Process mining and automation are connected to AI, large language models, and ChatGPT, but how exactly? Let's have a look at modern organizations, knowledge work, business processes, and use cases for automation with ChatGPT. Here's our agenda:
Why is This Relevant
Large Language Models in Knowledge Work & Future Outlook
Process Mining and Artificial Intelligence: Cognitive Automation
ChatGPT as a Tool for Process Mining Users
ChatGPT as a Tool for Process Mining Vendors
AI at OSNOVA
Conclusions
Why is This Relevant?
The mention of ChatGPT may elicit mixed reactions from people. The truth is that the potential of ChatGPT and large language models to change the way modern organizations work is pretty much limitless. It is also very relevant for process mining, as language model-based automation will be streamlined across business processes, at least during the first wave of implementations. This wave is happening already and will become apparent within 12 months. So I invite you to have a look at the function of the organization, the nature of knowledge work, and what we can conclude about the development of AI-based automation.
Organizations exist because they can complete certain tasks more efficiently than in an open market. Moreover, the efficiency, as well as the market power of modern organizations, largely depends on their ability to manage internal information flows. For our purposes, an atomic element of an information flow can be a purchase order, a quality notification, or a support ticket – what we usually derive case keys from in process mining.
Automation of knowledge work already has a tangible impact on the job market. It is still limited, although the Challenger Report in May 2023 attributes 4000 jobs lost in the USA to artificial intelligence, from a total of 80000. Content writers are being replaced by AI as we speak. And anecdotal evidence starts to emerge about software developers taking on 2-3 jobs: ChatGPT allows them to code three times faster, while companies' requirements and expectations are yet to catch up.
How are Large Language Models Used in Knowledge Work?
Large Language Models (LLMs) have revolutionized the way knowledge work is done. They are used to create proposals, white papers, technical documentation, marketing materials, and more. With the ability to generate human-like text, LLMs can produce high-quality content that is indistinguishable from that written by a human.
How much more is there really to the daily work of most people?
When we look from the perspective of the classic process mining project in order management or accounts receivable, most of it is about writing e-mails, creating invoices, and making documentation and slides.
And since ChatGPT can pass the law exam better than 90% of humans, it can be adapted to work through the details and customizations in companies, provided the right architecture & infrastructure.
Future Outlook
As the saying goes, it's difficult to predict things, especially the future, but let's try. To do so effectively, I propose the following building blocks - which are already happening - that we can extrapolate for the next months and years:
Job Deconstruction: Breaking down jobs into granular tasks to see which can be automated
Automation of Individual Tasks: Using AI to perform repetitive activities that require reasoning
Creation of Autonomous Agents: Using AI to perform complex tasks independently along end-to-end processes
Using job deconstruction for accounting, we can come up with the following task description:
Create invoice --> write a message & send it --> receive and interpret an answer --> trigger another action internal systems --> find mistakes and correct master data
Not only typical invoice-related activities are mentioned (what would be an entry point for process mining) but also communication, working with multiple systems, etc.
In the next 12-24 months, individual activities will be automated. It's clear since many of those are already automated by batch jobs & RPA. The qualitative difference is that LLMs can dynamically generate content and reason about it. For instance, it can decide that master needs fixing and suggest this as an action.
It is only logical to connect those activities with an orchestrator - a chatbot to guide other chatbots, control their execution, and aggregate + present their output. You can think of it as a manager of chatbots that itself is a chatbot. And just for reference, this is no science fiction, it is happening already with LangChain, HuggingFace Transformers Agents, AutoGPT, babyAGI, and some other tools. Those are in relative infancy - for now.
Process Mining and Artificial Intelligence: Cognitive Automation
To understand the connection between process mining and artificial intelligence, we need to think like this:
process mining is a tool to identify automation potential
AI is the tool to realize those automations
Of course, PM is doing much more than just identifying potentials, and AI use cases go beyond what we would consider process automation - we take this perspective to limit the scope of this article.
Moreover, it's helpful to consider three types of automation: workflow/backend, RPA, cognitive. I lump together workflow automation, batch jobs, automation scripts because the user usually doesn't see or care about those. RPA is based on clicks in the interface, and cognitive is a fancy term for machine learning-based automation. LLMs are an example of machine learning, hence fall into that category.
ChatGPT as a Tool for Process Mining Users
In the realm of process automation, ChatGPT is an invaluable tool for process mining users. It can help with:
automated communication with customers, partners, and vendors (customer service, order management, etc)
automation of document creation, routing, information gathering, preparation of documentation
generation of analytical findings in dashboards (trends, anomalies, distributions)
identifying errors in master data
ChatGPT as a Tool for Process Mining Vendors
And for the vendors, it offers powerful new capabilities:
Chatbot for self-service data exploration, similar to "Copilot for data teams" by Atlan
ChatGPT with code interpreter can find 2 errors in 35.000 data points, or analyze trends in the data on its own
Process mining implementation coach & generating project documentation (my Demo with ChatGPT)
Source system documentation, explanations, and feedback on architecture designs
SQL code generation, quality review, and error spotting
AI at OSNOVA
I don't only give recommendations to others about using generative AI. In my company, OSNOVA, we outperform large competitors by applying AI for:
customer research in the sales process with the Promptbook
generation of ideas for products
writing Python (which I know) and React code (which I don't) for demos and tests
writing e-mail sequences, LinkedIn posts, website content, documenting internal processes
transforming raw Teams/WhatsApp messages into documentation and proposals
Perhaps not surprisingly, this article has largely been written by ChatGPT:
I spoke the draft of the detailed agenda of these slides into text with Microsoft Dictate
I used ChatGPT to write out coherent paragraphs and bullet points, and edited them manually
I used the Tome app to automatically generate images and rewrite text for my conference slides (which this article is based on)
Conclusions
To sum up:
generative AI is here to revolutionize the nature of knowledge work
business processes are at their core designed to manage information flows
documents are atomic units of information flows and are very ripe for cognitive automation as presented by large language models
process mining customers and vendors alike already have multiple use cases they can implement today
starting is easy and often involves simply asking ChatGPT to do a particular task you're working on
For more advanced use cases, ping me on LinkedIn.