---
title: "Synthesis: The Prescriptive Process Monitoring Lineage (2014–2026)"
type: synthesis
tags: [prpm, prescriptive, lineage, synthesis, krumeich, groger, kubrak, ed-bpm, cep, llm-future]
sources:
  - "[[sources/2014-groger-prescriptive-analytics-bpo]]"
  - "[[sources/2015-krumeich-prescriptive-control-business-processes]]"
  - "[[sources/2022-kubrak-prescriptive-ppm-slr]]"
  - "[[sources/2021-dumas-process-mining-2-from-insights-to-action]]"
  - "[[sources/2026-padella-llm-features-ppm]]"
created: 2026-05-11
updated: 2026-05-11
---

# Synthesis: The Prescriptive Process Monitoring Lineage (2014–2026)

A consolidated timeline of how [[concepts/prescriptive-process-monitoring|PrPM]] crystallised as a distinct BPM sub-discipline, organised around four pivot points and three architectural traditions.

## The four pivots

| Year | Source | Pivot |
|---|---|---|
| **2014** | [[sources/2014-groger-prescriptive-analytics-bpo\|Gröger, Schwarz & Mitschang]] — *Recommendation-based BPO* | First explicit *recommendation-from-process-data* framing. Process warehouse + association-rule mining → operator recommendations. The "descriptive → predictive → **prescriptive**" pyramid is named. |
| **2015** | [[sources/2015-krumeich-prescriptive-control-business-processes\|Krumeich, Werth & Loos]] — *Prescriptive Control of Business Processes* | First **enterprise architecture** for PrPM. Coins [[concepts/event-driven-bpm\|ED-BPM]] + EDPA. 4-component concept + 5-layer system architecture + 7 requirements. German steel-industry case study. Built on Complex Event Processing rather than the data-warehouse tradition. |
| **2021** | [[sources/2021-dumas-process-mining-2-from-insights-to-action\|Dumas]] — *Process Mining 2.0: From Insights to Action* | Keynote that consolidates the vision: PM 2.0 = action-oriented PM, of which PrPM is the core technical instrument. Sets the stage for the SLR consolidation phase. |
| **2022** | [[sources/2022-kubrak-prescriptive-ppm-slr\|Kubrak, Milani, Nolte & Dumas]] — PrPM SLR | The field's canonical SLR. 36 included papers. 6-dimensional characterisation framework (performance objective × metric × intervention type × modeling technique × data inputs × intervention policy). Identifies 6 research gaps. |

## Three architectural traditions

The 2014 and 2015 papers establish two distinct PrPM traditions that subsequent literature elaborates rather than reconciles. The 2026 LLM-PPM line is opening a third.

### 1. Data-warehouse / classifier tradition

Originates with [[sources/2014-groger-prescriptive-analytics-bpo|Gröger et al. 2014]]. Architectural assumptions:

- Cleaned, aggregated process data in a warehouse.
- Periodic (batch-window) analytics rather than continuous streams.
- Association-rule mining or supervised classifiers driving recommendations.
- Operator-in-the-loop intervention.

Most of the literature in [[sources/2022-kubrak-prescriptive-ppm-slr|Kubrak's SLR]] inherits this tradition — even when the modeling technique changes to LSTM, RL, or causal trees.

### 2. CEP / event-driven tradition

Originates with [[sources/2015-krumeich-prescriptive-control-business-processes|Krumeich, Werth & Loos 2015]]. Architectural assumptions:

- Real-time event streams (sensor data, transactional events, sub-second latency).
- Complex Event Processing engine as substrate.
- Bayesian networks / rule induction for probabilistic prediction.
- *Control* feedback loop into process engine — closer to industrial automation than to managerial recommendation.

This tradition is conceptually closer to **process control** in chemical engineering and to manufacturing-MES integration than to traditional BPM tooling. Strongly motivated by *process manufacturing* (steel, chemicals) — less obviously useful for transactional white-collar processes.

