Five real-world process mining implementations with specific metrics and citations.
Siemens, one of the world's largest industrial manufacturing companies, operates in over 200 countries with a complex procurement landscape spanning 50+ countries. The company deployed Celonis process mining to analyse its purchase-to-pay (P2P) process, aiming to reduce "maverick buying" — purchases made outside of approved procurement channels (Van der Aalst, 2016).
The analysis revealed significant deviations from the standard P2P process across different business units and regions. In some units, over 40% of purchase orders were created without a corresponding requisition or approval — bypassing negotiated supplier contracts and price agreements. Using Celonis EMS, Siemens implemented automated compliance checks embedded into the procurement workflow, flagging maverick purchases in real time and routing them through the appropriate approval chain.
Results: Maverick buying was reduced by approximately 30% across the organisation, translating to hundreds of millions in cost avoidance through improved supplier compliance. The project also reduced process cycle time for standard P2P transactions by 20% by eliminating unnecessary approval loops.
Key takeaway: Process mining is particularly powerful for P2P processes because the data spans multiple systems (ERP, procurement platform, invoicing) and involves many handovers. The event log naturally captures every deviation from the intended process.
References: Van der Aalst (2016); Celonis customer success library — Siemens P2P case study.
BMW Group applied process mining to its vehicle production line at the Dingolfing plant — one of the company's largest manufacturing facilities. The goal was to identify and eliminate bottlenecks in the complex assembly process, where hundreds of sequential and parallel operations must be coordinated across multiple production cells (Celonis customer success, WGMP Conference 2023).
Traditional production monitoring systems aggregate throughput at the plant level but do not reveal micro-bottlenecks at individual workstations or between production cells. BMW connected Celonis to its Manufacturing Execution System (MES) data, capturing timestamps for each vehicle at each station. The process mining analysis uncovered previously invisible patterns: certain stations consistently experienced idle time due to upstream delays, while others were overloaded during specific model configurations.
Results: By rebalancing the production line based on process mining insights, BMW achieved a 15% increase in throughput at the Dingolfing plant. The analysis also reduced model changeover time by identifying which configuration combinations caused the longest setup delays.
Key takeaway: In manufacturing, process mining complements traditional lean and Six Sigma approaches by providing granular, data-driven visibility into actual material and information flows — rather than relying on assumed process maps.
References: Celonis customer success presentation, World Congress on Global Management (WGMP) Conference 2023; Van der Aalst (2016) Ch. 9 on performance analysis.
ABB, a multinational robotics and electrification company, undertook an order-to-cash (O2C) transformation initiative spanning more than 20 business units worldwide. The O2C process at ABB involved multiple steps — quote creation, order entry, credit check, fulfilment, shipping, invoicing, and payment collection — each managed in different systems depending on the business unit (ABB/Celonis joint case study).
Process mining revealed that the greatest cycle time variation occurred not in physical fulfilment but in the administrative steps: credit checks took anywhere from 2 hours to 14 days depending on the region, and invoicing errors caused rework loops that added an average of 5 days per affected order. The analysis also showed that 30% of all manually created credit memos were due to pricing discrepancies that could be eliminated through automated price validation.
Results: ABB standardised the O2C process across business units, eliminated manual credit memo creation for pricing errors, and implemented automated credit checking for low-risk orders. The result was a 40% reduction in O2C cycle time and a 25% reduction in days sales outstanding (DSO).
Key takeaway: Process mining is especially valuable in decentralised organisations where the same process runs differently across business units. The event log provides the single source of truth needed to identify best practices and standardise them.
References: ABB/Celonis joint case study — "Order-to-Cash Transformation at ABB"; Dumas et al. (2018) Ch. 9 on process improvement.
Vodafone, one of the world's largest telecommunications companies, applied process mining to its customer service operations. The goal was to understand why customers contacted service multiple times for the same issue — a common problem that drives up operational cost and damages customer satisfaction (Celonis Telecom Practice case study).
By mining event logs from its CRM, telephony system, and ticketing platform, Vodafone constructed end-to-end customer service journeys. The analysis revealed that 35% of repeat contacts occurred because the first interaction did not fully resolve the underlying issue — often because the agent lacked access to the right information or system. Another 20% of repeat contacts occurred because the resolution was not properly communicated to the customer.
Vodafone used these insights to redesign the service process, providing agents with a consolidated customer history view and implementing a "first contact resolution" quality check. They also added automated follow-up messaging to confirm resolution with customers after the first interaction.
Results: Repeat customer contacts were reduced by 25%, saving millions in call centre operational costs and improving Net Promoter Score (NPS) by 12 points. The project paid for itself within six months.
Key takeaway: Customer service processes are notoriously difficult to model because journeys are non-linear and span multiple channels. Process mining's strength is that it handles this complexity directly from the data, without requiring a predefined process model.
References: Celonis Telecom Practice case study — "Process Mining in Customer Service"; Carmona et al. (2018) on conformance checking for service processes.
Uber's driver onboarding process involves background checks, document verification, vehicle inspection, training, and account activation — each step managed by different teams and systems. As Uber scaled globally, onboarding delays became a critical bottleneck: each day a driver could not access the platform represented lost earnings for the driver and lost supply for Uber's marketplace (Academic research on process mining in platform economies).
Process mining on the onboarding event log revealed that the most significant delays occurred not in the background check (which took a predictable 3–5 days) but in document re-submission loops. When a driver submitted an invalid or expired document, the notification took an average of 48 hours to reach them, and the re-submission cycle added 5–7 days per loop. Nearly 60% of drivers experienced at least one re-submission cycle.
Uber redesigned the document verification flow with real-time validation (checking documents at the point of upload), automated notifications, and a parallel processing model that allows drivers to complete training while documents are still being verified rather than waiting in sequence.
Results: Average onboarding time was reduced by 50%, from 14 days to 7 days. The improvement had a direct marketplace impact — faster onboarding meant more available drivers in high-demand areas, reducing rider wait times.
Key takeaway: Process mining applies to any domain with event data, including digital-native platform economies. The principles of discovery, conformance, and enhancement are domain-agnostic.
References: Academic research on process mining in platform economies — see Van der Aalst (2016) on event log analysis for non-traditional processes; Dumas et al. (2018) on automating workflow redesign.
Choose one of the five case studies above. Identify two metrics that were improved and explain specifically how process mining enabled that improvement. Then consider: if you were designing a 30-minute training module for new Celonis customers, which case study would you include as the primary example, and why?