Business Process Modelling and AI

Traditionally, business processes are designed through domain expertise to model and execute the allowed or expected behaviour of e.g. a system or supply chain. AI continues to be increasingly adopted by organisations, changing their service offerings and the way they operate. In this respect, the development of AI engineering benefits from BPM and vice versa. On the one hand, the development and deployment of AI-based systems is a process which needs to be managed to ensure correct, compliant and responsible application of these systems. On the other hand, AI-based techniques can be used to ensure provable compliance of processes against regulations and organisations’ own business rules. This not only includes automated compliance checking of processes, but also automated generation of processes based on a given goal, which allows business agility due to reduction of manual design efforts and real-time anticipation to disruptions.
With a combination of human and autonomous agents executing processes, AI-based techniques are used to efficiently and reliably recognise the goal of that agent early on in the process and predict its future actions.


Automated process generation

Due to the size and complexity of rules originating from legislation and business requirements, it is increasingly challenging to ensure that each business process adheres to those rules. Business processes have to be designed such that they not only compliant, but also understandable and practical in their execution, which is a tedious and error-prone task.

We work on automated generation and adaptation of (understandable) business processes based on business rules, laws and regulations using AI techniques such as automated planning. The resulting business processes are guaranteed to be compliant and corresponding to the organisation’s goals and business rules. Furthermore, this allows organisations to adapt to different markets where tools provide required changes to their processes. Finally, these techniques allow to dynamically alter business processes during runtime, in order to anticipate on changing business environments and disruptions. Consequently, each business process is resilient to disruptions and able to actively adapt to accommodate new requirements.

Process variability

The existence of different process variants is inevitable in many modern organisations. Variability in business process support requires a flexible business process specification that supports the required process variants and is at the same time compliant with policies and regulations. However, manual specification of such models and constraints is complicated and error-prone. As such, we develop technology that allows to represent different process variants in a single model, and subsequently derive variability rules from this model providing a flexible process that supports the underlying models while being compliant by design.

Goal recognition

In this project, we utilise existing process mining techniques to solve the Goal Recognition problem, which aims to infer a possible goal of an autonomous agent (e.g., a robot or human) based on its observed behaviour. Existing approaches either require pre-defined plan library models or full domain models. In many real-world scenarios, however, these models are not available or require significant effort to be created.

We develop an approach that learns knowledge models by observing agents’ behaviours, which can be subsequently applied to recognise the goal of a newly observed behaviour. As such, this allows to:

  • Perform goal recognition in the absence of domain knowledge;
  • Perform goal recognition in non-stationary environments;
  • Recognise and incorporate new intentions of autonomous agents.

CSIRO Mission Alignment

Our research aligns with the vision of the Trusted AgriFood Exports Mission.


Research Team Involved

Trustworthy Processes Team