### 3. LLM-based tradition (emerging, 2026)

[[sources/2026-padella-llm-features-ppm|Padella, de Leoni & Dumas 2026]] flag prescriptive extension of [[concepts/llm-based-ppm|LLM-PPM]] as future work. Distinct architectural assumptions:

- Pre-trained foundation model (Gemini / Claude / GPT class).
- In-context learning rather than supervised retraining.
- Recommendations as natural-language reasoning + structured action proposal.
- Particularly suited to *data-scarce* settings where the warehouse tradition requires data the organisation lacks.

No concrete LLM-PrPM paper exists yet. Open empirical question: can an LLM combine the warehouse tradition's interpretive flexibility with the CEP tradition's latency budget?

## Kubrak's 6-dimensional framework as a unifying vocabulary

Where the three traditions disagree on *infrastructure*, they overlap on the **decision content**. [[sources/2022-kubrak-prescriptive-ppm-slr|Kubrak et al. 2022's]] six dimensions provide a comparison vocabulary regardless of architectural tradition:

| Dimension | Krumeich 2015 (CEP) | Gröger 2014 (warehouse) | Padella 2026 future LLM-PrPM |
|---|---|---|---|
| Performance objective | KPI optimisation (time / quality / resources) | Process-instance outcome improvement | Both — context-dependent |
| Performance metric | Cycle time, quality units (QU), throughput | Various — case-level | Both KPIs in Padella et al. (Total Time + Activity Occurrence) |
| Intervention type | Control-flow (production-step choice) | Resource + process-flow recommendations | Open (paper does not specify) |
| Modeling technique | Bayesian network + rule induction over CEP | Association-rule mining over warehouse | LLM in-context learning |
| Data inputs | Sensor + transactional event streams | Process warehouse aggregates | Trace + global attributes + reasoning examples |
| Intervention policy | Continuous (every event) | Periodic (warehouse update cadence) | Per-query (LLM call) |

## Research gaps from Kubrak's SLR, still open

1. **In-vivo validation** — most evaluations are historical back-tests; field deployments remain rare.
2. **Intervention discovery** — no principled method to *mine* candidate interventions from logs; they are hand-specified.
3. **Causal policies** — accounting for cause-effect and second-order effects is the emerging frontier ([[concepts/causal-process-discovery]]).
4. **Explainability and feedback loops** between PrPM systems and end-users.
5. **Dimension coverage** — temporal metrics dominate; quality, compliance, revenue under-served.
6. **Terminology** — *proactive adaptation*, *next-step recommendation*, *next-best action*, *on-the-fly resource allocation* coexist; a common vocabulary is needed.

## Practical entry points

- For a manager wanting **recommendation-style PrPM today**: read [[sources/2014-groger-prescriptive-analytics-bpo|Gröger 2014]] for vision, [[sources/2022-kubrak-prescriptive-ppm-slr|Kubrak 2022]] for tool selection.
- For a manufacturing-operations engineer: read [[sources/2015-krumeich-prescriptive-control-business-processes|Krumeich 2015]] for architecture, then go to the underlying CEP literature (Etzion & Niblett 2011 not yet in `raw/`).
- For an APM-adjacent researcher: read [[sources/2021-dumas-process-mining-2-from-insights-to-action|Dumas 2021]] for the vision, then [[sources/2026-calvanese-agentic-bpm-manifesto|the APM Manifesto]] for how PrPM fits into the agent-centric paradigm.
- For someone building LLM-PrPM: there is no ingested precedent. Closest is [[sources/2026-padella-llm-features-ppm|Padella 2026]] for the predictive half + the open prescriptive-extension flag.

## Related

- Concept hub: [[concepts/prescriptive-process-monitoring]].
- Adjacent concept: [[concepts/event-driven-bpm]].
- LLM-PPM hub: [[concepts/llm-based-ppm]].
- Sibling synthesis: [[syntheses/ppm-landscape]] for the predictive layer beneath PrPM.
- Reading list: [[syntheses/llm-bpm-reading-list]].