Source-linked AI summary
AI-Augmented Business Process Management Systems: A Research Manifesto
Marlon Dumas, Fabiana Fournier, Lior Limonad, Andrea Marrella, Marco Montali, Jana-Rebecca Rehse, Rafael Accorsi, Diego Calvanese, Giuseppe De Giacomo, Dirk Fahland, Avigdor Gal, Marcello La Rosa, Hagen Völzer, Ingo Weber
TL;DR
The manifesto addresses how to make business-process execution more adaptable and AI-augmented while retaining process awareness and constraints. It defines ABPMSs and outlines their lifecycle and characteristics, then identifies research challenges as an open call for methods that realize the vision. Its proposed progression is modular, and the applicability and effectiveness of augmentation techniques depend in part on process predictability and repetitiveness.
Problem
The manifesto addresses how to realize process-aware information systems that use AI to make business processes more adaptable while respecting process restrictions.
Source: §1 · p. 2Method
Results
The manifesto presents ABPMS characteristics and challenges as an open, non-exhaustive call for research, with modular progression from BPMS to ABPMS shaped by designer choice.
Source: §5 · p. 14Takeaways & Limitations
Designers can prioritize a modular progression toward ABPMSs, whose autonomy is framed by operational assumptions, goals, and environmental constraints.
Source: §5 · p. 14The process’s predictability and repetitiveness may affect which augmentation techniques apply and how effective they are.
Source: §1 · p. 3
Abstract
from arXiv · showhide
AI-Augmented Business Process Management Systems (ABPMSs) are an emerging class of process-aware information systems, empowered by trustworthy AI technology. An ABPMS enhances the execution of business processes with the aim of making these processes more adaptable, proactive, explainable, and context-sensitive. This manifesto presents a vision for ABPMSs and discusses research challenges that need to be surmounted to realize this vision. To this end, we define the concept of ABPMS, we outline the lifecycle of processes within an ABPMS, we discuss core characteristics of an ABPMS, and we derive a set of challenges to realize systems with these characteristics.
1 VISION AND MOTIVATION
The manifesto defines an ABPMS as a process-aware system using trustworthy AI within restrictions to continuously adapt and improve business processes. It motivates this vision with execution flows that need not be predetermined and with AI capabilities tailored to process characteristics.
1 VISION AND MOTIVATION
- More execution data and AI support systems with non-predetermined flows and on-the-fly discovery, validation, and enablement of improvements.Because AI can decide whether and how tasks are executed, trust is crucial to acceptance.Source: §1 · pp. 2–3
- An ABPMS is a process-aware information system that uses trustworthy AI within restrictions to adapt and improve business processes against performance indicators.A BPMS need not implement every described characteristic or lifecycle step, but must incorporate AI to drive business-process execution.Source: §1 · pp. 2–3
- An ABPMS supports processes, tracks compliance with a process frame, orchestrates activities, and uses AI to attain process goals within that frame.It can detect situations where restrictions cannot be met and make process-level changes, such as inserting activities.Source: §1 · pp. 2–3
- AI for ABPMSs must account for processes’ discrete temporal nature, compliance requirements, tactical goals, and coordination among stakeholders.The system should also detect and respond to situations that may take execution outside its frame.Source: §1 · p. 3
- The process’s predictability and repetitiveness affect which augmentation techniques apply and how effective they can be.The manifesto outlines a possible AI-enabled lifecycle and contrasts ABPMS characteristics with those of a conventional BPMS.Source: §1 · p. 3
2 AI-AUGMENTED BPMS: A LIFECYCLE
The lifecycle combines a framed process with a perceive–reason–enact execution cycle, augmented by steps for explaining, adapting, and improving. Framing sets goals and constraints, while the process may emerge during execution and evolve as the system or agents reframe it.
2 AI-AUGMENTED BPMS: A LIFECYCLE
- Basic lifecycle steps are frame, enact, perceive, and reason; explain, adapt, and improve are advanced steps specific to ABPMSs.The basic steps also occur in traditional BPMSs, but are augmented with AI in this lifecycle.Source: §2 · p. 4
- The lifecycle has two main actors: the ABPMS and agents, which may be human or digital entities using and controlling it.The lifecycle figure represents steps as rounded rectangles and their sequencing as arrows.Source: §2 · p. 4, Figure 1 · p. 4
- Framing supplies initial goals and constraints, which may include procedures, norms, regulations, best practices, and KPIs, and can encode competing goals.Frames can use different specification languages and range from rigid process models to more open articulations of goals, constraints, strategies, and performance metrics.Source: §2 · pp. 4–5
- The enacted process can emerge from the ABPMS rather than being directly specified by the initial constraints, with framing determining its execution mode.Execution ranges from a detailed imperative process to goal-only framing that leaves the system freer to choose how to proceed.Source: §2 · p. 5
- Process-aware execution repeatedly perceives the environment, reasons over data and goals, and enacts actions, keeping execution within the frame.Perception can draw on heterogeneous, incomplete, or uncertain data; enactment may be direct or mediated through a human or software agent.Source: §2 · pp. 5–6
- Perceived information may describe facts about the world or the process being enacted, and the system can update either kind of knowledge.Execution can also trigger adaptation, improvement, or explanation, revising process knowledge, constraints, and goals; agents may reframe the system too.Source: §2 · p. 6
- A system counts as an ABPMS only when process-execution steps run under explicit process-aware framing.AI capability alone, including NLP, deep learning, or reinforcement learning, is not sufficient without this framing component.Source: §2 · p. 6
3 CHARACTERISTICS OF AN ABPMS
An ABPMS is expected to combine framed autonomy, actionable conversation, adaptation, self-improvement, and explainability. Together, these characteristics describe systems that act within boundaries, respond to people and changing contexts, and assess their process outcomes.
