Most Companies Buy Process Mining Software.
Few Build a Process Mining Program.
Large enterprises generate an enormous volume of transactional evidence about how work moves. ERP systems, procurement platforms, finance modules, order management systems, and supply chain tools all contain fragments of process behavior. Those fragments rarely produce a shared operating picture.
Finance may understand the invoice lifecycle. Procurement may understand the purchase order path. Supply chain may understand fulfillment delays. Each function may be looking at accurate data, and the organization may still lack a common view of the end-to-end process. A Process Mining Center of Excellence exists to create that common view and turn it into disciplined action.
In programs I have built and led across healthcare, public-sector platforms, and large-scale commerce operations, the industries differed but the operating challenge was consistent. Large organizations usually have effort, meetings, dashboards, and transformation portfolios.
What they often lack is a governed mechanism for proving where process inefficiency accumulates, deciding which opportunities matter most, and measuring whether interventions produced real improvement. The technology is the execution surface. The operating model determines whether it delivers.
Charter and Decision Rights
The first artifact the CoE must produce is its own charter: why the capability exists, which enterprise processes are in scope, who owns the shared process model, how candidates will be prioritized, and how findings will move into execution.
The charter should make clear where CoE responsibility ends and business unit responsibility begins. The CoE owns the analytical foundation, the methodology, the shared data model standards, the KPI library, and the executive-ready findings. Business units own operating decisions, remediation execution, target commitments, and confirmation of realized value.
The CoE must also publish a decision rights matrix early. The CoE owns modeling standards, KPI definitions, conformance logic, and executive reporting standards. Business functions own policy decisions, operational redesign, and realized value confirmation. IT owns source access, integration design, and pipeline reliability. Enterprise architecture owns repository standards and BPMN conventions. Clear decision rights reduce confusion when findings become politically sensitive, and process evidence often challenges established assumptions.
Roles and the Business Relay Network
Named roles must be in place before the backlog grows. The CoE leader owns the capability roadmap, governance model, prioritization process, and executive narrative. Analysts build analyses, define KPIs, interpret variants, and translate process evidence into business findings. Data engineers own extraction, event logs, object relationships, and pipeline reliability. Process architects align mined baselines with BPMN 2.0 models and target-state designs. Transformation program managers move opportunities from diagnostic packages into delivery plans and benefits realization.
Business relays are the most underinvested component of most programs. Process mining can surface a deviation, a loop, or an unusual variant. A relay can explain whether that pattern represents waste, policy, regulatory necessity, supplier behavior, or system constraints. Reliable diagnosis requires both mined evidence and operational interpretation, and the operating model must sustain that relationship deliberately.
Why Object-Centric Process Mining Changes the Analytical Baseline
Traditional process mining traces a single case identifier through an event log. This works when a process is one-to-one: one object, one flow, one outcome. Enterprise processes carry far more complexity. A single purchase order may reference dozens of invoices. A sales order may spawn multiple fulfillment legs with independent scheduling, warehousing, and logistics events. Forcing this reality into a single-case model produces flattened process graphs, distorted cycle times, and misidentified automation candidates.
Object-Centric Process Mining, or OCPM, models multiple object types simultaneously and captures the relationships between them. In Celonis, the Object-Centric Data Model, or OCDM, is the architectural foundation on which every analysis is built. The resulting process graph reflects how work actually flows across object boundaries. Standing up this capability requires deliberate architectural decisions about which objects to include, how to define their relationships, and how to govern model extensions as new processes are onboarded. Getting those decisions right is a governance discipline, and it determines the analytical quality of everything built on top.
Data Pipeline Architecture and Model Governance
Analytical credibility depends entirely on data pipeline quality. A process graph built on incomplete event logs or misaligned timestamps will produce findings that experienced process owners will correctly challenge.
In enterprises running Oracle Fusion Cloud, the primary event log sources span Procurement, Payables, Receivables, Order Management, Supply Chain, and General Ledger. Extraction can run through direct Celonis EMS connectors or through an intermediate data lake layer, commonly AWS or Azure. The intermediate layer makes the event log assembly logic visible, testable, and auditable, which matters when data lineage and audit readiness are program requirements.
The OCDM must be governed as an enterprise asset: documented modeling standards, a defined extension policy, a model catalogue describing each object type and its source, and change review gates. A revised timestamp definition can alter throughput reporting. A source-system enhancement can create the appearance of process improvement when the underlying process did not change. Version control, change logs, and model review boards are the governance mechanisms that protect analytical trust.
