FRID-AI | BPM 2026

Process Intelligence Meets Agentic AI

As organizations move beyond AI-assisted automation toward autonomous, goal-driven systems, the intersection of Process Intelligence and Agentic AI is redefining the future of BPM. BPM 2026 FRID-AI: Process Intelligence Meets Agentic AI explores how intelligent agents can leverage process knowledge, operational data, and organizational context to make informed decisions, orchestrate workflows and continuously optimize business operations.

Friday, October 2 - Program Schedule

BPM 2026 FRID-AI: Process Intelligence Meets Agentic AI

Time	Session	Speaker
9:00 – 9:30 a.m.	Opening
9:30 – 10:30 a.m.	Keynote	Marco Montali Free University of Bolzano (Italy) Title: Framing AI-augmented processes Session Chair: Fabrizio M. Maggi, Free University of Bolzano (Italy)
10:30 – 11:00 a.m.	Coffee Break
11:00 – 11:30 a.m.	Focus Talk	Timotheus Kampik SAP-Signavio / Umeå University (Sweden) Title: Agentic Business Process Management: Where We Are and Where to Go Session Chair: Andrea Marrella, Sapienza University of Rome (Italy)
11:30 a.m. – 12:30 p.m.	Scientific Panel	Topic: BPM in the age of Agentic AI and LLMs. What changes, what remains? Moderator: Andrea Marrella, Sapienza University of Rome (Italy) Participants: - M. Montali, Free University of Bolzano (Italy) - T. Kampik, SAP-Signavio / Umeå University (Sweden) - A. Gal, Technion - Israel Institute of Technology (Israel) - S. Rinderle-Ma, Technical University of Munich (Germany) - A. Metzger, University of Duisburg-Essen (Germany) Session Chair: Claudio Di Ciccio, Utrecht University (Netherlands)
12:30 – 1:45 p.m.	Lunch Break
1:45 – 2:15 p.m.	Invited Talk	Soren Frederiksen CTO of mindzie Title: Insight in Minutes: What an AI Agent Needs to Do Process-Aware Data Analysis Session Chair: Claudio Di Ciccio, Utrecht University (Netherlands)
2:15 – 2:45 p.m.	Invited Talk	Eric Roovers Head of Customer Value, ARIS Title: From Process Models to Enterprise Context: Why Agentic AI Needs a Process-Aware Knowledge Layer Session Chair: Claudio Di Ciccio, Utrecht University (Netherlands)
2:45 – 3:30 p.m.	Industrial Panel	Topic: The Missing Elements for Effective AI-Enabled BPM in Enterprises Moderator: Claudio Di Ciccio, Utrecht University (Netherlands) Participants: - Soren Frederiksen, mindzie, Toronto (Canada) - Jessica Nahulan, IBM Toronto (Canada) - Eric Roovers, ARIS (Netherlands) - Rudy Kuhn, Celonis (Germany)
3:30 – 4:00 p.m.	Break
4:00 – 5:00 p.m.	Keynote	Sheila McIlraith University of Toronto, Canada Title: Auditing, Monitoring, and Intervention for Compliance of Agentic AI Systems Session Chair: Arik Senderovich, York University, Toronto (Canada)
5:00 – 5:30 p.m.	Closing

Speakers

Marco Montali

Free University of Bolzano (Italy)

Marco Montali is Professor of Computer Engineering at the Faculty of Engineering of the Free University of Bozen–Bolzano, Italy, where he coordinates the BSc programme in Informatics and Management of Digital Business. He has made pioneering contributions at the intersection of artificial intelligence, process science and information systems engineering, with a particular focus on the modelling, analysis and mining of multi-perspective processes and multi-agent systems. A unifying theme of his research is the combination of model-driven and data-driven techniques to engineer safe and controllable information systems and agents, with particular attention to their evolving behaviour. He has authored or co-authored more than 300 publications and received ten best-paper awards and two test-of-time awards. His scientific service includes programme-chair roles at BPM 2018, RuleML+RR 2019, ICPM 2020 and CBI 2021, and general-chair roles at ICPM 2022, EDOC 2022 and AIxIA 2024. He will serve as Programme Chair of PETRI NETS 2027. He also served a three-year term on the Steering Committee of the IEEE Task Force on Process Mining. He is a EurAI Fellow and a corresponding member of the Accademia dei Georgofili. Beyond academia, he regularly engages with the public, policymakers and industry on artificial intelligence and its societal implications.

A long-standing challenge in process science is to balance flexibility and control: process actors should remain free to determine how best to act, while their actions must comply with process constraints. As AI becomes increasingly intertwined with process management, tackling this challenge is more important then ever. Framed autonomy has been consequently introduced as central concept in AI-augmented processes: process orchestration is partially delegated to one or more autonomous agents, which may pursue distinct goals, while remaining within a process frame, that is, within the behavioural boundaries imposed by the process. In this presentation, I examine the formal and algorithmic foundations of process framing. I argue that declarative formalisms based on (first-order) temporal logics provide natural and increasingly expressive languages for specifying frames over behaviour, time and data. I then discuss how such frames can be made operational through verification and synthesis, conformance checking, and monitoring. The analysis unfolds along two dimensions. The first concerns data: from control-flow constraints to data-aware specifications, and from case-centric traces to object-centric and relational event data. The second concerns agency: from centralised orchestration among cooperative actors to decentralised execution by multiple autonomous agents, potentially pursuing different goals. These extensions challenge established notions of process instance, state, conformance and responsibility.

Timotheus Kampik

SAP-Signavio / Umeå University (Sweden)

Timotheus Kampik is a Fellow of the Wallenberg AI, Autonomous Systems and Software Program and Assistant Professor at Umeå University, Sweden. He also serves as Principal Scientist for Process Intelligence at SAP, having worked in R&D and professional service roles at Signavio since 2012. His core research interests are business processes, multi-agent systems, and symbolic reasoning.

