Can AI Model Business Processes Like a Human Analyst?

We’ve just wrapped up another BPMN business process modeling training. And once again, the same question came up: “What about AI? Can it model processes like an analyst would?”

Here are some observations and thoughts on the topic.

Current State: AI Modeling Is Already Available

Several vendors — including Camunda, Appian, and BonitaSoft — have already introduced AI-powered modeling capabilities. But in practice, the results are underwhelming:

• models are overly simplistic,
• the logic is often shallow,
• subprocesses are missing,
• and the structure lacks the hallmarks of good BPMN modeling style.

In many cases, it’s easier to build a model from scratch than to fix what AI generated.

Why Isn’t AI There Yet?

The main reason lies in the nature of the data. Process models are stored as XML files with rigid structures. For large language models (LLMs), this format remains difficult to process deeply — especially at the level required for professional-grade modeling.

So the issue isn’t just with the algorithms — it’s also with the format of the input data.

Where AI Is Already Useful

Despite its limitations, AI is already proving effective in specific areas:

1. Process Mining

AI performs strongly here because it works with structured data extracted from IT system logs. The models it produces are close to real-life processes, though vendors often avoid displaying them in BPMN format — to sidestep questions about the proper use of notation elements.

2. Converting Text-Based Procedures into Process Models

LLMs can already extract structure from textual regulations and generate initial draft models. These models are simplified, but they provide a helpful starting point for analysts and can save significant time.

What’s Next?

Will AI eventually be able to model processes at the level of a skilled analyst? That’s likely just a matter of time. More and more companies are storing their processes in the cloud — in vendor-managed databases. Access to large volumes of historical models will help algorithms learn and improve modeling quality.

AI will still struggle with exceptions, custom business rules, and nuances like choosing between subprocesses and call activities. But the overall quality will continue to improve. In the coming years, AI will increasingly act as a support tool — a co-pilot for analysts.

Conclusion

AI won’t replace skilled business process analysts anytime soon. But it can already make their work easier — generating drafts, accelerating modeling and analysis, and enhancing how we interact with process data. It’s not replacing analysts — it’s extending their capabilities.