---
title: "Agentic Business Process Management: Practitioner Perspectives on Agent Governance in Business Processes"
type: source
tags: [apm, governance, practitioner-study, qualitative, empirical, industry, adoption]
authors: [Vu Hoang; Klievtsova Nataliia; Leopold Henrik; Rinderle-Ma Stefanie; Kampik Timotheus]
year: 2025
venue: "Proceedings of the 23rd International Conference on Business Process Management (BPM 2025), Springer, pp. 29–43. arXiv:2504.03693v2 [cs.SE]"
kind: paper
raw_path: "raw/ABPS/Practitioner Perspectives on Agent Governance in Business Processes.pdf"
key_claims:
  - Defines agentic Business Process Management (ABPM) as the governed deployment of autonomous software agents to achieve business-process goals, integrating agent-based abstractions into process-oriented design and analysis.
  - Positions a historical "Agents for BPM" (agents as BPM support) vs "BPM for Agents" (BPM as discipline governing agents) distinction — the latter is the novel contribution of ABPM.
  - Empirical basis: 22 semi-structured interviews with BPM practitioners across 5 countries (Germany 12, India 7, Australia/Austria/Malaysia 1 each) and 8 roles (Consultant 9, Application Owner 5, CoE Expert 3, IT Architect, Process Analyst, Consulting Manager, Senior AI Prog Manager); qualitative content analysis (Mayring) with deductive–inductive coding.
  - No interviewee had actual practical experience with agentic AI — all assessments are expectation-based, often against the backdrop of RPA.
  - Practitioners anticipate six benefits (enhanced efficiency, reduced manual work, improved data quality, error reduction, cost savings, compliance improvement, scalability, risk reduction) and five risks (bias in AI models, over-reliance on automation, job displacement, lack of clarity, cybersecurity threats, governance issues, unauthorized decisions, data breaches).
  - Six adoption recommendations: clear business goals, legal/ethical guardrails, human-agent collaboration, customised agent behaviour, risk management, safe integration with fallback options.
  - Four BPM-specific implications: human-agent balance, human-role redefinition (peer vs supervisor), evolution beyond static process structures, new agent-aware performance metrics.
  - Responsibility for agent failures is predominantly seen as *shared* across organisations, developers, users, application owners, AI developers, and process managers — with organisations and "shared responsibility" the two most-cited categories.
created: 2025-04-01
updated: 2026-04-21
---

# Vu, Klievtsova, Leopold, Rinderle-Ma & Kampik 2025 — Practitioner Perspectives on Agent Governance in Business Processes

Empirical paper published at **BPM 2025** (23rd International Conference on Business Process Management, pp. 29–43; arXiv preprint `2504.03693v2`, 3 July 2025). Authors span **SAP Berlin** (Vu, Kampik), **TU Munich** (Klievtsova, Rinderle-Ma), **Kühne Logistics University Hamburg** (Leopold), and **Umeå University** (Kampik). Rare in the agentic-BPM corpus for being a **practitioner-facing qualitative empirical study** rather than a theoretical / architectural paper.

## Summary

The paper addresses a gap noted implicitly by the [[sources/2026-calvanese-agentic-bpm-manifesto|APM Manifesto]]: the theoretical agentic-BPM literature (APM, ABPMS, A-BPMS keynote) is largely detached from current industry practice. The authors conduct 22 semi-structured interviews with BPM practitioners to investigate how organisations can effectively govern AI agents in business processes. The research question is phrased as: *"How can BPM support the responsible integration of AI agents into organizations, ensuring both effective business outcomes and social accountability?"*

§2 traces a **historical timeline** of agent concepts in BPM, organised along two streams:

- **Agents for BPM** — agents as means to support BPM tasks (Agent-Oriented Programming '90s, RPA '10s, ABPMS '20s, GenAI agents mid-'20s).
- **BPM for Agents** — BPM as a discipline that governs agent-based systems (Normative MAS, Process Choreographies, Agent System Mining, and finally Agentic BPM mid-'20s).

On this basis the authors introduce **Definition 1 (Agentic Business Process Management)**: *"ABPM describes (i) the deployment and execution of autonomous software agents to achieve business process goals and (ii) the application of agent-based abstractions for the process-oriented design and analysis of autonomous software agents."* The definition is deliberately technology-agnostic and spans the BPM lifecycle's deployment/execution and design/analysis phases.

