. These "digital breadcrumbs" are the raw material of process mining.
Purchase Order: PO-2204 — Full Timeline
Jan 10 · 09:02 AM
Purchase Order Created
Jan 10 · 11:30 AM
3-Way Match Initiated
Jan 11 · 08:45 AM
Goods Receipt Before Approval ⚠
Jan 13 · 02:15 PM
PO Approved
Jan 15 · 10:00 AM
Invoice Received
Jan 16 · 04:30 PM
Payment Cleared
Where Do Footprints Come From?
ERP Systems
SAP, Oracle EBS — every document change logs a line item in the change document table (CDPOS, CDHDR)
SAP, Oracle EBS — every document change logs a line item in the change document table (CDPOS, CDHDR)
CRM Systems
Salesforce logs every stage transition in opportunity history — the backbone of O2C mining
Salesforce logs every stage transition in opportunity history — the backbone of O2C mining
ITSM Platforms
ServiceNow, Jira — ticket status transitions form a clean event log for incident process mining
ServiceNow, Jira — ticket status transitions form a clean event log for incident process mining
Custom Applications
Any system with an audit trail or state machine can be connected — even legacy systems via ETL
Any system with an audit trail or state machine can be connected — even legacy systems via ETL
Anatomy of an Event Log
The event log is the input to every process mining algorithm. It must contain at minimum: Case ID, Activity, and Timestamp.
P2P Process — Event Log (20 cases, 87 events)
| Case ID | Activity | Timestamp | Resource | Duration | Status |
|---|---|---|---|---|---|
| PO-001 | Create PO | 2024-01-02 09:00 | Alice | — | Start |
| PO-001 | Approve PO | 2024-01-02 14:00 | Bob | 5h | On-time |
| PO-001 | Goods Receipt | 2024-01-05 10:00 | Carol | 2d 20h | Compliant |
| PO-001 | Invoice Received | 2024-01-06 09:30 | System | 23h | Matched |
| PO-001 | Payment Cleared | 2024-01-10 16:00 | Finance | 4d 6h | Complete |
| PO-002 | Create PO | 2024-01-03 11:00 | Alice | — | Start |
| PO-002 | Goods Receipt | 2024-01-04 08:00 | Dave | 21h | Deviation |
| PO-002 | Approve PO | 2024-01-05 15:00 | Bob | 1d 7h | Late |
| PO-002 | Invoice Received | 2024-01-07 10:00 | System | 1d 19h | Price Diff |
| PO-002 | Payment Cleared | 2024-01-15 12:00 | Finance | 8d 2h | Late Pay |
Assembling the Data Model
Before process mining runs, raw tables from source systems must be assembled into a structured data model that maps cases to activities to timestamps.
P2P Data Model — Table Structure
CASE TABLE
🔑 PO_Number (PK)
Vendor_ID
Company_Code
Purchase_Org
PO_Date
Net_Value
Currency
Document_Type
Vendor_ID
Company_Code
Purchase_Org
PO_Date
Net_Value
Currency
Document_Type
ACTIVITY TABLE
🔗 PO_Number (FK)
Activity_Name
Timestamp
User_ID
System_Module
Change_Type
Field_Name
Old_Value · New_Value
Activity_Name
Timestamp
User_ID
System_Module
Change_Type
Field_Name
Old_Value · New_Value
ATTRIBUTE TABLES
Vendor Master
Material Master
Cost Center
User Roles
GL Accounts
Currency Rates
SLA Thresholds
Material Master
Cost Center
User Roles
GL Accounts
Currency Rates
SLA Thresholds
Data Extraction Sources (SAP P2P)
| SAP Table | Contains | Used For |
|---|---|---|
EKKO | PO Header data | Case attributes |
EKPO | PO Line items | Item-level analysis |
CDHDR / CDPOS | Change documents | Event log (activities + timestamps) |
MSEG | Goods movements | GR/GI events |
BKPF / BSEG | FI documents | Invoice & payment events |
LFA1 | Vendor master | Vendor enrichment |
Data Model → Process Graph
The process graph (Directly-Follows Graph) visualizes how activities connect. Edge thickness represents frequency; edge color represents throughput time.