Candidate Prioritization
A formal prioritization framework must be in place before demand pressure builds. The framework I have used scores candidates across value potential, feasibility, and evidence confidence, structured around Quality, Cost, and Delivery, or QCD, criteria. Quality scoring assesses defects, rework, and compliance issues. Cost scoring assesses labor effort, working capital impact, and cost-to-serve. Delivery scoring assesses cycle time, SLA adherence, and fulfillment delay.
The CoE converts those scores into candidate heat maps. Candidates in the high-value, high-feasibility quadrant are near-term priorities. High-value, low-feasibility candidates go into a development backlog. Publishing this framework and using it visibly in every steering committee is how the program maintains analytical independence and keeps prioritization grounded in evidence and insulated from internal pressure.
KPIs, Baselines, and Target Models
Celonis uses PQL, Process Query Language, for defining KPIs, conformance checks, and analytical expressions. A mature program maintains a governed PQL-based KPI library organized by domain and analytical purpose: diagnostic KPIs for cycle time and rework, conformance KPIs for match rates and approval bypass frequency, and value KPIs for working capital impact and cost of rework.
Every candidate that enters the program gets a frozen baseline at intake. Without it, value realization claims twelve months later have no credible reference point. Once validated with business relays, the as-is baseline is exported as BPMN 2.0 into the enterprise architecture repository and placed under version control. The target process model then defines expected operating performance with measurable QCD outcomes. Process simulation in Celonis can test whether removing a step, automating an exception path, or standardizing a variant is likely to improve throughput before the organization commits.
Diagnostic Packages and the Investment Case
When the CoE completes an initial process analysis, the output is a structured diagnostic package. The opportunity inventory lists findings by type: cycle time accumulation points, conformance deviations, rework loops, automation candidates, and control gaps. Each finding includes frequency of occurrence, estimated QCD impact, and the data supporting the estimate.
The investment case is a handover document. It includes frozen baseline KPIs, target KPIs with the expected delta, the QCD value estimate with the calculation methodology explicit, the implementation approach, the named executing owner, and the value realization measurement plan. The moment a business unit leader commits to execution, the CoE shifts to monitor: tracking KPIs against the frozen baseline and updating the value scorecard when realization is confirmed.
The value scorecard is the program’s primary credibility instrument with executive leadership. Pipeline value is tracked separately from committed value, and committed value is tracked separately from realized value. Many transformation programs estimate value confidently and prove it rarely. The scorecard is the governance mechanism that closes that gap.
AI Integration
AI belongs in the program wherever it reduces the time between data and decision without introducing opacity that undermines analytical credibility. Guided analysis features that surface anomalies and suggest investigation paths reduce the analytical burden on CoE staff. Agent integrations that monitor KPIs and trigger automated or human-in-the-loop interventions require careful design of trigger logic, intervention protocol, and audit trail. Every agent-mediated action must be traceable, and maintaining that traceability is an operating model responsibility.
The relationship between the Celonis Process Intelligence Graph and any Enterprise Knowledge Graph should be defined explicitly before agents are deployed at scale. Leaving the governance of that context flow ambiguous creates analytical debt that is expensive to resolve.
The Maturity Progression
Programs that sustain themselves share a common arc. The diagnostic phase stands up the architecture, establishes the operating model, and completes initial analyses of the highest-priority candidates. The operational phase runs a continuous analysis cadence and updates the value scorecard with confirmed realized value. The institutional phase treats process mining baselines as the reference point for any significant process change and uses AI-enabled monitoring to reduce the manual effort required to maintain baseline coverage.
Most programs stall between diagnostic and operational: governance gaps, baseline discipline failures, and relay networks engaged for the initial analysis but never sustained. Avoiding that stall is a leadership challenge, and the operating model is the instrument for meeting it.
At maturity, a Process Mining Center of Excellence gives the enterprise a reliable memory of how work actually happens. It maintains observed baselines, identifies costly deviations, connects process behavior to business outcomes, and measures whether transformation delivered the value that was promised. The tools, Celonis OCPM, Oracle Fusion event logs, AWS data pipelines, PQL KPI libraries, Process Adherence Manager, BPMN 2.0 exports, and process simulation, each play a role.
The operating model determines whether those tools create durable business value.
Nabeil Sarhan, MBA, is a dynamic technology delivery manager with over 15 years of experience in tech, cybersecurity, and computing scalability. He excels in leading diverse teams and delivering enterprise-class systems across industries such as healthcare, finance, and retail. Nabeil’s passion for solution design, systems architecture, and performance optimization makes him a sought-after consultant. He holds degrees from Harvard, MIT, and Bryant University. Connect with Nabeil on LinkedIn