Getting ready for ServiceNow Otto/AI

Matt Dodd
ServiceNow Employee

on ‎08-26-2024 02:19 AM - edited Thursday

Getting Ready for ServiceNow Otto/AI
What to plan before you switch it on

ServiceNow Otto can lift productivity and improve the experience for your people and your customers. A smooth rollout depends on more than the technical setup. This guide covers the readiness steps that make the biggest difference, across three pillars: foundation, governance, and adoption and scale.

What has changed recently

Since this guide was last refreshed, we have released Zurich (Q4 2025) and Australia (Q2 2026), and recently Brazil has also become Generally Available (GA) on November 5th 2026). The biggest shifts for AI, related to this article are:

  • AI governance: AI Control Tower has grown from a visibility dashboard into a full governance layer covering discovery, risk, security, guardrails, observability, and value measurement. It has also recently been renamed to the Super Intelligence Control Tower.
  • Model provider choice: you can route between model providers, including Anthropic (Claude), Microsoft Azure OpenAI, and Google Gemini, and make specific model selections. The ServiceNow-hosted NowLLM offering has also shifted to use the Gemma 4 26B-A4B model. (Model Provider Updates)
  • Knowledge Center: finds duplicate knowledge, supports AI-assisted merging, and generates content to fill gaps.
  • Now Assist Readiness Evaluation: a free store application that scans your instance and gives actionable insights on many of the topics below.
  • ServiceNow Otto: our AI conversational experiences now carry this brand, launched at K26.

We also publish a ServiceNow Otto Quick Start Guide that covers the basics of definitions and components. It is a useful companion to this article.

Pillar 1: Foundation

Get your data and search right. The first pillar covers the knowledge, records, and search that Now Assist depends on.

1. Knowledge management

Now Assist is only as good as the knowledge it can find. Up-to-date, relevant articles give you accurate, actionable answers. The usual knowledge management practices apply, and they matter even more here:

  • Retire old and outdated content so Now Assist does not surface it.
  • Collect feedback on articles and use it to refine the content.
  • Schedule regular audits to find gaps, outdated content, and areas to improve.

Access: unless you configure otherwise, AI can surface answers from every knowledge source a user can access. Use user criteria to keep searches within the right areas. Think beyond requesters and agents. A university must separate faculty and staff from students, and a retailer must separate shop floor staff from managers.

Knowledge creation: decide which users can generate knowledge with ServiceNow Otto. Wider access can reduce your knowledge backlog.

External sources: if knowledge lives elsewhere, consider external content connectors. Check the ServiceNow Store for any licensing considerations.

Knowledge Center: the best starting point for elevating your knowledge content. It covers the steps above and more. Read Introducing Knowledge Center with Advanced Editing, powered by ServiceNow Otto.

2. Service Operations Workspace readiness

Most agent-facing Now Assist skills are designed for Service Operations Workspace (SOW), not legacy backend forms. This is a common blocker that is easy to miss. Assess your SOW adoption and any related customizations before you start. If your agents still work mainly in classic forms, plan that transition as part of your readiness. Many of the AI skills do still work in the older UI, but you'll get better support and experiences in the workspaces.

3. Conversational catalog support

If customers should request catalog items through Virtual Agent, review the supported variables for conversational catalog skills. See what makes catalog items non-conversational in premium chat.

Once the Now Assist plugins are active, the Conversational Catalog Overview Dashboard shows which items are already conversational and what each remaining item needs. Start with your most requested items for quick wins. Victor Chen has written a helpful article on this topic, and there is an AI Academy session to watch.

4. Incident and case data quality

Incident and case summarization works on records with more than 50 words. The real value sits in complex records with many activity log updates, status changes, and reassignments.

When an issue is ready to resolve, resolution notes generation summarizes what happened and how it was fixed. Interestingly some customers decide not to use this feature, but i make the case for why you should in another community article.

