top of page
  • Writer's pictureJames Newman

TOTW #43 - Using ChatGPT with Celonis and BPMN Diagrams

One way we have incorporated recent advancements in generative AI technology with process mining is by using Celonis to generate a BPMN diagram of an organization's business process using real data and then feed it into ChatGPT to get the following information:


  1. Process Overview: Provide a general description of the process, outlining its primary purpose and overall flow.

  2. Element Analysis: Examine specific elements within the process, such as tasks, gateways, and events, to understand their roles and functionalities.

  3. Flow Analysis: Analyze the sequence flows to understand how the process progresses from one step to another and identify possible paths within the process.

  4. Error Checking: Review the BPMN structure for any inconsistencies, errors, or deviations from standard BPMN practices.

  5. Edit Suggestions: Based on the analysis, suggest possible improvements, modifications, or optimizations to enhance the process efficiency or clarity.


Some of the suggestions we have generated in the past from number 5 include the following:


  1. Gateway Conditions: The exclusive gateways lack condition expressions. It's beneficial to add conditions to the outgoing sequence flows of the exclusive gateways to explicitly define the decision criteria.

  2. Process Variables: Consider adding process variables that might influence the decision at the exclusive gateways.

  3. Documentation: Adding documentation elements to tasks and gateways could enhance understanding and clarity, especially for complex decisions.

  4. Event Types: Depending on the process's complexity, incorporating intermediate events (like timers or messages) could help model real-life scenarios more accurately.

  5. Sub-Processes: If any task is complex, consider breaking it down into sub-processes for more detailed modeling.

  6. Error Handling: If applicable, include error handling mechanisms using boundary events.


In order to get these insights, we generated the BPMN from the entire dataset. You could also try using the top 10 or 20 variants to get insight into only the "happy path." Or you could do the longest variants by cycle time to query about the worst offenders in your data. We then used the following ChatGPT application:

And just like that, you are able to creating BPMN out of Celonis!

118 views0 comments


bottom of page