§3 describes the qualitative methodology: semi-structured interviews of 60–90 minutes, transcription, familiarisation, then deductive–inductive coding per Mayring (2019). 22 participants across Professional Services (10), Manufacturing (5), Telecommunications (3), and smaller sectors; Consultant (9), Application Owner (5), Center of Excellence Expert (3), and other roles. Notably **none reported prior hands-on experience with agentic AI** — answers reflect expectations informed by existing RPA/automation experience.

§4 reports findings under six themes:

1. **Understanding.** 10 of 22 were familiar with "agentic AI"; views ranged from "an evolution of RPA" to "digital assistant" to "orchestration layer".
2. **Benefits.** Enhanced efficiency (6), reduced manual work (6), improved data quality (5), error reduction (4), cost savings (4), compliance improvement (3), scalability (3), risk reduction (2).
3. **Risks.** Bias in AI models (5), over-reliance on automation (4), job displacement (4), lack of clarity (3), cybersecurity threats (3), governance issues (3), unauthorized decisions (3), data breaches (3).
4. **Use cases.** Task Automation (7), Process Monitoring (5), Customer Service Automation (4), Predictive Analytics (4), Quality Control (3), Data Structuring (3), Financial Risk (3), Supply-Chain Optimization (3), Master Data Maintenance (2).
5. **Requirements.** Governance Structures (8), Process Integration (7), Compliance with Guidelines (6), Employee Training (5), Cost Management (4), Preconfigured Use Cases (3), Early Testing (3), Access Control (3).
6. **Autonomy and adaptability.** Configurable autonomy — flexibility in low-risk areas, restriction in high-risk scenarios (system changes, external communications); start-simple-then-grow adoption.
7. **Human involvement and responsibility.** Agents as decision-support, not decision-makers, in complex scenarios. **Responsibility** most often attributed to *organisations* (10) and as *shared responsibility* (10), with AI developers (7), users (7), application owners (4), process managers (3), dedicated AI department (2) trailing.

§5 distils **six adoption recommendations**: (1) Business context — clear goals, expected benefits, cost assessment; (2) Guardrails — legal/ethical/organisational boundaries; (3) Human-agent collaboration — defined roles, intervention points; (4) Customization — tailor autonomy level, oversight, documentation, reasoning transparency; (5) Risk management — monitor performance, prevent undesired adaptation/evolution; (6) Adoption — seamless integration, fallback mechanisms.

Four **BPM-specific actions** are proposed: (1) **Human-agent balance** in workflow design; (2) **Human involvement** — peers vs supervisors; (3) **Process structures** evolving beyond static models toward real-time-adjustable structures; (4) **Performance metrics** — agent-impact metrics on efficiency, decision quality, organisational outcomes.

## Key claims

- Agentic BPM (ABPM) is the governed deployment of autonomous software agents to achieve business-process goals, applying agent-based abstractions to process design/analysis.
- The "Agents for BPM" vs "BPM for Agents" distinction frames the novelty of ABPM: BPM as a discipline that governs agent systems rather than just uses them.
- Practitioners anticipate agentic AI benefits and risks largely through the lens of RPA, with no participant reporting hands-on agentic-AI experience.
- Configurable autonomy with risk-proportional restrictions, not all-or-nothing autonomy, is the practitioner consensus.
- Responsibility for agent failures is perceived as shared across organisations, developers, users, and domain owners — not locatable in a single role.
- Static BPMN-style models are inadequate; process structures must evolve to support real-time agent-driven adjustments.
- Agent-specific performance metrics are needed alongside traditional BPM KPIs.
- Cultural acceptance of autonomy-to-agent transfer is a first-order adoption barrier — not just a technical issue.

