Overview
This article is intended to serve as a companion to the 'Enabling AI Agents in Source-to-Pay Operations' webinar video, for those who prefer to read rather than watch an hour-long video. It can serve as a reference, and expands on some points mentioned in the video. Please feel free to provide us with any feedback in the comments below.
If you are looking for more opportunities to learn about Now Assist capabilities in Source-to-Pay, check out the related resource links above. Each listed ServiceNow University course includes a simulator instance with Now Assist enabled, along with content that details the capabilities and explains how to configure and use them. Our documentation further expands on how to configure and leverage our Now Assist capabilities.
Family release: Zurich
Monthly release: October 2025
Store applications:
Understanding Now Assist
To discuss AI Agents in Source-to-Pay, we need to ensure we have a shared understanding of what this solution entails. For further information on Now Assist capabilities, please review the Now Assist forum and Now Assist FAQs, linked in the Resources section above.
Foundations
The technologies that form the foundation for Now Assist capabilities are the ServiceNow AI Platform, Small/Large Language Models (SLMs and LLMs), and the Generative AI Controller.
The ServiceNow AI Platform is a unified platform with a single data model that spans the entire enterprise, featuring workflow and integration capabilities to engage with and act on Systems of Record. This platform holds enterprise data relevant to your business and is highly configurable.
Small/Large Language Models (SLMs + LLMs) are the underlying AI models that power generative AI capabilities. These models have been trained on large datasets comprising millions to trillions of points of context. They are designed to understand and generate human-like text based on the input they receive, and are generally trained on publicly available data.
The Generative AI Controller is a ServiceNow application that functions as a hub to manage interactions between the ServiceNow AI Platform and language model providers. ServiceNow offers OEM integrations with select models and provides several connectors for a variety of models and providers. Customers can also bring their own model provider licenses to the platform or integrate with their own models. The Generative AI Controller manages security, compliance, and governance aspects of these interactions.
Language Model Inputs
Now that we have established foundations, we can discuss the core components of interacting with Language Models: Prompts and Context.
Prompts, put simply, are text you provide to a language model that will cause a response to be returned. It can be helpful to think of the input box for tools like ChatGPT and Microsoft Copilot as the space where you write a prompt. The Language Model responds, using the information it receives in the prompt to guide its selection of words (or other content) to provide back as a response.
Context, on the other hand, is information you provide to a language model for it to leverage while processing the prompt. Context on its own does not trigger a response. Since language models are pre-trained, context is what allows you to present information that the model is otherwise unaware of.
OOTB and Custom GenAI Capabilities
In ServiceNow, we package prompts and context together into Skills. These are reusable Generative AI actions, comprising prompts tailored to various situations and models. They enable the handling of user inputs to populate a prompt before sending it to the language model. ServiceNow ships hundreds of skills, with Source-to-Pay shipping a significant number of these and introducing more with time. Customers can build their own skills or copy ServiceNow skills and refine or replace them with functionality tailored to their business needs.
Individual Agents and Underpinning Tools
Agents are made by combining one or more skills with one or more Tools. Agents are designed to take discrete actions or groups of actions to achieve an outcome. An example may be an Agent that queries and summarises all Invoice Exceptions for a given period, making observations on key themes and creating cases for Supplier Managers or fulfillers to take action as appropriate. Agents are a core step towards Autonomous Procurement, allowing actions to be identified and taken and engaging your fulfillers for key, valuable tasks.
Tools are the mechanisms through which agents perform specific tasks or interact with external systems. They can be Flows, Scripts, Knowledge Graphs, and more. This evolution shifts Generative AI capabilities from summarising, generating, and reasoning into the realm of taking action. Tools can act before or after skills in an Agent, allowing reasoning, summarisation, or generation to occur at different stages of the process based on inputs and outputs.
Combining Agents into Agentic Workflows
Agentic Workflows are made by gathering one or more agents together. Each agent has a description of its purposes and activities, as well as the circumstances under which it should be used. The Agentic Workflow itself also includes a description of its purpose, activities, and the outcome it aims to achieve. The Agentic Workflow acts as an orchestrator, processing the given input to determine which agent(s) should be used and triggering these agents to perform their actions. Agentic Workflows enable the definition and automation of complex workflows, allowing for more sophisticated and autonomous operations within Source-to-Pay.
Source-to-Pay Operations Skills, Agents, and Agentic Workflows
As of the October 2025 store release in Zurich, Source-to-Pay Operations ships two Agentic Workflows, one standalone AI Agent, seventeen Skills, as well as Now Assist for Document Intelligence, Virtual Agent, and AI Search capabilities. This list is continually evolving, and further details about the current capabilities can be found in ServiceNow Documentation using the Now Assist links in the Resource section above.
Approaching Use Cases
Agentic Workflows are stored in the ServiceNow database as 'Use cases', in the [sn_aia_usecase] table. This reflects their nature; an Agentic Workflow addresses a given process or use case. When considering use cases where Skills and AI Agents may add value, consider the following elements:
- Probabilistic:
Generative AI is at the core of this technology, and as described above, it predicts the next appropriate words to address a given input. This nature introduces probability, meaning that the output is likely to vary from execution to execution. When identifying appropriate use cases, ensure you understand and accept that components where Generative AI will be involved in decision-making or producing output will vary in nature. This approach works well in cases where tools or other technologies are used to perform the deterministic (yielding the same result every time) components, and Generative AI is used for reasoning and communication.
- Specific Context:
Individual skills and AI Agents should be defined with a high degree of context, and ideally supported with context or data relevant to the particular execution. While more general skills may be effective in certain circumstances, they introduce higher rates of variance compared to use cases where the individual AI Agent or Skills have a defined context, such as an understanding of domain knowledge.
- Clear Goal:
The desired outcome is well-defined and measurable. Much like any employee, Agentic Workflows, as well as AI Agents and Skills, are best supported with clear expectations and outcomes. This also improves reportability and the measurement of value, as defining the goal of these technologies enables the measurement of success and achievement rates.
- Defined Process:
The task follows a structured process that can be automated or guided. This pairs with all of the above elements, as poorly defined processes lead to inconsistent outcomes and expectations, as well as risking a lack of relevant context, resulting in wildly varying results when attempting to apply Generative AI.
The above elements are not exhaustive, but provide a good yardstick to consider before proceeding with ideating or implementing a new use case.
Demonstration and Next Steps
The demonstration begins at 23 minutes and 35 seconds into the webinar. It covers how to access Now Assist Admin, enabling and copying Skills, accessing AI Agent Studio, reviewing Agentic Workflows and AI Agents, as well as enabling and copying an Agentic Workflow.
For next steps, consider exploring the related resources above, particularly the ServiceNow University courses, which include simulator instances with Now Assist capabilities enabled. If you are already licensed for Now Assist in Source-to-Pay, you can also explore these capabilities in your own ServiceNow instance by installing the relevant Store applications present in the links above.
Conclusion
We hope this video and article provided valuable insights and education on Now Assist in Source-to-Pay Operations, including how to enable it on an instance to use and enhance this functionality.
We hope this video has answered your questions and provided a clear understanding of the capabilities available to you. If you have any further questions or topics you'd like us to explore in future videos, please post them in the comments section below. Your feedback is invaluable in helping us improve and deliver content that meets your needs.
Check out the Source-to-Pay Operations forum for more resources
