Successful workflow automation requires evaluating the level of sophistication your processes demand. Below are four levels of GenAI-driven automation, each offering unique benefits and use cases.
This approach integrates GenAI capabilities directly into existing software, such as Now Assist by ServiceNow. The user interacts through chat, and an LLM has the agency to call existing APIs. By constraining the agent's actions to a few well-defined workflows, you improve reliability. This interface also enables automation beyond the software's existing capabilities, for instance creating multiple tickets with a single prompt. Initially, software vendors may charge extra for these features, but as they become more common, they will likely be included in the base price. This level will not be replaced by generative process automation (GPA) because direct API access is faster, cheaper, and more precise.
A thin GenAI-enabled layer uses (low-code) tools with LLM features to span entire business processes. This consolidates multiple point solutions into more streamlined, end-to-end workflows. Processes that are fairly standard or should be standardized can greatly benefit from this approach. Examples include marketing, sales, and customer service processes. I expect the range of processes that can be automated at this level to expand quickly, as foundational model providers add capabilities like multi-modality and interactive voice. However, as Levels 1 and 3 continue to improve, fewer processes may remain at this level.
Here, the generative automation tool has direct access to a user's computer to record and automate tasks. This is ideal for highly customized or fluid scenarios that do not fit standard solutions. Consider any repetitive ad-hoc task you might have, such as renaming files, extracting data, or populating spreadsheets. Because GenAI can define a workflow (via natural language, video analysis, etc.) and also execute it (through computer vision, etc.), GPA may have a major impact for end users. Improvements in accessing other agents and computer interactions will make this level of automation very powerful.
This final level bridges the virtual and physical worlds with robotics or other devices. It enables end-to-end automation for processes requiring both software orchestration and real-world actions. For instance, a robot can deliver a package while updating its status in the relevant software.