P2P Process — Directly-Follows Graph
Conformance Check Results
Fitness Score
0.74
Precision
0.82
Generalization
0.88
Create PO → Approve PO — Compliant (10 cases, 100%)
Create PO → Goods Receipt — Deviation (5 cases, 25%) — GR before approval violates 3-way match
Approve PO → Reject → Create PO — Rework loop (5 cases, 23%) — avg +3.2 days cycle time
Goods Receipt → Invoice → Payment — Compliant (all cases)
Case-Centric vs Object-Centric Process Mining
Traditional process mining assigns every event to exactly one case. Object-centric PM allows one event to be related to multiple objects simultaneously — much closer to how real processes work.
Case-Centric View
Case-Centric Model — One case ID per process instance
| Case ID | Activity | Timestamp | Limitation |
|---|---|---|---|
| PO-001 | Approve PO | Jan 10 14:00 | Convergence issue |
| PO-001-LINE1 | Goods Receipt | Jan 12 10:00 | Flattened — duplicated events |
| PO-001-LINE2 | Goods Receipt | Jan 13 11:00 | PO-001 split into 2 artificial cases |
| PO-001-LINE1 | Invoice Cleared | Jan 15 09:00 | Approve PO event duplicated |
Problem: When a PO has multiple line items, Goods Receipts, and Invoices — the classic approach either duplicates events or splits cases artificially. Both distort the model.
When to Use Object-Centric?
Use OCPM when your process has many-to-many relationships — one PO can have multiple invoices, one invoice can span multiple POs. Classic examples: P2P (PO → GR → Invoice), O2C (Order → Delivery → Invoice → Payment), Logistics (Shipment → Multiple Packages → Multiple Routes).
Process Mining Techniques
Three primary technique families — each answering a different question about your process.
Process Discovery
Automatically construct a process model from an event log — no prior knowledge required. Algorithms: Alpha Miner, Heuristics Miner, Inductive Miner, Fuzzy Miner.
Conformance Checking
Compare actual execution against the reference model. Quantify fitness, precision, recall, and identify exactly where and how deviations occur.
Process Enhancement
Extend models with additional data — performance (time, cost), decisions (if/else branches), and resource perspectives (who does what, when).
Discovery Algorithm Comparison
| Algorithm | Best For | Handles Noise | Short Loops | Complexity |
|---|---|---|---|---|
| Alpha Miner | Clean, simple logs | No | No | Low |
| Heuristics Miner | Noisy, real-world logs | Yes | Partial | Medium |
| Inductive Miner | Guaranteed sound models | Partial | Yes | Medium |
| Fuzzy Miner | Complex, spaghetti processes | Yes | Yes | High |
Real-World Case Studies
How enterprises across industries use process mining to drive measurable outcomes.
P2P — Manufacturing
BMW Group
Identified ₹500M+ in payment discounts being left on the table due to late invoice processing. Celonis revealed that 38% of invoices were parked manually without auto-posting — a SAP configuration gap invisible to auditors.
€170M
Savings captured
80%
Touchless invoices
6 wks
To first insight
O2C — Pharma
Pfizer
Order-to-cash cycle exceeded 35 days on average. Process mining discovered that 22% of orders were blocked for credit checks that were ultimately approved — adding 4+ days per case with zero risk reduction benefit.
-8d
DSO reduction
$1.2B
Working capital freed
22%
False blocks removed
ITSM — Telecom
Vodafone
IT incident resolution averaged 14 hours. Mining ServiceNow data revealed that 60% of P2 incidents were reassigned more than twice before resolution — a skills-routing failure, not a volume problem.
-40%
Resolution time
95%
First-assign rate
€18M
SLA penalty savings
Logistics — Retail
Siemens
Supply chain delays were costing €2M/month in expediting fees. Object-centric process mining mapped the PO → Delivery → GR → Invoice flow across 12 plants simultaneously, revealing bottlenecks at the port-of-entry GR step.
€24M
Annual savings
-31%
Expediting costs
12
Plants analyzed