Prescriptive Process Monitoring (PrPM)

A family of runtime methods that recommend interventions during the execution of a business process case so as to optimise the process with respect to an explicit objective. Where PPM stops at a forecast, PrPM closes the loop to action: predict that a case will run late, then also tell the worker what to do to prevent it.

Canonical SLR: Kubrak, Milani, Nolte & Dumas 2022. Earliest BPM articulations: Gröger, Schwarz & Mitschang 2014 (rBPO on a process warehouse) and Krumeich, Werth & Loos 2015 (CEP / event-driven BPM architecture for the process manufacturing industry). Keynote vision: Dumas 2021 — Process Mining 2.0: From Insights to Action. Emerging LLM-based future direction: Padella, de Leoni & Dumas 2026 explicitly flag a prescriptive extension of LLM-PPM as next work.

Position in the analytics spectrum

PrPM occupies the prescriptive cell of the descriptive → predictive → prescriptive analytics pyramid. In van der Aalst's refined process-mining framework (cartography / auditing / navigation × offline / online), PrPM is Navigation × Online: the model is used to guide a live journey, not just audit a past one. concepts/operational-support (detect / predict / recommend) is the nearest precursor name van der Aalst used for the same cell.

PrPM is also the BPM-flavoured instance of what other disciplines call process control. Distinct from chemical-engineering process control (sources/2007-bao-lee-process-control-passivity) and statistical process control — same word, very different machinery.

The six-dimensional framework (Kubrak et al. 2022)

Every PrPM method can be characterised along:

Dimension	Typical options
Performance objective	Optimise case outcome (desired label probability) · Optimise efficiency (cycle/processing time, resource utilisation)
Performance metric	Cycle time, processing time, defect rate, revenue, deadline violations, cost
Intervention type	Control-flow (next activity, alarm, path change) · Resource (assign / reassign) · Data (rarely)
Modeling technique	Nearest-neighbour / similarity; decision tree; classical ML; LSTM; RL (online and offline); causal / uplift trees
Data inputs	Control-flow, resource, temporal, domain-specific
Intervention policy	when and to whom — see concepts/intervention-policy

Orthogonal axes on interventions:

Frequency — continuous (every step) vs discrete (only when triggered).
Purpose — optimizing (KPI-driven; may be correlation- or causality-based) vs guiding (similarity to historical traces).

Research gaps (Kubrak et al. 2022)

Method lineage (representative)

Design-time predecessor

Before runtime, data-driven interventions, BPM already had a design-time, expert-driven catalogue of process improvements: the concepts/bpr-heuristics (Reijers & Liman Mansar 2005). PrPM can be read as the runtime, per-case, data-driven successor to this catalogue. Both share the concepts/devils-quadrangle as evaluation scaffolding.

Connections to APM

The APM Manifesto (sources/2026-calvanese-agentic-bpm-manifesto) positions PrPM techniques as what an agent invokes when playing the Recommend role of concepts/conversational-actionability. PrPM also supplies the adaptation trigger signal feeding concepts/self-modification.

concepts/predictive-process-monitoring · concepts/llm-based-ppm · concepts/event-driven-bpm · concepts/intervention-policy · concepts/operational-support · concepts/process-mining-spectrum · concepts/causal-process-discovery · concepts/interventionist-theory-of-causation · concepts/bpr-heuristics · concepts/devils-quadrangle · syntheses/ppm-landscape · syntheses/prescriptive-process-monitoring-lineage