What is ServiceNow Now Assist? A Beginner Guide with Real Enterprise Examples

Now Assist in Plain English

If you have used ChatGPT, Copilot, or any AI assistant at home, you already understand the basic idea behind Now Assist. Now Assist is ServiceNow's built-in AI assistant that helps employees and support agents work faster by summarizing information, suggesting responses, generating content, and finding answers — all within the ServiceNow platform they already use every day.

The critical distinction: Now Assist is not a separate chatbot you deploy and maintain. It is embedded directly inside ServiceNow workflows — inside the incident form, the knowledge base, the service catalog, the HR case, the customer service interaction. When an agent opens a ticket, AI suggestions appear right there. When an employee searches the portal, AI-enhanced answers appear automatically. There is no separate tool to log into.

The Simplest Way to Think About It

Think of Now Assist as three AI assistants in one:

Assistant	Who It Helps	What It Does	Real Example
Agent Assist	Support agents (IT, HR, Customer Service)	Summarizes case history, suggests responses, drafts resolution notes, recommends similar past incidents	An L1 agent opens a ticket with 47 comments spanning 3 weeks. Instead of reading everything, they click "Summarize" and get a 5-line summary of the issue, actions taken, and current status in 3 seconds.
Requester Assist	Employees / End users	Helps people find answers and submit better requests through the self-service portal	An employee types "my VPN is not connecting" in the portal. Instead of just showing search results, Now Assist generates a direct answer: "Try these 3 steps first..." with a link to the full KB article.
Creator Assist	Developers and admins building on the platform	Generates code, flows, catalog items, and playbooks from natural language descriptions	A developer types "create a flow that sends a Slack notification when a P1 incident is created" and Now Assist generates the Flow Designer configuration.

Every Now Assist Skill — What It Actually Does

Now Assist is organized into skills — individual AI capabilities that can be turned on or off per module. Think of skills like apps on your phone: you enable the ones you need.

The skills fall into four families. ITSM skills help IT agents: incident, chat, and change summarization, resolution notes generation, KB article generation, and suggested chat replies. HRSD and CSM skills apply the same ideas to HR cases and customer cases, adding CSM-specific capabilities like email reply drafting and voice call summarization. Creator skills help builders: flow generation, code assist, and catalog item, playbook, app, and dashboard generation from natural language.

For the full skill-by-skill breakdown — what each skill does, where it appears in the UI, and how to configure it — see the companion post "Now Assist Skills Deep Dive: Every Skill, Every Module, and How to Set Them Up" on this blog.

How Now Assist Works Under the Hood

Understanding the architecture helps you troubleshoot when AI suggestions are poor and explain to stakeholders why certain things work or do not work.

The Flow of an AI Request

User clicks "Summarize" on an incident
        ↓
Now Assist reads the incident data:
  - Work notes, comments, activity log
  - Related records (tasks, CIs, changes)
  - Knowledge articles linked to the incident
        ↓
ServiceNow constructs a prompt:
  - System instructions + context data + user request
  - Guardrails applied (PII filtering, topic restrictions)
        ↓
Prompt sent to LLM (Large Language Model):
  - ServiceNow's own LLM infrastructure (Now LLM)
  - OR configured third-party model (Azure OpenAI)
        ↓
LLM generates response
        ↓
Response filtered through guardrails:
  - PII detection, harmful content check
  - Relevance verification
        ↓
Summary displayed to user in the Now Assist panel
  - User can accept, edit, or dismiss

Key Architecture Concepts

Concept	What It Means	Why It Matters
Now LLM	ServiceNow's own large language model, hosted on ServiceNow infrastructure	Your data stays within ServiceNow's boundary — no third-party data exposure. This matters for compliance-sensitive industries (healthcare, finance, government).
Skills	Individual AI capabilities that can be enabled/disabled per module	You control exactly which AI features are active. Enable summarization for ITSM but not for HRSD if you want to phase the rollout.
Guardrails	Rules that filter what goes into prompts and what comes out of responses	Prevents the AI from exposing PII (names, SSNs, emails) in generated text and blocks off-topic or harmful content.
Context Grounding	The AI response is grounded in actual ServiceNow data (incident notes, KB articles), not just general knowledge	Reduces hallucination — the AI cites actual records instead of making things up. This is why clean data matters: garbage in = garbage out.
Now Assist Admin Console	The centralized administration hub where you manage all AI settings	One place to enable skills, review usage analytics, configure guardrails, and manage LLM connections.

