Event-Driven Business Process Management (ED-BPM)

A BPM paradigm in which business processes are monitored and controlled in real time on the basis of complex event patterns detected across heterogeneous event streams — typically sensor data, transactional records, and process-execution events — using Complex Event Processing (CEP) as the technical substrate. ED-BPM connects BPM's classical post-hoc analytics to operational sense-and-respond architectures.

The term was popularised in the German BPM research community in the early 2010s; Krumeich, Werth & Loos 2015 develop the paradigm into a full system architecture for prescriptive process control in process manufacturing.

Conceptual layers

The ED-BPM paradigm is typically organised around three event-abstraction layers (Krumeich et al. Fig. 1):

Layer	Examples	Quantity	Business impact per event
Complex events	Combinations of multiple aggregated events indicating critical situations	Few	High
Aggregated events	Enterprise-system events, event correlations	Many	Medium
Simple events	Single sensor signals, atomic process-system records	Very many	Low (individually)

A CEP engine continuously detects complex event patterns from streams of simple/aggregated events and triggers downstream actions — typically in a connected process engine, prediction layer, or notification system.

Event-Driven Predictive Analytics (EDPA)

A subcomponent of ED-BPM that extends real-time CEP detection with predictive analytics: forecasting complex events before they occur on the basis of preceding simple/aggregated events. Coined and developed by Krumeich et al. (citing Janiesch et al. 2012 and Redlich & Gilani 2012 as precursors). EDPA is what makes prescriptive process control feasible — the time gap between event prediction and proactive action determines business value (Krumeich Fig. 2).

Position within prescriptive process monitoring

ED-BPM and CEP provide one architectural realisation of prescriptive process monitoring (PrPM), distinct from the data-warehouse / association-rule approach in Gröger et al. 2014 and the data-stream / classifier approach in later work. The three approaches differ in:

Latency — CEP/EDPA operates on streams (sub-second); data-warehouse approaches operate on batch windows (hours/days).
Data substrate — CEP/EDPA fuses sensor + transactional + process events; data-warehouse approaches assume cleaned, aggregated process data.
Prediction model — CEP/EDPA leans on rule-induction + Bayesian networks; later approaches use deep learning.
Industry fit — ED-BPM was originally motivated by process manufacturing (steel, chemicals, pharmaceuticals) where high-frequency sensor data + non-linear material flows make CEP a natural fit. Less obviously useful for purely transactional white-collar processes.

Connections to APM

Recommend role — CEP-detected complex event patterns can drive APM-agent recommendation in real time.
Self-modification — ED-BPM architectures already include feedback loops from prediction outputs to process-engine control; the self-modifying ABPS research line extends this with adaptation logic.
Process-awareness — ED-BPM's emphasis on intra-process cross-event mining is one form of process-awareness an APM agent needs.

concepts/prescriptive-process-monitoring · concepts/predictive-process-monitoring · concepts/agentic-bpm · concepts/self-modification · concepts/business-process-simulation · sources/2015-krumeich-prescriptive-control-business-processes · sources/2014-groger-prescriptive-analytics-bpo · sources/2022-kubrak-prescriptive-ppm-slr