That summary feeds other AI features, including the Predictive Intelligence similarity framework, and helps agents working on similar issues later. Good record data in means good resolution notes out. For more ambitious use, you can also generate knowledge articles with Now Assist. These help end users resolve issues themselves, which supports shift left.

Pillar 2: Governance

Control, trust, and compliance. Set up the guardrails, owners, and approvals before you turn on capabilities.

6. Security and data handling

Many customers have questions about how AI handles their data. We aim to be as transparent as possible. Start with these references:

7. AI administrator roles

Decide who owns and manages the AI configuration in your organization. Your system administrator may or may not be the right person. Many ServiceNow AI/Otto capabilities come with their own roles, so you can share ownership across teams. Decide which roles can use skills such as summarization or knowledge generation, and configure them in the Otto AI admin settings.

8. AI Data Steward and the Super Intelligence Control Tower

The Super Intelligence Control Tower (aka AICT) is now the central governance foundation for enterprise AI. The AI Data Steward role helps shape how your organization adopts AI, including which model providers are allowed. Read the Control Tower documentation for more insights - this is an evolving space with a lot of new changes and innovations being released, so its worth staying across.  A central concept here is whether you choose your AI Assets to be fully managed by the control tower, or just visible (unmanaged)

 

Practical takeaway: if you are deploying AI for the first time, validate your governance foundation before you enable any AI skills or AI Agents. Think about whether you want the approval workflows active for any new AI Assets being added to establish your governance process upfront.

9. Model provider choice

Through AI Control Tower, you can select and route between different model providers, including Anthropic (Claude), Microsoft Azure OpenAI, and Google Gemini. NowLLM is also available with an Open Weights Gemma 4 model too. As part of your readiness, decide:

  • Your default model strategy, and which workloads justify a specific provider.
  • Any data residency requirements that should drive routing.
  • Which providers to allow-list for the Now Assist Skill Kit and AI Agent Studio.

In-country and sovereign customers should note that not all model providers are available in Australian government protected (IRAP) environments, US federal environments (FedRAMP), US Department of Defense environments, self-hosted setups, or other restricted environments. Some Now Assist features may also be unavailable there. Check the latest availability before you commit to a model strategy, using ServiceNow Support knowledge base articles KB1584492 and KB0743854.

10. Instance and store app versioning

Access to the latest Now Assist skills generally requires your instance to run one of the most recent releases. Staying current gives you the most value. The Upgrade Center makes each upgrade faster and smoother. Upgrade Suites are also used now to help package compatible applications together, so you can upgrade them at once and reduce version mismatches. Staying across all of the bundled updates in a Suite an be difficult too, so ensure to review the Release Notes for the entire Suite before just hitting the 'Update' button. From that page, check for the Suite Version by date (on the left), and then select the relevant patch for your instance (on the right) to get to the exact bundle of plugins that will be updated, and all the changes included.

Note: For tips on Navigating the AI Suites for Store Plugin updates, you can read more here.

Pillar 3: Adoption and scale

Roll out and grow. Once the foundation and governance are in place, focus on how your people adopt and extend the capabilities.

11. Instance customization

Many organizations tweak the out-of-box configuration to meet business needs. Be aware of how those changes affect Now Assist. For example, if you use a different field on the incident table or add a custom state to your process, check the out-of-box UI actions and the resolution workflow. These can affect how the new skills work.

Certain out-of-box skills can be copied. A copy lets you adjust the inputs, the role conditions that control when a skill runs, and add input sources such as related tables or emails. For more capability, the AI Skill Kit lets you build custom skills or modify existing ones. Read the AI Skill Kit FAQ to explore it further.