Setting Up Now Assist — Step by Step

This is the practical setup path for a ServiceNow admin enabling Now Assist for the first time:

Step 1: Verify Your Instance Version

// Navigate to: your-instance.service-now.com/stats.do
// Check "Build name" on stats.do — Australia is the current family.
// Now Assist skills began shipping in earlier releases and expand each family
// (Skill Kit, AI Agents, AI Control Tower). Confirm which skills ship on your build
// via the release notes for that family.

Step 2: Open the Now Assist Admin Console

// Navigate to: All > Now Assist > Now Assist Admin
// This is the central hub for all AI configuration.
//
// What you will see:
// - Dashboard: usage metrics, active skills, model status
// - Skills: list of available skills by module (ITSM, HRSD, CSM, Creator)
// - Settings: plugin management, LLM configuration, data sharing controls
// - Analytics: how often skills are used, acceptance rates, feedback

Step 3: Install the Now Assist Applications

// Now Assist capabilities ship as applications, not classic plugins.
// Install the one for each workflow you own:
//   Now Assist for ITSM     → incident summary, resolution notes
//   Now Assist for HRSD     → HR case summary
//   Now Assist for CSM      → customer case summary, email reply
//   Now Assist for Creator  → code assist, flow generation
//
// Install and update them from the ServiceNow Store, or on the
// instance via System Applications > Application Manager.
// The Now Assist Admin console manages them once installed and licensed.

Step 4: Enable Your First Skills

// In the Now Assist Admin Console:
// 1. Go to the Skills section
// 2. Find "Incident Summarization" (a good first skill — low risk, high visibility)
// 3. Click to enable
// 4. Configure:
//    - Which roles can use it (e.g., itil, admin)
//    - Whether it auto-runs or requires user click
//    - Any additional guardrail settings
// 5. Save and test with a sample incident

Step 5: Test and Validate

// Open an incident with a substantial work note history
// Look for the Now Assist panel (usually right side of the form)
// Click "Summarize"
// Review the generated summary:
//   - Is it accurate?
//   - Does it capture the key points?
//   - Does it expose any sensitive data it shouldn't?
// If satisfied, expand to more users; if not, check data quality.

Where Teams Get Confused — Common Misconceptions

Misconception	Reality
"Now Assist will fix our broken processes"	AI amplifies what you already have. If your incident categorization is a mess, AI suggestions will be a mess. Fix processes first, then add AI.
"We need to train the AI on our data"	Now Assist uses pre-trained models plus your instance data as context. You do not train a custom model — you configure skills and ensure your data (KB articles, work notes) is clean and complete.
"Now Assist replaces L1 support agents"	It augments them. Agents still review, approve, and edit AI-generated content. Now Assist handles the tedious parts (reading 50 comments, drafting notes) so agents can focus on judgment and empathy.
"AI suggestions are always accurate"	They are usually helpful but not always correct. That is why every AI output has an accept/edit/dismiss option. Treat AI suggestions like a junior colleague's draft — review before using.
"We can enable everything at once"	Phased rollout is critical. Start with 1-2 high-volume, low-risk skills (like incident summarization), measure results, then expand. Trying to enable all skills on day one overwhelms users and makes it impossible to measure impact.

The Data Quality Foundation

This is the single most important factor in Now Assist success. The AI generates responses based on what is in your ServiceNow data. If your data is poor, the AI output will be poor.

