Swaroop-Pydy
ServiceNow Employee
ServiceNow Employee

Introduction

 

This article is part of a series from the AI Center of Excellence Team at ServiceNow.

 

We’ve worked with many teams across different industries to help them roll out Now Assist for HR Service Delivery (HRSD). Even though every company is unique, we’ve started to see clear patterns in what makes these implementations successful and what tends to slow them down. 

 

This isn’t a formal instruction manual. Think of it more like a playbook built from real experience in the field. Whether you’re just starting to plan your rollout or looking to improve what you’ve already launched, these insights are meant to help you avoid common pitfalls and get the most value from the platform. 

 

Where Challenges Typically Happen 

Most challenges fall into three areas: people, process, and technology. It’s important to consider all three together. Even with the best technology, things can go off track without the right people and processes in place. 

 

People Challenges 

  • Teams sometimes aren’t sure how Now Assist fits into the day-to-day work of HR agents. This makes adoption harder. 
  • Some agents worry that AI output might be inaccurate or non-compliant. Others might trust it too much without review. 
  • Training and change management often come too late in the project or are too basic to build real confidence. 
  • There’s often no clear way for agents to give feedback after launch, and no one owns ongoing adoption and overall performance. 
  • Success metrics are not defined upfront, which makes it hard to measure results or show return on investment later. 
  • Often expectations are not set as to what Generative AI can do in regards to knowing attributes of the user. 

 

Process Challenges 

  • Teams often focus on the technology but skip defining which use cases matter most or how to measure success. 
  • There’s no clear way to connect AI usage to actual business outcomes, making adoption harder to scale. 
  • Existing process issues sometimes get overlooked during implementation, creating friction later. 
  • Knowledge articles may be outdated, inconsistent, or hard to navigate, limiting what the AI can effectively use (especially within Agent Recommendations space). 
  • Without a way for agents to give input on AI summaries or search results, improvements are hard to track. 
  • There’s often no clear governance over how AI is used, who can access it, and how it’s being monitored. 
  • HR Cases lack the details needed to generate high-value summaries and resolution notes. 

 

Technology Challenges 

  • Some teams try to implement Now Assist on older or unsupported HR workspaces, which might not be compatible to support GenAI features. 
  • Key conversational HR data stored in custom fields or tables that GenAI can’t access or not built to support out of box. 
  • Important plugins may be missing or misconfigured, causing delays or partial functionality. 
  • AI-generated summaries can sometimes pull in private or sensitive data if controls aren’t set up correctly. 
  • Teams may not be using Now Assist Guardian, which is a key tool to catch security risks like offensive content or harmful prompts. 

 

What Works Well in the Field 

 

Here are some practical steps teams have taken to move forward with confidence and clarity. 

Make Sure the Platform is Ready for GenAI 

 

Problem: Using older interfaces or overly customized platforms makes GenAI features unreliable or incompatible. 

 

What to Do: 

  • While majority of the key features work on both workspaces and Next Experience interfaces, it is recommended to Upgrade to the latest HR Agent Workspace if workspace is the primary interface for agents 
  • Run a readiness check to catch any hidden customizations that could interfere with AI behavior. This can be a questionnaire to evaluate customizations applied to key UI actions and metadata fields within your instance to determine possible conflicts 
  • Confirm that key plugins like "Now Assist for HR Service Delivery (HRSD)" and other contributing plugins are installed, up to date and functioning properly. 
  • Avoid customizing core fields like description, short description, or lifecycle states. These are often used by AI models. 
  • Cross-scope access allows one app’s scripts to interact with another app’s resources like tables or APIs. If access issues come up, review and update cross-scope policies to allow the right permissions. This also includes monitoring Application Restricted Caller Access records that are not at the state of Allowed. 

Example: One team started to see a "create knowledge" popup every time they closed a case. The issue was tied to an outdated Workspace. After upgrading the workspace plugin to the latest version, the problem was fixed, and the experience was much smoother. 

 

Control What the AI Can Access 

 

Problem: Without clear boundaries, AI might include sensitive information in summaries. 

 

What to Do: 

  • Ensure your internal security teams are in the know with the implementation 
  • Use access controls (ACLs and RBAC) to make sure only the right users can view sensitive content. 
  • Enable Now Assist Guardian to monitor what’s being sent to the AI engine, and to catch things like unsafe content or prompt injection. 
  • Establish governance around who can use AI, how it’s used, and how it’s monitored 
  • Train agents on best practices for where to log personal or sensitive information, so it doesn’t end up in the wrong place. 

Example: There is an instance where health-related details were pulled into AI summaries from unrestricted comment fields. This was resolved by excluding those fields and updating agent guidance on secure data entry. 

 

Train Agents and Keep Them in the Loop 

 

Problem: Without proper onboarding, agents either don’t trust the AI or trust it too much. 

 

What to Do: 

  • Use real examples in training to show how AI supports decision-making but doesn’t replace human judgment. 
  • Emphasize the “human-in-the-loop” approach. AI offers help, but the agent still owns the outcome. 
  • Set realistic expectations about the AI’s limitations. It’s only as good as the data it’s working with. 
  • Make it easy for agents to flag issues or give feedback inside the workspace. 
  • Schedule regular check-ins to gather input, answer questions, and make improvements based on what is working. 