12. Moving to ServiceNow Otto for Virtual Agent

Now Assist upgrades Virtual Agent from the older Natural Language Understanding (NLU) version to a large language model (LLM) version. Consider these points:

  1. Compatibility: each portal can have only one assistant. Choose an NLU-based Virtual Agent (legacy) or an LLM-based one using Now Assist. If you are new to Virtual Agent, you can skip the migration step.
  2. Topic migration: a topic migration utility is available. Do not migrate everything blindly. Evaluate where a topic is really needed, because the LLM-powered assistant often handles requests without one. Victor Chen's Migrating NLU Topics to LLM article compares the two technologies.
  3. Value from deflections: you can analyze and assess the value from Virtual Agent interactions.
  4. One or many assistants: each portal has one Virtual Agent, but you can create multiple assistants for different portals. Each can use different information sources and topics.
  5. Now Assist for Voice: the Australia release adds multilingual assistance, authentication, and real-time insights. If you support phone channels, factor this into your Virtual Agent strategy.

13. Developers and creators

Many capabilities can make your developers and creators more productive. Many of these are accelerators that need less planning than other areas. Check the latest release notes for the current skills, then decide which areas to target first. You may also want to try building ServiceNow apps with Claude Code and the ServiceNow SDK.

To get started:

Watch this space. Build Agent has grown from an IDE-only preview into a Studio-native experience. It can create AI agents, skills, and agentic workflows, generate and run test coverage, and support MCP client connections.

14. MCP and external tooling

The Model Context Protocol (MCP) Client is now generally available. It lets your AI agents and Build Agent reach approved external tools and data sources. For readiness:

  • Inventory the external tools and data sources you want to connect.
  • Confirm approval through your governance process. In the Australia release and beyond, unapproved MCP servers are hidden from AI Agent Studio.
  • Review MCP connections as part of your governance and security review, not as an afterthought.

15. AI agents and agentic workflows

Agents are a whole topic of their own. Once you have covered the basics in this guide, you can start planning. To find your first use cases:

  1. Explore the out-of-box options. Review the agentic workflows and see if any match your needs. This is a quick win. Use them as they are, or clone and configure them.
  2. Use the AI Agent Advisor. The AI Agent Advisor mines your instance for common use cases and suggests suitable agents and agentic workflows.

Additional resources

Impact services

Customers subscribed to ServiceNow Impact can use accelerators to prepare for Now Assist and AI agents. View the full set in the accelerator catalog.

Summary

Implementing ServiceNow Otto involves much more than switching on features. Get your knowledge and search right, set up governance before you enable capabilities (or at least early on), and then plan adoption and scale. Many of the topics above deserve deeper exploration of their own, so pick the ones that matter most to your organization. We would love to hear your field learnings in the comments below.

Labels:
  • 26,240 Views
Comments
priyaradhak
ServiceNow Employee
‎08-28-2024 12:25 AM

Great article! Thanks Matt!

Yogesh Shinde
ServiceNow Employee
‎01-06-2025 08:44 AM

Great article! One of the key aspects is the adoption of Service Operations Workspace (SOW), as most of the Now Assist Skills are designed to be used through SOW rather than backend forms. Therefore, SOW and its related customizations are crucial to check during the pre-assessment or instance assessment before starting the Now Assist implementation.

Additionally, it's essential to properly implement and adopt AI Search and Virtual Agent. AI Search is the most critical and widely used functionality since it is accessible to all users, unlike Agent skills which are limited to agents. I plan to add field learnings to this article and encourage others to do the same. Now Assist is an excellent product!

AshishKM
Kilo Patron
‎06-03-2025 04:20 PM

This is great summary about preparation on Now Assist learning.

I landed on this page from "Now Assist ITSM" course's best practice section. 

0 Helpfuls
ChristianSt
ServiceNow Employee
2 weeks ago

Great summary.

Link in

Pillar 2 — Governance (control, trust, and compliance), section 6. Security & Data Handling, #8 Now Assist Guardian

appears broken.

 

It probably should be updated to AI Guardian.

0 Helpfuls
Version history
Last update:
Thursday
Updated by:
ServiceNow Employee
Contributors