ChatGPT or Copilot for process models: where it works and where it stops

ChatGPT and Copilot can turn a good description into a first draft of a process. For client work, that is not the question. The difference is in what comes after that one model: forty models that look alike, deliverables that stay consistent with the model, validation with the client, and knowing where your client's data ends up.

We hear the question in demos almost every week, and it is a fair one. Every consultant has a chatbot open, and the models get better every six months. Why buy a separate tool for process modelling? Here is the honest answer.

What a general chatbot does well

Start with what works.

• A first draft from a description. Paste a transcript or a process description and you get steps, decisions and roles back as a starting point.

• Summarising a transcript. Bringing an hour-long interview down to the main points works well.

• Thinking along. "Which exceptions am I missing in this process?" often produces useful questions for the next conversation.

For a one-off sketch of an internal process, that can be enough.

Where it stops: the work around the model

On a client engagement, the first model is rarely the problem. The problem starts with the second, the tenth and the fortieth.

1. Ten consultants, ten styles

A consultancy does not sell one model but a way of working. When ten consultants each write their own prompt, you get ten styles: one uses lanes per department, another per system, a third names tasks differently. Each model can be right on its own, but side by side at a client with forty processes it is a patchwork that costs a reviewer hours.

One shared tool gives a medior and a senior the same kind of model. The medior makes the first version, the senior reviews the content rather than the form. At Bovemij, the first version of a model is about 80 percent right that way (read the case).

2. The model is the start, not the end

A client rarely wants just a diagram. They want the RACI, the work instruction, the requirements for the implementation partner. In a chatbot, each of those is a new conversation. Change the process after validation, and it is up to you to keep the RACI and the work instruction in line. In ModelMatic, the RACI matrix, SOPs and requirements come from the same validated model (see the features).

3. A model that lives in your chat gets validated by nobody

A process model only becomes valuable when the client recognises it. That means walking through it together, capturing feedback, adjusting, and keeping the model somewhere a colleague can pick it up next month. A chat conversation is not built for that. In ModelMatic you validate the model with the stakeholder and capture feedback as you go. You then export it in the BPMN 2.0 standard to the tools your client already uses.

And your client's data?

This is the point clients of consultancies raise most often themselves. An interview transcript holds names, systems, sometimes rates and weak spots. Pasting that into a personal ChatGPT account shares it with a service whose terms the client never approved.

To be fair: Copilot inside your client's own Microsoft 365 tenant is often well covered on data. There, the difference is not the data but the three points above.

ModelMatic processes and stores all customer data encrypted within the European Economic Area, and your data is not used to train AI models. The rest of the terms are in our trust center. For what a client can expect around ISO 27001 and GDPR, see process modelling for EU consultants.

Side by side

• The same kind of model from every consultant. ChatGPT or Copilot: depends on each person's prompt. ModelMatic: one way of working for the whole team.

• RACI, SOP and requirements that follow the model. ChatGPT or Copilot: separate conversations you keep in sync yourself. ModelMatic: from one validated model.

• Validating with the client and capturing feedback. ChatGPT or Copilot: outside the tool. ModelMatic: in the tool.

• Keeping and sharing models with colleagues. ChatGPT or Copilot: chat history. ModelMatic: one shared place for the team.

• Where client data goes. ChatGPT or Copilot: unclear with a personal account, often covered with Copilot in the client's own tenant. ModelMatic: within the EEA and not used for training.

When the difference starts to count

A one-off sketch for internal use is something a chatbot can do. The sum changes when your team models for clients every week, when models are made and reviewed by several consultants, and when deliverables follow. Then the question is not whether a chatbot can produce a first draft, but how many hours your team loses on everything around it.

Want to compare on one of your own processes? Book a demo and bring your own transcript.

On notation: BPMN 2.0 is an open standard from the Object Management Group (specification).