aiExpert in practice- #3 – BPMN Process models

Source diagram

The source of this model is a sub-set of the E-mail voting example which is delivered in the Sparx ‘EAExample’ model, which is installed when you install EA.

This diagram can be found at: Example Model.Analysis and Business Modeling.”BPMN 2.0 Examples”.Process Diagrams.E-mail Voting Example, in the EA v17 version of EAExample.

The purpose of this document is to describe what the results were when parts of this model were analyzed by the aiExpert ‘BPMN2.0 best practice GPT5 Detailed’ query, which is used to examine BPMN process diagrams.

Please note that this is using the out-of-the-box function of aiExpert, with no tailoring of the prompts to suit the domain.

Please also note that this is not an implied criticism of this bit of modelling – we are just using to give some idea of the kinds of feedback you might expect when you apply aiExpert to your own models.

This diagram seems to be designed to highlight the range of support that EA has for BPMN modeling.

We chose the ‘BPMN2.0 best practice GPT5 Detailed’ query for this analysis, which collects all the data about all aspects of the BPMN process model which are shown in the diagram, and submits this to the LLM with the prompts:

System instructions: “You are an expert in the style and structure of BPMN 2.0 Process models.”

Rules:

Here are some parts of a BPMN Process.
Check if they conform to the best practices for the structure BPMN processes.
Use the 7PMG styles rules. Also use the Bruce Silver ‘BPMN Method and style’ guildelines
When you respond, grade the severity from 1 – a minor issue, to 5 – a serious problem.
Always include the ID of each element with any feedback

LLM used: gpt-5-mini

Data collector: ConnectorDetail

Important note:

We present the details of the Ai analysis largely without comment. This is not our model: we don’t understand the domain, so we are not in a position to say whether the feedback from the LLM is good or bad, useful or useless.  What we can say is whether the feedback might be useful, or whether it seems silly.

In the ‘suggested fix’ information which is returned by the AI, we highlight in italics the part of the input message to which the feedback refers.

Some of the feedback from the AI model

In this example, the LLM returned a lot of feedback. Below are some examples.

Main Process: ‘Discussion Cycle’

Severity – 2

Issue: Sub-process ‘Discussion Cycle’ has loopCharacteristics: Standard but no loop condition or multi-instance detail. If intended to repeat, provide loop condition or convert to multi-instance with clear semantics.

Suggested fix: Change “Discussion Cycle: loopCharacteristics: Standard (no condition provided).” to: “If the subprocess should repeat until a condition is met, add a loop condition expression (e.g. ‘while issuesRemain == true’). If it should run in parallel instances, use multi-instance characteristics instead.”

Activity :   E-mail discussion Dealine Warning (sic)

Severity – 1 (lowest)

Issue: Name contains a typo: ‘Dealine’ should be ‘Deadline’. Misspellings reduce professionalism and may confuse readers.

Suggested fix: Change “E-mail discussion Dealine Warning” to: “Rename to ‘E-mail discussion Deadline Warning’.”

Activity: Moderate E-mail discussion

Issue #1

Severity 3

Issue: The task ‘Moderate E-mail Discussion’ contains element ‘7 days’ (ID 95582). Tasks/sub-process containers should only contain child flow elements (if it’s a subprocess) — embedding an independent intermediate event inside a task is invalid/ambiguous.

Suggested fix: Change “Task ‘Moderate E-mail Discussion’ contains 95582 ‘7 days’.” to: “Remove the contained element from the task. If the timer is part of a subprocess, make it a child intermediate event of that subprocess, or model the wait as an intermediate event on the sequence flow outside the task

Issue #2

Severity – 3

Issue: Task is typed as User task but implementation is set to ‘##WebService’. A user task should not have a web service implementation. This inconsistency confuses readers and any execution mapping.

Suggested fix: Change “implementation: ##WebService; taskType: User” to: “If this is a human task, clear the implementation (or set to an appropriate UI form reference). If it is an automated service, change taskType to ServiceTask and keep implementation as ##WebService.”