Krumeich, Werth & Loos 2015 — Prescriptive Control of Business Processes

Springer Business & Information Systems Engineering (BISE) research paper from the Institute for Information Systems (IWi) at DFKI Saarland University. Published online 7 December 2015 (assigned to volume year 2016 — both dates appear in citations). One of the earliest articulations of prescriptive process control in BPM, predating most of the literature catalogued by Kubrak et al. 2022's SLR and contemporaneous with Gröger et al. 2014.

What the paper does

A design-science research artifact (Hevner) for a system architecture that takes predictive process monitoring beyond predicting the future to actively controlling it via KPI-driven optimisation. Three lenses make the contribution distinctive:

Event-Driven framing — re-uses Complex Event Processing (CEP) from financial-services & manufacturing operations as the technical substrate. Coins ED-BPM and EDPA (Event-Driven Predictive Analytics) as the umbrella concepts.
Industry verticalisation — explicitly targets the process manufacturing industry, particularly steel, with its characteristic divergent/analytical material flows, production cycles, non-linear outputs, and customer-specific end products.
Borrowed conceptual heuristic from Condition-Based Maintenance (CBM) — section 2.4 develops CBM as a methodological analogy for prescriptive process control.

The four-component concept (Figure 4 in the paper)

#	Component	What it does
1	Process blueprints with complex event patterns	Process models extended to embed CEP queries at decision/control points.
2	Pattern induction + detection	Cluster mining for significant event patterns; CEP-engine runtime detection in event streams.
3	Event-based process forecasting	Bayesian network (probabilistic) prediction of remaining process behaviour and KPIs, conditioned on detected complex events.
4	KPI-based prescriptive control	Optimisation function o(a, b, c) over time, quality, resource-consumption KPIs; selects the (most-optimal) onward process branch.

The enterprise architecture (Figure 5)

Five technical layers, top to bottom:

Dashboard with KPI Controller — managers tune KPI weights.
Process Engine Layer — Process Models · Process Instances · Simulation Engine · Optimisation Engine.
Prediction Layer — Event-based prediction models (rule induction · neural net · etc.).
Complex Event Processing Layer — EPA₁..EPAₙ · Event Processing Engine · Event Model · Event Rules.

Event stream flows: pre-processed events → CEP detection → predicted complex events → prediction layer → process engine → control feedback.

Seven requirements derived from the case study

Sensor instrumentation of physical processes.
CEP engine integrating heterogeneous event streams.
Event models + event rules specifying detection logic.
Intra-process cross-event mining to discover predictive complex-event patterns.
Embedding of prediction targets (cycle time, branch likelihoods, exception probabilities) directly into process models.
Cross-correlation of contextual (sensor) and intra-process data.
KPI-driven optimisation feedback loop from prediction outputs to process engine.

Position in the wiki's prescriptive-process-monitoring lineage

This paper is an early ancestor of the prescriptive-process-monitoring literature later catalogued in Kubrak et al. 2022's SLR:

Year	Source	Contribution
2014	sources/2014-groger-prescriptive-analytics-bpo	Concept of "recommendation-based BPO" using process warehouse + association-rule mining.
2015	This paper (Krumeich et al.)	CEP-based prescriptive control architecture for process manufacturing; 4-component concept; 7 requirements.
2022	sources/2022-kubrak-prescriptive-ppm-slr	SLR consolidating the field; covers 36 included papers from 2014–2021.

Limitations

Single case study — German steel manufacturer. Generalisability to discrete-manufacturing / service-industries explicitly flagged as future work.
Revelatory rather than evaluative — design-science evaluation is conceptual (descriptive utility) + single-case demonstration. No quantitative end-to-end deployment.
Bayesian-network prediction model — chosen for tractability + interpretability; not benchmarked against contemporary deep-learning PPM alternatives (Evermann 2017, Tax 2017 had not yet been published at the time).
Optimisation function kept simple (weighted sum); does not address multi-objective Pareto frontiers, stochastic dominance, or risk-aware KPI weighting.

Connections

Concepts:

concepts/prescriptive-process-monitoring — early canonical reference.
concepts/event-driven-bpm — introduces ED-BPM and EDPA in the wiki.
concepts/predictive-process-monitoring — embedded prediction component is PPM-as-subroutine inside PrPM.
concepts/business-process-simulation — simulation engine in the process engine layer.

Methods:

methods/process-discovery-methods — complex-event-pattern mining as a discovery variant.

Sources cited that exist in the wiki:

sources/2014-groger-prescriptive-analytics-bpo — concurrent PrPM-vision paper.
sources/2018-dumas-fundamentals-of-bpm — BPM lifecycle foundation.

Syntheses:

syntheses/prescriptive-process-monitoring-lineage — anchor source for the lineage.
syntheses/ppm-landscape — referenced as PPM→PrPM bridge.