3.1 (Framed) Autonomous
- Framing bounds ABPMS actions; designers can use meta-framing to control which parts of the frame the system may change independently.More context can support informed decisions, while an adjustably autonomous system can ask agents for information it lacks.Source: §3.1 · p. 7
3.2 Conversationally Actionable
- Conversationally actionable ABPMSs proactively communicate about process actions, goals, and intentions, adapting conversation to user preferences, information, and situation.They can make recommendations and discuss benefits and drawbacks with users.Source: §3.2 · pp. 7–8
3.3 Adaptive
- Unlike conventional BPMSs that require human intervention for ad-hoc instance changes, ABPMSs may autonomously react to real-time changes in process context.Examples include reassigning work when a person's certification changes or redirecting flows when a truck is unavailable.Source: §3.3 · p. 8
3.4 Self-Improving
- Beyond reacting to external changes, an ABPMS should predict upcoming changes and prepare for them, such as scheduling maintenance when transport load is low.It may reserve capacity in other trucks to take over affected orders.Source: §3.4 · p. 8
- ABPMS decisions weigh competing process measures, such as operational cost and waiting time, and may affect whole processes or individual cases, temporarily or permanently.For maintenance, the system may weigh longer delivery time and higher cost against higher output quality.Source: §3.4 · p. 8
3.5 Explainable
- Process-aware explainability links historical framing and process outcomes to assess output validity, and supports faithful explanations and recovery or retraction from undesirable conditions.The paper treats fairness and auditability as part of broader business explainability, while addressing safety in framing.Source: §3.5 · pp. 8–9
4 RESEARCH CHALLENGES AND OPPORTUNITIES
The manifesto identifies five research directions for ABPMSs: situation-aware explanations, human-augmenting automation, run-time adaptation, perspective agility, and actionable conversations.
4.1 Situation-aware Explainability
- Situation-aware explanations should connect process traces, causal reasoning, and contextual factors across enactments, supporting real-time justifications and retraction when inconsistencies arise.Event knowledge graphs are proposed to represent behavioral and causal interdependencies over time.Source: §4.1 · pp. 9–10
4.2 Augmented Process Automation
- ABPMSs should augment rather than replace workers, escalating contextualized decisions to people, accepting overrides, and using corrective actions to improve their learning and autonomy.This responds to manual identification of automation opportunities and varied work practices that make full automation difficult.Source: §4.2 · pp. 10–11
4.3 Automated Process Adaptation
- ABPMSs should adapt processes at run time with minimal intervention, moving beyond predefined handlers because contingencies are incomplete or become obsolete; SmartPM still needs task families specified at design time.SmartPM builds recovery procedures without predefined handlers, but extensions are needed for new data types and tasks during run time.Source: §4.3 · pp. 11–12
4.4 Perspective Agility
- Perspective agility requires describing and inferring processes across chosen combinations of formalisms and entity behaviors, including structures that emerge at run time.Prior BPM integration largely targeted imperative and declarative formalisms; linked specifications and extensible graphs are proposed for broader support.Source: §4.4 · pp. 12–13
4.5 Actionable Conversations
- ABPMS interfaces should answer queries and act, proactively alert users, and recommend interventions; NLP and machine learning could tailor conversations using vocabulary and prior interactions.Existing conversational approaches include domain-knowledge and scalability challenges.Source: §4.5 · pp. 13–14
5 CONCLUSION
The conclusion presents ABPMSs as AI-enabled systems that reason about process state to improve performance, while framing the proposed characteristics and challenges as an open, modular research agenda.
5 CONCLUSION
- ABPMSs reason about current or multiple process states to determine actions that improve process performance, within operational assumptions, goals, and environmental constraints.Source: §5 · p. 14
- ABPMSs should engage human agents conversationally, explain their actions, and recommend process adaptations or improvements.Source: §5 · p. 14
- The proposed characteristics and challenges are non-exhaustive and support modular progression from BPMS to ABPMS according to designer priorities.The manifesto frames them as an open call for future methods and techniques.Source: §5 · p. 14