Unlocking Customer Success with AI-Powered Search

[Esha Kinra1]

From Frustration to Fluidity 

Search sits at the heart of every self-service experience. Inefficient search leads to user frustration, redundant support tickets, and reduced productivity. For a global platform like Now Support, enhancing search directly impacts case deflection, user satisfaction, and operational efficiency. 

Now Assist in AI Search transforms static keyword lookup into a contextual, AI-driven retrieval delivering precise, actionable answers without users sifting through irrelevant documents. 

Generative AI Meets Enterprise Search 

Traditional search engines rely primarily on keyword matching and relevance ranking. Now Assist in AI Search elevates this by combining semantic vector search with large language model (LLM) reasoning to generate contextual answers sourced from verified enterprise knowledge. 

Key technical improvements include: 

• Unified Results: Knowledge Articles, Catalog Items, and Virtual Agent topics are aggregated into a single, semantic-aware response layer. 

• Contextual Summarization: The system parses article sections, attachments, and reusable knowledge blocks to produce concise summaries. 

• Hybrid Ranking: Keyword precision and semantic similarity are combined for optimal relevance. 

• Governance & Transparency: Responses are drawn only from approved content and labeled Powered by Now Assist. 

For employees and customers, it feels like asking an expert and getting the right answer right away.  

Building the AI Search Foundation 

A reliable AI search experience depends on high-quality data, structured configuration, and ongoing governance. 

1. Prepare Data Sources 

• Identify tables and fields to be indexed (Knowledge Articles, Catalog Items, Virtual Agent topics, etc.). 

• Validate content, archive outdated material, and ensure clarity. 

• Configure semantic index settings for optimal AI retrieval. 

2. Create Semantic Index 

• Define a semantic index linking to each data source and its attributes. 

• Apply chunking strategies to segment long articles into smaller passages for accurate retrieval. 

• Chunking Configurations: 

• Chunking unit: Divide text by words or sentences. 

• Chunk size: Passage length (default 250 words or 15 sentences). 

• Expanded snippet size: For Small-to-big chunking, determines added context. 

• Chunking Strategies: 

• Fixed-Size Chunking: Splits text into uniform passages; simple and effective for most content. 

• Small-to-Big Chunking: Selects top relevant chunks, expands context, and re-divides for indexing; ideal when key info spans multiple sections. 

Fine-tuning these parameters balances context and precision, improving retrieval performance. 

3. Set Up Search Profile 

• Link semantic indexes with retrieval scripts in the Genius Result. 

• Configure typo handling, result improvement rules, synonyms, stop words, and tuning parameters. 

• Enable multi-content experience for aggregated results across knowledge sources. 

4. Indexing and Testing 

• Trigger indexing for all semantic sources. 

• Validate queries against a Golden Set of representative queries to confirm correct retrieval. 

5. Optimize and Maintain 

• Monitor AI search responses via dashboards. 

• Refine indexing, chunking, and result improvement rules as content evolves. 

• Re-index after major updates to maintain accuracy. 

6. Continuous Improvement 

• Track Search Relevancy, performance, “no result” queries, and content gaps. 

• Iterate on indexing, tuning, and content promotion to improve precision, recall, and user satisfaction. 

Enabling Now Assist in AI Search 

A thoughtful rollout combines precision with momentum. Here’s a tested framework you can adapt for your organization: 

Phase 1: Pilot and Fine-Tune 
Start small, with one portal and a defined knowledge base.  Configure your Search Profile using synonyms, stop words, and tuning rules to refine precision. Use passage chunking to divide longer articles into smaller, meaningful sections that help the model produce sharper answers. Build a Golden Set of 20 to 50 common queries with desired outcomes and use it as a benchmark each time you fine-tune configurations.  

Phase 2: Expand and Optimize 
Once the pilot performs well, gradually add new content sources and user groups. When connecting external repositories like SharePoint or Confluence, ensure data synchronization schedules and permissions are current to avoid duplicates or stale results. 
Use analytics dashboards to identify search gaps, low-performing content, and “no result” trends. Then refine and promote the articles that deliver the most accurate answers. 

Phase 3: Govern and Scale 
Establish clear ownership for AI Search configuration and governance. Use the dashboards to track performance, identify regressions, and prioritize improvements. Treat Now Assist as a living system that learns from how people use it. Regular reviews, golden set testing, and data-driven updates will keep it accurate and reliable at scale.  

You can find more detailed setup guidance in the official ServiceNow documentation here 

Measurable Impact 

Since introducing Now Assist in AI Search on Now Support, we have seen measurable improvements that demonstrate both technical and strategic value. The Search Success Rate increased by 15 percentage points, preventing roughly 16,000 support cases and delivering an estimated $1.36 million in annual cost avoidance.  

These results reflect disciplined engineering with clean data, careful tuning, and analytics-driven iteration. Now Assist transforms knowledge into a governed, self-improving asset, delivering authoritative answers quickly while maintaining compliance. Users find information faster, support teams focus on complex issues, and enterprise knowledge adapts to usage. 

Key Lessons and Takeaways 

From our experience, organizations can implement AI search responsibly and effectively by focusing on three principles: 

• Make search serve solutions, not links: Generative AI can cut through the clutter and bring customers directly to the right answer. 

• Ground responses in trusted knowledge: Retrieval-augmented generation ensures answers are accurate, reliable, and context-aware. 

• Measure what matters: Tracking Search Success Rate gives us a clear, customer-centric view of impact and shows just how powerful these changes are. 

What’s Next 

Search will continue to be a core part of digital services. With Now Assist in AI Search, we have improved how information is found, delivering answers that are fast, reliable, and easy to understand, all based on trusted enterprise knowledge. In the future, we plan to enhance contextual understanding, integrate more content sources, and support multiple languages while keeping accuracy and transparency at the forefront. When used responsibly, AI Search acts as an intelligent layer that adapts over time, respects policies, and helps the organization make the most of its knowledge.