Data Area	What "Good" Looks Like	What "Bad" Looks Like	Impact on Now Assist
Work Notes	Detailed, timestamped notes describing what was done and why at each step	Empty or one-word notes like "done" or "fixed"	Summarization is only as good as the notes. No notes = nothing to summarize.
Knowledge Base	Well-structured articles with clear steps, updated regularly, covering common issues	Outdated, incomplete, or nonexistent KB articles	KB Generation and Search Assist depend on existing knowledge quality. AI cannot create good articles from nothing.
Categorization	Consistent use of categories and subcategories with clear definitions	Incidents miscategorized or left as "Other/General"	AI recommendations for routing and resolution depend on pattern matching against historical data. Bad categorization = bad patterns.
Resolution Notes	Clear description of what fixed the issue and what the root cause was	Copy-paste of the short description or "Resolved per user"	Resolution Notes Generation learns from past closures. If past closures are garbage, generated notes will be too.

Measuring Success — What Leadership Cares About

Metric	What It Measures	How to Track in ServiceNow	Target Benchmark
Average Handle Time (AHT)	Time agents spend per ticket	Performance Analytics indicator on incident table	15-30% reduction after Now Assist adoption
First Contact Resolution (FCR)	Percentage of tickets resolved on first interaction	Report: resolved_at within first assignment period	5-10% improvement
Self-Service Deflection	Tickets avoided because users found answers in portal/KB	Virtual Agent + KB feedback analytics	20-40% deflection rate for enabled categories
Agent Satisfaction	How agents feel about the AI tools	Post-shift surveys, feedback button on Now Assist panel	70%+ positive feedback
AI Acceptance Rate	How often agents accept AI suggestions vs. dismiss them	Now Assist Admin Console analytics	Above 60% suggests good quality; below 30% suggests data quality issues
Knowledge Article Coverage	Percentage of common issues covered by KB articles	Map top 50 incident categories to existing KB articles	80%+ coverage for top categories

Implementation Roadmap — Phased Approach

Phase	Duration	What You Do	Success Criteria
Phase 0: Foundation	2-4 weeks	Audit data quality. Clean KB articles. Standardize work note practices. Baseline current metrics (AHT, FCR, deflection).	Top 20 incident categories have KB articles. Work note compliance above 80%.
Phase 1: Pilot	4-6 weeks	Enable 1-2 skills (start with Incident Summarization and Resolution Notes) for a pilot group of 10-20 agents.	AI acceptance rate above 50%. Agents report time savings. No data quality issues flagged.
Phase 2: Expand	6-8 weeks	Enable for all ITSM agents. Add Chat Summarization and KB Generation. Start CSM or HRSD pilot if applicable.	AHT reduced 15%+. FCR improved. Positive agent feedback 70%+.
Phase 3: Optimize	Ongoing	Review analytics weekly. Tune guardrails. Enable Creator skills. Explore AI Agents for automated resolution. Feed learnings back into data quality improvements.	Continuous improvement in metrics. Expanding to more modules and use cases.

Now Assist vs. Other AI Tools — Why Not Just Use ChatGPT?

	Now Assist	ChatGPT / Copilot (External AI)
Data access	Reads your incident data, CMDB, KB, and user records directly — no copy-paste needed	You must manually copy data into the AI, risking PII exposure and losing context
Context	Knows the specific record, the related CIs, the assignment group, the SLA status	Has no awareness of your ServiceNow data unless you provide it
Integration	Actions happen inside ServiceNow — generated notes go directly into the field, generated flows deploy immediately	You must copy AI output and paste it back into ServiceNow manually
Governance	Guardrails, audit trails, role-based access, data stays within ServiceNow boundary	No guardrails — data sent to external servers, no audit trail, potential compliance violations
Compliance	Designed for enterprise compliance (SOC 2, HIPAA, FedRAMP depending on instance)	May violate data handling policies if you paste ticket data into external AI