What LLMs can bring to process intelligence

The immense capacity for data generation that has occurred in recent decades, and the massive processing of data in the form of remote servers (what we call cloud) has been joined by the exploitation of artificial intelligence techniques. 

What we call generative artificial intelligence, the latest leap in this race for results that still has no known limit, allows us to interact with information as never before: thanks to large language models (LLMs), people can ask systems questions using their own language and receive natural answers to specific queries with a degree of precision that no one could have imagined.

How relevant is this for process intelligence? Well, a quick answer would be to enumerate all the process intelligence platforms that are currently integrating LLM-based chatbots or natural language interfaces that allow a more user-friendly experience. Truth is, there is a genuine interest in using LLMs for purposes beyond the always useful marketing. Which purposes? Here we describe some of the most salient applications of this new technology:

• First and foremost, process analytics. For years, process mining tools have struggled with the problem of achieving straightforward perceived value. It is hard for users of these platforms to get to clear insights without a thorough manual exploration of the process flows, including filtering attributes, isolating variants of interest and finding bottlenecks. LLMs help solve this problem by allowing users to ask questions using their own language, which are then interpreted and translated into analytical operations that satisfy the provided input.

• Once a process has been analyzed and a diagnosis is available, we can also rely on LLMs to provide guidance for its enhancement. However, this will depend on the kind of information the LLM has been trained on and it is probable that, for very specialized processes, more documentation needs be injected beforehand into the LLM (via prompting or fine-tuning).

• Enhancing a process often means considering automation, which also benefits directly from language models. These can transform simple instructions in text into a set of linked operations, eliminating the need to manually define triggers and actions and decreasing the time needed to implement automated operations within the process.

• Data preparation is another key task that can benefit from the use of large language models. Why? Because the only way to guarantee the trustworthiness of process intelligence tools is by preparing and ensuring high quality data is available. This is a hard task that often requires consultancy hours for understanding and combining raw data from different sources and systems, such as joining different tables, transforming data structures or aggregating events at different granularity levels. In this context, LLMs are helpful in assessing the user on how to proceed with these operations and suggesting ways to create a unified source of process data.

Although the application of this new technology to process intelligence still requires extensive research and testing, the possibilities LLMs open are so vast and the number of potential use cases is so high that, even once the current AI hype fever has worn out, we still see this as a technology that has come to stay. The main reason? A door has opened that allows non-specialized users to interact using their own words with process intelligence tools, which are otherwise hard to use and understand without further knowledge.

P.S. This first post has been written without help from any LLM. However, this newsletter aims to contribute to the dissemination of LLMs and their application to process intelligence, so rest assured we will be using them in the future for a more agile writing process.