## Framing distinctions

- **Agents for BPM vs BPM for Agents.** The historical "timeline" figure is the key conceptual device: the paper positions ABPM squarely in the "BPM for Agents" stream.
- **Configurable autonomy ≠ full autonomy.** Practitioners uniformly reject across-the-board autonomy; they envisage a gradient scaled to task risk.
- **Agent as peer vs supervisor.** Two distinct modes of human-agent collaboration surface in the data — a question the paper flags for future BPM research.
- **Benefits vs requirements are not symmetric.** The most-cited benefits (efficiency, data quality) differ structurally from the most-cited requirements (governance, integration, compliance) — adoption is gated on governance, not on capability belief.
- **Decision-support vs decision-maker.** Practitioners pull agents decisively toward decision-support in complex scenarios, preserving human decision authority.

## Positioning vs related work in this wiki

- **Companion to [[sources/2026-calvanese-agentic-bpm-manifesto|APM Manifesto]] and [[sources/2026-dumas-agentic-bpms-pyramid|A-BPMS keynote (Dumas et al. 2026)]].** Where those papers are theoretical/architectural, this is the corresponding **empirical practitioner study**. The Dumas et al. 2026 keynote cites this paper as ref [12].
- **Contrasts with [[sources/2023-dumas-ai-augmented-bpms|ABPMS Manifesto (2023)]]:** same agentic territory, but ABPMS is designer-oriented while this paper is practitioner-oriented. Cites ABPMS as ref [6] positioning ABPM as "BPM for Agents" continuation.
- **Connects to [[concepts/ai-adoption]]** — adds a BPM-specific facet to the broader AI-adoption evidence base ([[sources/2025-korst-wharton-gen-ai-enterprise-adoption]], [[sources/2025-handa-which-economic-tasks-ai]]). Unlike those, this paper captures BPM-specialist perceptions rather than cross-sector knowledge-worker usage.
- **Complements [[sources/2024-kampik-large-process-models]]** (Kampik's other APM/ABPM contribution — the "Large Process Models" vision) and [[sources/2025-calvanese-autonomy-business-process-execution]] (framed-autonomy paper).
- **Explicitly references "Process Autonomization" (Janiesch, Kowalkiewicz & Rosemann 2025)** and subject-oriented BPM ([[concepts/declarative-process-modelling]] lineage).
- **Sibling to [[sources/2025-fettke-explainable-autonomous-business-processes]]** (the XABPs paper) — both 2025 BPM-community outputs shaping the empirical/theoretical scaffolding for the APM Manifesto.

## Connections

**Concepts (existing):** [[concepts/agentic-bpm]] · [[concepts/framed-autonomy]] · [[concepts/self-modification]] · [[concepts/explainability-apm]] · [[concepts/conversational-actionability]] · [[concepts/perceive-reason-act]] · [[concepts/ai-adoption]] · [[concepts/abps-autonomy-levels]] · [[concepts/behavioral-variability]] · [[concepts/process-architecture]] · [[concepts/bpm-lifecycle]]

**Entities:** [[entities/timotheus-kampik]] · [[entities/stefanie-rinderle-ma]] · [[entities/henrik-leopold]] · [[entities/hoang-vu]] · [[entities/nataliia-klievtsova]]

**Related sources:** [[sources/2026-calvanese-agentic-bpm-manifesto]] · [[sources/2026-dumas-agentic-bpms-pyramid]] (cites this paper, ref [12]) · [[sources/2023-dumas-ai-augmented-bpms]] · [[sources/2025-calvanese-autonomy-business-process-execution]] · [[sources/2024-kampik-large-process-models]] · [[sources/2025-fettke-explainable-autonomous-business-processes]] (sibling workshop output) · [[sources/2025-korst-wharton-gen-ai-enterprise-adoption]] (cross-sector adoption comparator) · [[sources/2024-xu-the-agent-company-benchmark]] (agent-capability comparator)

**Syntheses:** [[syntheses/abpms-to-apm-evolution]] · [[syntheses/apm-manifesto-core-messages]] · [[syntheses/apm-business-themes]] · [[syntheses/llm-bpm-reading-list]]

## Open questions raised by the source

- How to quantitatively validate the qualitative findings across industries (sample was 5 countries, Germany-heavy)?
- How do practitioner expectations evolve once hands-on agentic-AI experience accumulates?
- How should BPM governance frameworks differ structurally from RPA governance (a specific BPM-research ask)?
- What agent-specific performance metrics best complement traditional BPM KPIs?
- How to operationalise "configurable autonomy" in BPMN-style notations lacking frame/guard-rail constructs?