Example: In one case, AI helped summarize a complicated leave request. This saved agents time, but they also learned to change the output to remove personal information by following policy rules. It’s a great example of how training helps build trust and keep people responsible. 

 

Improve the Knowledge Base That Feeds the AI 

 

Problem: Case deflection and search adoption is low if your knowledge base isn’t clean or well-organized. 

 

What to Do: 

  • Run a content audit. Remove duplicates, clean up formatting, and retire articles that are outdated. 
  • Create hyper-focused knowledge articles and reduce links. 
  • Provide content in text alongside images, videos and attachments 
  • Maintain consistency and quality via governance. 
  • Tag articles with clear metadata like topic and audience, so they’re easier for the AI to find and rank. 
  • Add summaries and section headings to improve readability for both agents and AI models. 
  • Assign owners to each knowledge article so there’s accountability for keeping content up to date. 
  • Create a regular process to review and improve knowledge content based on what agents need. 

Example: There are implementations that deliver weak search results and irrelevant agent recommendations due to inconsistent tagging. A targeted cleanup and retagging project led to a measurable improvement in response to relevance. 

 

Roll Out Features Gradually and Plan for Growth 

 

Problem: Turning everything on at once creates confusion and makes it hard to measure what is working. 

 

What to Do: 

  • Define the scope of use cases early. Be clear on what you’re testing first and how you plan to expand later. 
  • Set measurable success goals for each use case like reduced handle time, faster resolution, or improved accuracy. 
  • Start with passive features like summarization and resolution notes generation. Add more advanced tools as a fast follower. 
  • Make changes early using feedback and data, and make sure the process also supports successful AI use, not just the technology. 

Example: One customer waited to enable chat-based suggestions until they had cleaned up their KB and tuned their tagging. When they rolled it out, the agent's experience was much better. 

 

Architecture Recommendations

 

Before You Launch 

  • Ensure HR Agent Workspace is upgraded, and other required plugins are installed and up-to date. This ensures your HR team has access to the latest user interface and AI-powered tools that improve productivity and case resolution speed. 
  • Index your HR knowledge bases with accurate and complete metadata. This will help AI understand and retrieve the right content, which improves the quality of recommendations across the module 
  • Avoid customizing core HR tables unless necessary. Customizations can break AI features, complicate upgrades, and reduce compatibility with future ServiceNow releases. Consider configuration over customization wherever possible. 
  • Use ServiceNow’s diagnostic tools to identify and fix any misconfigurations, plugin issues, or data integrity problems before going live. This helps ensure a smooth launch. 
  • Test summarization, recommendations and other capabilities using real HR cases. Validate that the AI is providing accurate, relevant, and helpful outputs by running it against actual HR scenarios. This helps build trust and ensure readiness. 
  • Turn on Now Assist Guardian for monitoring. Enable this feature to track how AI is being used, detect anomalies, and ensure compliance with your organization’s policies. 
  • Organizational Change Management (OCM) is a key part of any successful rollout. Plan your OCM activities early and follow through to make sure everyone understands the change and adopts it fully. 
  • Establish a governance model that defines ownership, decision-making processes, and review cycles for AI configurations, knowledge content, and user feedback. This ensures long-term success and accountability. 
  • Identify areas for use of Result Improvement Rules (RIR) to help boost relevant content when necessary.  

 

After You Launch 

  • Customizations that override or conflict with AI components can cause unexpected behavior. Design your future implementations to not customize out of box components associated with AI functionality. This makes maintenance and upgrades easier. 
  • Keep your internal upgrade plans aligned with ServiceNow’s AI updates by staying on top of release schedules, plugin changes, and new features through release notes, community updates, and training. 
  • Maintain up-to-date documentation for all configurations, decisions, and customizations. This helps with troubleshooting and future enhancements. 
  • Continue to gather feedback and monitor usage analytics regularly to determine how users interact with the implemented AI features. Identify what’s working, what’s underused, and where users are struggling. 
  • Outdated or irrelevant articles reduce AI effectiveness. Set a recurring schedule to review and refresh knowledge base content. Encourage teams to contribute new articles, update existing ones, and share insights. A rich, well-maintained knowledge base is the foundation of effective AI. 
  • Continue to train agents and embed an ongoing OCM plan. Make sure agents understand how to use AI tools effectively and that adoption is supported through regular communication and reinforcement. 

 

Final Thoughts 

 

Getting Now Assist for HRSD up and running is only part of the journey. The bigger success comes from making sure people know how to use it, trust it, and have the right content and processes in place around it.  Start small, define success clearly, and build from there. With the right foundation and ongoing support, GenAI can become a valuable part of your HR operations. 

 

If you find this helpful, feel free to share it with your team. If you’ve developed additional practices or faced unique challenges, we welcome your insights to help strengthen the broader ServiceNow community. 

 

P.S. Insights shared here are my own and do not reflect those of my team, employer, partners, or customers. 

 

Comments
b__SilvanaD
Giga Explorer

Excellent article, Swaroop! Very insightful!

Version history
Last update:
‎06-24-2025 01:50 PM
Updated by: