Robert Kugel's Analyst Perspectives

Agents are all the rage—and for a good reason. They are a way to automate work almost effortlessly so that repetitive and boring tasks get done with the least amount of effort on the part of the operator. In business, agents can be a boon for customer satisfaction and a way to improve worker productivity. They are alluring, with an almost unlimited number of potential use cases. However, as currently described, general-purpose agents and chatbots can’t reliably perform many important business processes that involve collaboration in a complex set of iterative steps. For this, another form of agentic artificial intelligence-assisted process management, which I’m calling generative automation, is necessary and often delivered as off-the-shelf functionality in business software. It’s generative because it is constantly refining itself to reflect real-world business conditions and practices. Like other agent systems, generative automation applies machine learning in a way that mimics how humans behave based on observed states and their context to direct the actions that follow. Rather than serving as a general-purpose tool, generative automation is an application-specific design technique that makes it a more practical and sustainable approach to using augmented intelligence in computer-aided process execution. Generative automation is only a concept now, but I assert that by 2028, more than one-third of business applications will include some form of it to shorten process cycles, increasing productivity, organizational agility and customer satisfaction. AI agents can easily automate relatively short, repetitive tasks, enhancing finite condition decision-making and providing personalized assistance across various business functions. Chatbots and personal assistants are increasingly able to improve individual productivity and, by eliminating dull, repetitive work, promote individual satisfaction. What I am calling generative automation is something new. It allows for far more complex coordination and communication between multiple people in completing time-consuming processes managed in an application. Importantly, it does so without extensive human programming, and it enables the process to evolve in line with business conditions and practices. The software continuously molds itself to how work needs to be done rather than requiring organizations to adapt to how the application is designed. Generative automation is best described as augmented intelligence because, by using machine learning techniques applied to process mining methods, it enables systems to recognize when a particular step in a task can be performed automatically with little or no risk of a misstep or presents options for decisions or next steps when certainty falls below a defined acceptable threshold. The human effort and time required to complete a business process is the function of the interventions needed and lags created by serial motions within a process, including contact-and-response, review, approval and execution. Generative automation collapses or eliminates lags by its ability to convert these to concurrent actions or fully automate some or all steps depending on the context. By significantly compressing the time between steps in a collaborative process, this type of software can shorten cycles in cases where the degree of complexity in human interactions and their required coordination is relatively high. Generative automation could be preferable to agents for handling collaborative, iterative processes with complex, non-linear decision nodes, especially where the actors and the next steps in a process are defined by conditions and context that evolve over time. For example, many tasks in the accounting close follow iterative paths involving multiple participants, as do supply chain management events where a delivery delay can set up a complex choreography of collaborative decision-making to deal with the delay, preferably in a relatively optimal fashion. In theory, and described at a high level, any agentic system can do almost anything. But, like most software, the devil is in the details. A generative automation approach minimizes the cost and effort required to design and maintain a reliable and scalable automation system within a business application. Such a system can never pretend to be autonomous. It must be able to defer to human decision-making and action when conditions lack sufficient certainty. It must be able to continuously learn from these interventions with a deep understanding of the context of the situation. Generative automation is built on process mining, which applies algorithms to computer event log data to identify trends, patterns and details of how a process unfolds. Process mining analyzes event data from the logs of software applications to understand how processes are designed to be performed and how they are actually performed. Typically, the technology will show that there are many variations in how a process unfolds and the context in which those permutations occur. The results can be used to uncover the source of bottlenecks, delays, unseen risks and unnecessary workloads that, in turn, allows organizations to institute improvements. You can gain a deeper understanding of process mining from this analyst perspective by my colleague, David Menninger. Process mining also provides the necessary data to train generative automation models, applying machine learning to develop and train process workflows to replicate and optimize how tasks are performed. ML is a subfield of artificial intelligence that uses technology to process information similarly to how humans do. This includes improving accuracy as more data is processed in ongoing model training cycles. Just as generative AI can replicate many sophisticated human tasks, generative automation replicates the condition and evolution of how repetitive work is performed in enterprises. Generative automation guides workflows where the routing, roles and responsibilities are determined by the context of the existing conditions at this node of a process. The context is defined by how the process is typically performed but is mindful of the chain of events and conditions leading up to the current node in the process. The ability to train these generative workflows is built around process mining and analysis rather than some idealized definition of a process because this method incorporates the reality that business processes are not always performed as planned. Even after process mining and analysis eliminate unnecessary and unproductive variations, some variations will remain, especially for complex interactive activities. Moreover, businesses evolve to address change, so continuous training of generative automation business process models is necessary. Generative automation’s core use cases cover conditions under which applying machine learning techniques to automate and assist in executing complex, interactive processes, organizations can gain productivity or compress decision-execution cycles. It complements the power of general-purpose agents for use cases where these cannot handle the complexities or are not adaptable enough. Although, in theory, generative automation can be applied anywhere, the most practical application is within enterprise applications because this simplifies the use of process mining. As is the case for all AI-led efforts, accurate, consistent and timely data is essential. When process mining is used across platforms or applications, it faces the challenge of having to deal with incompatible and incomplete processes. Omissions and errors in system logs will create inaccurate process maps, especially in missing steps or tracking the time it takes to complete a process. Narrowing the scope to a single platform or software application minimizes the challenges posed by third-party process mining and analysis software because it simplifies the task of creating and continuously training process execution models. This sort of application-specific process mining can deal with data quality issues because the high-quality, consistent and comprehensive event logs are an integral part of the software, eliminating problems with siloed data, inconsistent data formats and incomplete records. The main drawback of this approach is that it is limited to processes executed within the specific application, but that isn’t the point. Generative automation is just a concept now, but application software providers should investigate how to apply it to enable software to adapt and optimize processes in enterprises. Like other forms of AI and generative AI, generative automation gives business applications providers a once-in-a-generation opportunity to differentiate offerings in mature markets. Regards, Robert Kugel

Robert Kugel's Analyst Perspectives

Generative Automation: An Alternative for Process Optimization

Our Analyst Perspective Policy

Our Analysts

David Menninger

Jeff Orr

Keith Dawson

Mark Smith

Matt Aslett

Matthew Brown

Robert Kugel

Stephen Hurrell

Ventana Research

Subscribe to Email Updates

Posts by Month

Posts by Topic

Analyst Perspectives Archive

Company

Resources

Helpful Links