Getting started with AI Data Explorer

olgaredkina · an hour ago

Getting Started with AI Data Explorer (AIDE)

Enable your business users to explore data in plain language — no SQL, no BI training required. This guide covers everything from evaluating fit to enabling, configuring, and rolling out AI Data Explorer to your users.

This guide covers: What AI Data Explorer is and whether it fits your needs · Prerequisites and licensing · Enabling and configuring AIDE · Validating it works · Rolling out to users · Troubleshooting

What is AI Data Explorer?

AI Data Explorer (AIDE) is ServiceNow's AI-powered analytics exploration experience, available as part of Platform Analytics. It lets business users — IT managers, service owners, team leads — ask questions about their ServiceNow data in plain English and receive instant answers as charts, summaries, and follow-up suggestions. No SQL, no report builder, no BI tool required.

Behind the scenes, AIDE uses a semantic layer that maps natural language questions to your ServiceNow tables and fields, then passes that context into a large language model to construct and run the query. Results appear as interactive visualizations with AI-generated narrative and suggested next questions.

💬 Natural language questions
Ask in plain language. Get charts, summaries, and follow-up suggestions in seconds.

🔍 Contextual exploration
Chain follow-up questions using "these" to build deeper analysis without losing context.

👥 Collaborate in real time
Share explorations with colleagues. Co-edit, tag teammates, and build shared analysis together.

⚡ Act on insights
AI-generated action recommendations translate findings into prioritized next steps.

Is it right for your organization?

Before investing time in setup, confirm AIDE fits the questions your users are actually asking.

Where AIDE works well today

IT managers tracking open, overdue, or SLA-breached incidents by team or priority
Service delivery teams reviewing case volumes, aging, or assignment across CSM or HR
CMDB health checks — assets approaching retirement, incomplete configuration items
Ad-hoc operational questions without waiting for a scheduled report or dashboard refresh
Team leads doing weekly reviews across a single data source
Any user who needs a quick data check — current state, trends, counts, or breakdowns — across any ServiceNow table they have access to

Current limitations — all on the roadmap for future releases

⚠️ The following are current boundaries, not permanent ones. Support for each is planned in future releases.

Limitation	Details
Partial indicator support	Performance Analytics indicator data is partially supported. Automated indicators can be queried directly (as of July 2026 store release). Formula and data snapshot indicators are not yet supported.
No text field analysis	AIDE answers through aggregation queries — counting, grouping, filtering. It can't analyze the content of text fields (descriptions, comments, notes) for sentiment or themes.
No causal or predictive questions	"Why is this happening?" and "What will happen next?" are outside current scope. AIDE surfaces what the data shows, not inferences or predictions.
Single query per question	AIDE executes one query per question. Questions requiring two independent queries need to be broken into separate questions. Related-table conditions within a single query (dot-walking, reference fields) work fine.
No definitional questions	"What is the definition of SLA?" will not work — questions must be analytical and answerable with a chart or table.

Supported data sources

🗄️ ServiceNow Tables
All task-extended tables (Incident, Change, Problem, Case…), key CMDB tables, Users, and Groups are included out of the box. Custom and non-task tables can be added manually.

🔗 Database Views
Combine fields from multiple SN tables into one queryable entity. Lets users ask cross-table questions in a single query. Added the same way as regular tables.

🌐 External Data
Data from outside ServiceNow via Zero Copy Connectors (SQL databases, ERP systems). Requires WDF Professional — discuss licensing separately with your account team.

Prerequisites & licensing

Licensing — two paths

Path	What you need	Notes
AI Native package	Any AI Native license (Foundation, Advanced, or Prime)	AIDE is included. Recommended path for new customers.
WDF v2 + AI add-on	Workflow Data Fabric Standard or Professional (Data Fabric Credit Pack) plus any license with AI capabilities (e.g. Pro+ package)	Both components required. External data sources additionally require WDF Professional v2.

If you're unsure which path applies to your instance, check with your ServiceNow account team or review your entitlements in the Customer Service portal.

Technical requirements

Plugin: AI Data Explorer (sn_pa_ai_canvas) — must be installed and active
AI Search: must be active on your instance

User role

Users who need access to AI Data Explorer require the following role assigned to their profile:

🔑 now_assist_explorer_user

This role is required both to use AIDE and to edit shared explorations.

Enabling AI Data Explorer — implementation checklist

Follow these steps in order. The health check in step 4 will catch most configuration issues before you expose AIDE to users.