With the advent of current-generation agentic AI, BPM experiences a paradigm shift, from orchestration-orientation to agent-orientation. To support this shift, Agentic Business Process Management (ABPM) embraces the dichotomy between agents as natural entities with micro-level goals and processes as artificial abstractions for aligning work with organizational objectives. This talk gives an overview of ABPM, based on perspectives jointly developed by academics and industry experts from, e.g., IBM, Google, Meta, Salesforce and SAP, during a Dagstuhl seminar and in a position paper recently published in the Information Systems journal. It also discusses recent research results relating to ABPM, as well as a research roadmap towards enabling future-generation agent-oriented organizations.

Soren Frederiksen

CTO of mindzie

Soren Frederiksen is the CTO and Co-founder of mindzie, a process intelligence company whose platform helps organizations discover, understand, and improve how their processes actually run. He has spent his career building software at the meeting point of data and operations, with two successful company exits along the way. Based in the Toronto area, he works hands-on across process-aware data analysis, BPMN, and applied AI, and is currently focused on how AI can help people author process models they can trust.

Ask a large language model the same question about your event log twice and you can get two different answers. It will name a bottleneck that is not in the data and sound exactly as certain as when it is right. An analysis you cannot reproduce is not an analysis. This talk is what it takes to fix that. I will show the mindzie agent doing real process-aware data analysis on customer data: one way in to the log and no other, code it writes and runs itself, a method that makes it discover before it concludes, and every figure traceable back to what it read. It runs inside the customer's own box, on premise or in their own cloud, and on our models or the ones their company already licenses. Then the measured result. Twenty of these reports, root cause, SLA attainment, KPI audit, return on investment, each built end to end in five to eight minutes. The analysis has stopped being the bottleneck. A proof of concept that took six to eight weeks now takes two, and nearly all of that time now goes on interviewing people and understanding the business. I will finish on where this leads: a factory of those boxes, one to a project, and agents moving between them.

Eric Roovers

Head of Customer Value, ARIS

Eric Roovers is Vice President and Head of Customer Value at ARIS. He leads a global team of Value Architects who help organizations create sustainable value through Process Intelligence and Business Process Management. With over 20 years of international experience, Eric is a strategic leader guiding organizations in designing and scaling impactful transformation initiatives.

Large Language Models have demonstrated remarkable reasoning and communication capabilities, yet they continue to struggle in complex enterprise environments where decisions depend on organizational context, business rules, process dependencies, and cross-functional coordination. While recent research has focused on improving AI models themselves, a critical question remains: What enterprise context must be provided for AI agents to operate safely, reliably, and at scale? This talk explores the emerging role of process-aware enterprise knowledge layers as the foundation for agentic AI. We argue that business processes represent more than operational workflows; they encode organizational intent, governance structures, decision rights, compliance obligations, and the relationships between people, systems, data, and outcomes. Without access to this context, enterprise AI agents risk producing recommendations that are technically plausible but operationally incorrect. Drawing on real-world enterprise transformation initiatives, this presentation examines how process intelligence, governance models, and semantic business architectures can be combined to provide the grounding required for trustworthy AI. We will discuss how process models evolve from documentation artifacts into dynamic context systems that enable AI agents to understand objectives, constraints, responsibilities, and business impact. The talk concludes by proposing a research and practitioner agenda for Process Management in the age of Agentic AI, where BPM moves beyond process optimization and becomes a core mechanism for governing, explaining, and scaling intelligent enterprise systems.

Sheila McIlraith

University of Toronto, Canada

Sheila McIlraith is a Professor in the Department of Computer Science at the University of Toronto, a Canada CIFAR AI Chair (Vector Institute), and an Associate Director and Research Lead at the Schwartz Reisman Institute for Technology and Society. McIlraith’s research is in the area of AI sequential decision making, broadly construed, with a focus on human-compatible AI and AI Safety. McIlraith is a Fellow of the ACM and AAAI, a Schmidt Sciences AI2050 Senior Fellow, and is currently serving as a member of the Canadian AI Safety Institute (CAISI) Research Council. McIlraith has received a number of honours and recognitions for scholarly work and innovation in teaching.

Advanced AI systems, notably Large Language Models (LLMs), are increasingly being used as a core technology in the deployment of products and services---often in the form of agentic AI systems. This trend towards AI-enabled products and services presents a set of governance challenges that transcend AI governance efforts related to frontier AI models, requiring them to additionally support safe and lawful adoption of AI-enabled products and services, often in the form of jurisdiction-, sector-, or product-specific behavioural requirements. In this talk, we discuss one particular dimension of AI governance: how to monitor and audit AI-enabled products and services throughout the AI development lifecycle. Combining principles from formal methods with SoTA machine learning, we propose techniques that enable developers, and third-party evaluators, to perform offline auditing and online (runtime) monitoring of product-specific (temporally extended) behavioral constraints such as safety constraints, norms, rules and regulations with respect to black-box advanced AI systems, notably LLMs. We further provide practical techniques for predictive monitoring, such as sampling-based methods, and we introduce intervening monitors that act at runtime to preempt and potentially mitigate predicted violations. Experimental results show that by exploiting the formal syntax and semantics of Linear Temporal Logic (LTL), our proposed auditing and monitoring techniques are superior to LLM baseline methods in detecting violations of temporally extended behavioral constraints; with our approach, even small-model labelers match or exceed frontier LLM judges. Our predictive and intervening monitors significantly reduce the violation rates of LLM-based agents while largely preserving task performance. (Joint work with Parand Alamdari and Toryn Klassen.)