Large Process Models: Process Management in the Age of Generative AI
The breakthroughs in Generative AI have taken the (tech) world by storm and have the potential to bring about disruption and significant changes to many industries and domains.
An organization is a sum of its business processes and the many ways it operates and engages with stakeholders, so the intriguing question is… how will Generative AI affect or is already affecting the way organizations model, analyze, govern, and overall transform their business processes?
At SAP Signavio, we believe that with SAP Signavio Process Transformation Suite, we have the tool to leverage and reap the benefits of this technology for our customers in their process transformation efforts. The process observability approach that we take is unique in that it integrates data and knowledge from a wide range of sources – not only graphical models and event logs – and integrates them with the experiences of all involved people. Generative AI can reinforce the already strong value proposition we offer to our customers, allowing organizations to understand process dynamics fast and effectively (achieve an even faster time to insight), drive enhancements through recommendations (accelerating time-to-adapt), and monitor and predict the impact of process improvements and automations.
In this blog post, we will introduce the foundations we are building at SAP Signavio, moving from Large Language Models (LLM) to Large Process Models (LPM), while leveraging SAP´s 50 years of unique process, customer, and industry insights and expertise.
The evolution from Foundational Models (FM) to Large Language (LLM) Models
Foundational Models (FM) were mentioned for the first time by Stanford researchers to summarize a new level of Machine Learning models, closely related to the concept of transfer learning. Such models are trained on broad data sets that can be adapted to a wide extent, which makes it possible to specialize them for a certain niche or use case with only a very small set of training samples.
This achievement was a big breakthrough, especially from a commercial point of view. Traditional machine learning approaches require a massive amount of training data. Collecting and preparing such data is very time-consuming and comes with high costs and significant time investment. Having an already pre-trained Foundational Model and only finetuning it for specialization reduces the operational costs of building a new model, opening up a myriad of opportunities for new business uses cases.
Large Language Models (LLMs) are foundation models (FM) trained on large amounts of text data, consisting of billions of parameters. Given a prompt — a natural language description of a task — LLMs can generate text and perform text-based tasks.
Large Language Models (LLMs) are — enabled by recent breakthroughs in deep learning with the transformer neural network architecture — the fast-moving frontier of real-world artificial intelligence. Promising applications of LLMs are emerging in the enterprise software industry. While general-purpose LLMs are already now augmenting day-to-day knowledge work such as copywriting, specialized models are trained for domains such as software engineering, finance, and HR.
We enhance Large Language Models (LLM) to Large Process Models (LPM)
However, as LLMs are statistics-based tools that re-use large corpora of often poorly curated, human-generated text, their behavior is unpredictable, at least at times socially not acceptable, and frequently illogical, which currently limits the applicability of LLMs in many business contexts.
So.. can Large Language Models (LLMs) be applied to the business process management space?
Structurally, process models are logical sentences, hence many advancements based on LLMs technologies can be leveraged in the process management space. LLMs can train a broad knowledge that allows users to let the AI “write a recipe with these ingredients”. However, if a user would ask a standard generic AI “what is the biggest issue in my production processes?” they will not get the right answer. This is why it´s important to finetune and augment a generic LMM to achieve a Large Process Model (LPM). Given a safe and sound integration with classical algorithmic tools and structured data, the LPM then enables novel insights that in a classical setup would not even surface, thus drastically increasing process observability.
A Large Process Model (LPM) with access to unstructured and semi-structured organizational knowledge can leverage the very valuable know-how of decades of process experience from thousands of experts, as well as years of performance data of thousands of organizations. Large Process Models (LPM) can be further finetuned and augmented with context-specific, automatically tailored process and other business models, analytical deep-dives, and improvement recommendations from day one, substantially decreasing the time and effort required for generating insights.
The Impact of LPMs on Business Process Management Software
We expect that Large Process Models (LPMs) will become a reality as the main facilitators of intelligent process management. In particular, we expect LPM-augmented software to enhance or provide new business process management capabilities. We added them here, ranked from short-/mid-term to long-term:
1. Facilitating automated process analysis with contextualized knowledge.
We expect LPMs to support in enhancing process analysis capabilities with vast amounts of organizational knowledge, for example to produce process automation recommendations as well as structural process improvement suggestions.
2. Expanding the process intelligence universe by turning unstructured process information into insights.
LPMs could help generate business process models and process analyses, directly from unstructured process information and data that is abundant in organizations. This can drastically reduce the time and effort needed to understand and improve processes and operations.
3. Automating continuous improvement with the human in control.
The surge of LPM-based approaches that auto-generate human interpretable and verifiable analyses queries, compiling them to automatically tailored deep-dive process investigations. Based on them, process change actions could be automatically inferred and triggered in a human-in-the-loop manner to improve process operations in record-time.
4. Enterprise general artificial intelligence for self-driving organizations.
Finally, we expect the algorithmic toolbox around LPMs to continuously expand so that LPMs can automatically adjust their inferences based on past performance, and further augment the data and knowledge corpora to move towards the vision of a fully self-driving organization, enabled by enterprise general intelligence.
The breakthroughs in Generative AI have taken the (tech) world by storm and have the potential to bring about disruption and significant changes to many industries and domains, including Business Process Management. At SAP Signavio, we aim to leverage and reap the benefits of this technology for our customers in their process transformation efforts. Generative AI can reinforce the already strong value proposition we offer to our customers with SAP Signavio Process Transformation Suite, shortening time-to-insights, time-to-adapt and improving process monitoring.
In this blog post, we have introduced the foundations we are building at SAP Signavio, moving from Large Language Models (LLM) to Large Process Models (LPM), while leveraging SAP´s 50 years of unique process, customer, and industry insights and expertise. We also reviewed the potential impact of LPMs on Business Process Management capabilities.
In further blog posts we will continue to explain our vision for this topic and our efforts towards bringing this technology closer to our customers. Stay tuned!
What challenges does your organization currently face in process management that you think Large Process Models (LPMs) could help to solve? Share your thoughts in the comments below!
Visit our new www.signavio.com/process-ai website to learn more about this topic!