Install the plugin
Navigate to System Definition > Plugins and install sn_pa_ai_canvas (AI Data Explorer). Confirm the plugin status shows as Active.
Activate AI skills
Go to Now Assist Admin > Skills > Data and Analytics and ensure all skills related to AI Data Explorer and Query Generation features are active.
Enable Record Level Analysis
Within the Analytics Exploration skill settings, enable Record Level Analysis. This improves the quality of AI-generated insights, especially for drill-down questions.
Validate using the Health page
Navigate to Query Generation > Health. Check all of the following before proceeding:
- AI Search is active and running
- LLM connection is active
- Entities and Dimensions counts are greater than zero and indexed — if either shows 0, the semantic layer has not built yet. Wait for the background job to complete before continuing.
Assign user roles
Assign the now_assist_explorer_user role to everyone who needs access. Users without this role will not be able to open or edit explorations.
Run the smoke test
Open AI Data Explorer from the main menu. Ask: "Show me open incidents." You should receive a chart or list result within a few seconds. If you get "Unable to understand," see the Troubleshooting section.

⚠️ Entities/Dimensions showing 0? Check that AI Search is running and that the semantic layer build job has completed. The build runs on a schedule after plugin activation — on a large instance it can take 30–60 minutes on first run.

Configuring for your data

Out of the box, AIDE works well for standard ITSM data. For best results with your organization's specific tables and terminology, a small amount of configuration goes a long way.

Step 1 — Check what tables are already included

Navigate to Query Generation > Entities. If a table is listed and marked Active, it's already in the semantic layer. Task-extended tables (Incident, Change, Problem, Case, and their extensions) are included by default. Check here before adding anything.

Step 2 — Add tables that are missing

For custom tables or tables outside the task hierarchy, use one of two paths:

Via Query Generation: Query Generation > Administration > Semantic Table Config > New
Via System Tables: System Definition > Tables > open the table record > list action "Enable for Query Generation"

After adding, confirm the table appears in Entities with Active = true. Run a test question and check Logs to verify it resolves correctly.

⚠️ Reference fields: If a question uses a reference field — for example, caller_id.department — the referenced table (in this case, Department) must also be added to the semantic layer. Without it, that dimension will not resolve.

Step 3 — Add business terminology with Manual Segments

Segments translate business terminology into specific filter conditions. AIDE auto-generates segments from your existing reports and PA indicators, but you can add custom ones for org-specific terms your users will ask about.

Navigate to Query Generation > Manual Segment Config > New. Set the Name to match how users would phrase it naturally (e.g. "Sev1", "VIP caller", "overdue"), then define the filter condition.

💡 Start with 3–5 manual segments for the terms your users will ask most. Common examples: "Sev1" (priority = 1), "overdue" (due_date < now AND state != resolved), "VIP" (custom field = VIP). These have an outsized impact on first-use accuracy.

Step 4 — Tuning when something resolves incorrectly

The LLM will occasionally make a bad decision on its own — retry the same question a few times before tuning. Only tune when a failure is repeatable. There are three levers:

Entity Description
Wrong table selected or "unable to understand." Add synonyms users actually say. Keep to 1–2 sentences. Example: "IT incidents, outages, service disruptions, and IT support tickets" — not just "Incident table."

Dimension Label / Usage
Wrong field selected or queried incorrectly. Update Label to match org terminology. Add Usage Instructions to teach matching strategy and expansion rules for complex fields.

Manual Segments
Business term maps to a specific filter. Name = what users say. Filter = the condition. Best for fixed terms like "Sev1", "VIP", "at-risk."

ℹ️ Always check ACLs before tuning. If a user gets wrong results because they lack read access to the intended table or fields, it looks identical to a tuning problem. Verify permissions first.

Validating it works

Once AIDE is enabled, ask questions using your own real data — not generic examples. Questions that matter to your users are the best test of whether AIDE is working correctly in your environment.

How to validate a question is working correctly

Ask a question you know the answer to
Start with a simple factual question you can verify — for example, the count of open incidents in a specific group. If the number looks right, the query resolved correctly.
Check the query directly in the exploration
After getting a result, expand the View Source section in the exploration panel. This shows the actual query AIDE constructed. Verify it reflects your intent — the right table, fields, and filters.
If the result looks wrong, check the logs
Navigate to Query Generation > Logs, find your query by user and date, and inspect which Entities, Dimensions, and Segments were matched. This tells you exactly where the mismatch happened — wrong table, wrong field, or wrong filter.
If logs show a configuration gap, tune
Use the guidance in the Configuring for your data section above to adjust the entity description, dimension label, or add a manual segment. Only tune when a failure is repeatable — not on first occurrence.

💡 Test the follow-up pattern too. Ask a question, then follow up with "Of these, how many are [X]?" The word "these" preserves context from the previous result. Without it, each question starts fresh. This is the most common pattern users will rely on — confirm it works before rollout.

Rolling out to users

Access

Assign the now_assist_explorer_user role to users who need access. Users can then find AI Data Explorer via unified navigation (search "AI Data Explorer") or launch it contextually from a list or chart within Platform Analytics.

Setting expectations — share this with your users

The single most important thing to communicate is the mental model: AIDE answers one analytical question at a time, about one data source. Users who understand this have dramatically better experiences than those who expect it to work like a general chatbot.

✅ Works well Simple, specific questions answerable with a single chart: counts, breakdowns, trends, filters by field values.	✅ Chain questions with "these" Use "Of these [records], …" to narrow results and preserve context. Without "these", each question starts fresh.
✅ Cross-table questions Break into multiple sequential questions, or ask your admin to set up a Database View for common cross-table needs.	⏳ Coming later "Why is this happening?", formula indicator-based analysis, and sentiment/text analysis are planned for future releases.

Tips to share with users

Keep questions short and specific — simple questions work better than complex ones
Use "these" in follow-up questions to keep filters active from the previous result
If you get "Unable to understand," try rephrasing with the table name (e.g. "incidents" instead of "tickets")
Try the AI-suggested follow-up questions — they often surface the next useful angle automatically
Save useful explorations as Analytics Documents to refer back to or share with your team

When things don't work

For admins — diagnosing failures

Always start with Query Generation > Logs. Find the entry by user and date, check which Entities, Dimensions, and Segments were matched, then open Query Constitutor Output to see what was constructed. This tells you exactly where in the pipeline the failure happened.

Symptom	Likely cause	Fix
"Unable to understand"	No matching entity found	Improve entity description with synonyms; verify AI Search is running
Wrong table selected	Ambiguous table names	Add Semantic Usage Instructions to the entity to disambiguate
Wrong table (permissions)	User lacks read ACL on intended table	Grant table/field read access — always check ACLs before tuning
Wrong or missing field	Dimension disabled or unclear label	Activate the dimension; update Semantic Label to match user terminology
Referenced field not resolving	Referenced table not in semantic layer	Add the referenced table as an Entity and enable it
Wrong filter applied	Segment mis-matched	Fix segment name/description; disable conflicting segment
No filter when expected	No matching segment; phrase too vague	Create a Manual Segment with natural-language name matching how users phrase it
Health page shows 0 Entities	AI Search not running or build job not complete	Check AI Search status; wait for scheduled build job to complete

💡 LLM variability: The same question may produce slightly different results across attempts. Only tune when failure is repeatable — not on first occurrence. Retry the same question 2–3 times before making any configuration changes.

For users — quick fixes

What happened	Try this
"Unable to understand"	Rephrase with the table name (e.g. use "incidents" instead of "tickets"). Keep the question simple and specific.
Results seem wrong	Try rewording the question. Add context like a time range or specific field value. Ask your admin if a particular term needs to be configured.
My follow-up lost context	Use "these" explicitly — e.g. "Of these incidents, how many are Sev1?" Without "these," each question starts fresh.
A table I know exists isn't found	The table may not be added to the semantic layer yet. Ask your AIDE admin to add it via Semantic Table Config.
I can't access AI Data Explorer	Ask your admin to assign you the `now_assist_explorer_user` role.

Resources & further reading

📖 AI Data Explorer — Official Documentation
Complete product documentation including configuration reference, release notes, and feature details. Updated with each release.
🎬 AI Data Explorer: Explore Better, Decide Faster (Sway)
A concise, visual feature overview covering key capabilities, use cases, and the end-user experience. 15-minute read.
🔧 Semantic Layer Tuning Overview
Detailed guidance on tuning entity descriptions, dimension labels, usage instructions, and segments to improve query accuracy for your organization's data.
💬 ServiceNow Community — Data & Analytics
Ask questions, share configurations, and connect with other AIDE practitioners in the community.

💬 Have a question, found a configuration tip, or hit something not covered here? Reply below — community knowledge makes this guide better for everyone.