For many ITSM leaders, the first question on the path to AI adoption is a familiar one: “Is my data ready?”. It’s an understandable concern-after all, traditional machine-learning systems demanded massive datasets and rigorous data hygiene before any value could be realized. But in the era of ServiceNow’s Now Assist and agentic AI, the rules have changed. Yes, data matters. But you need far less than you think, and data alone is not what determines whether your AI initiatives succeed. After personally working with early AI deployments across dozens of ITSM organizations, one theme stands clear: Your understanding of your processes, personas, and workflows is just as important as the data you provide.
This blog outlines what data actually matters, what misconceptions to let go of, and how to prepare both your information and your workflows to accelerate value from Now Assist for ITSM and AI Agents.
Why You Need Data Foundations for AI (and What “Good Enough” Actually Looks Like)
AI models, whether generative or deterministic, need to understand one core concept:
What does “correct” look like in your organization?
This comes from the examples you already have in your system:
- Resolution notes showing how issues were solved
- Knowledge articles capturing the “known good” solutions
- Categorization and assignment patterns
- Policies, runbooks, and other procedural guidance
These artifacts create the guardrails that help AI ground its reasoning in real organizational context. Without them, an LLM can still attempt to answer (and often answer well), but it may hallucinate a category, fabricate missing details, or propose resolution steps that never existed.
The good news: You don’t need thousands of pristine examples for the AI to start performing.
Why the Amount of Data You Need Has Changed in the Age of Agentic AI
Under older ML approaches-like Predictive Intelligence-customers needed 10,000–30,000 labeled records to train a regression model. That meant long onboarding cycles, manual cleanup, and “data readiness” becoming a blocker for every new field you wanted to predict.
Agentic AI has flipped this model. Today’s Agentic and reasoning based LLM AI can:
- Reason even with zero examples
- Infer patterns by reading your knowledge and recent incidents
- Use similarity AI to retrieve and learn from 4–5 related cases, not tens of thousands
- Generalize across workflows using foundational LLM intelligence
This shift dramatically lowers the barrier to entry. Customers routinely go live with messy, partial, or imperfect data and still see immediate ROI with use cases like incident triage agents or incident summarizer skills. The requirement isn’t “big data.” It’s representative guidance-just enough examples for the AI to understand your norms.
Process Mapping Matters Just As Much As Data
This is the most overlooked insight for IT leaders beginning their AI journey.
When AI acts-classifies an incident, recommends a resolution, or executes a multi-step workflow-its effectiveness depends heavily on your process clarity, not just your historical data.
Why Process Mapping Is Critical
AI needs to know:
- Which decisions matter
- What inputs are required
- What a “good” output looks like
- Where to hand off to humans
- Where automation can safely execute
Teams that invest time upfront in mapping their workflows consistently see higher accuracy, more reliable automation, and faster time-to-value.
Use Personas + Workflows to Define Success
A highly effective technique is running workshops around:
- Target personas (Service Desk Agent, Change Manager, Network Ops Analyst)
- End-to-end workflow steps
- Inputs → decisions → outputs
- Steps that can be offloaded to AI
This also feeds into the concept of Agent Evaluations (“Agent Evals”)-PM-defined rubrics for what good AI responses should contain:
- Length
- Required fields/data
- Tone
- Format
- Decision logic
Good AI results start with good definitions of “good.”
The Minimal Data You Actually Need
Below is a realistic breakdown of what’s strongly recommended, nice-to-have, and not necessary when beginning your implementation journey with Now Assist for ITSM.
- Task Resolution Data (Most Important)
These are your gold mines for grounding LLM outputs.
Strongly Recommended
- Incidents + Resolution Notes
- Assignment Groups + Priorities
- Knowledge Articles (can also be external via content connectors)
- Alerts + Configuration Items (CIs)
- IT Service Catalog items + Record Producers (if relevant)
Why? This is the material AI uses to pull similar cases and recommend accurate resolutions.
- Operational Data (Helpful, Not Required)
- Change records with test/backout plans
- Operational logs or alerts
These enhance reasoning for AI Agents, especially in ITOM/ITOps scenarios, but they are not blockers to starting.
- Foundational Data (Simple but Valuable)
- Assignment Group descriptions
- Updated list of Configuration Items in CMDB
- Business rules, priority matrices, routing logic for deterministic logic
This data ensures the AI doesn’t infer deterministic things you already know precisely.
If you have clear logic, don’t make the LLM guess it.
The great news is that as you adopt GenAI skills like resolution note generation and KB generation, Now Assist AI will help provide a flywheel of context to improve future results more quickly by encoding the learning and information.
The ITSM AI Adoption Journey: Crawl → Walk → Run
Most customers progress naturally across four phases:
- AI Search
Let users and agents retrieve knowledge and historical incidents conversationally. - LLM Now Assist Virtual Agent
Handle inquiries, triage, and simple requests with generative chat. Focus on prebuilt, OOTB topics and agents. - GenAI-Assisted Workflows (Skills)
Draft incidents, summarize tickets, recommend solutions, write change plans, etc. Focus on prebuilt, OOTB skills. - AI Agents (Agentic Workflows)
Full multi-step automations that classify, diagnose, execute, and close the loop. Focus on activating and adjusting the triggers of prebuilt, OOTB agentic workflows.
Across all phases, one principle holds: OOTB First, Custom Second
You get faster time to value, lower maintenance, and higher-quality responses by starting with the out-of-the-box Now Assist capabilities and only customizing once you understand your patterns and opportunities. We've configured common use cases OOTB and they will be on by default for entitled customers - from skills to agents across incident and request fulfillment. That means day 1 you can take advantage of preconfigured ai search, skills, and agents for IT. For each AI type, you can customize and extend it to personalize it and apply it to additional organizational processes and bespoke data but we recommend doing so only after evaluating OOTB.
So, What Should ITSM Leaders Actually Do to Prepare implementing Now Assist AI for ITSM?
Here is the short list of actions that create outsized impact:
- Ensure your resolution notes and knowledge base reflect reality. They don’t need to be perfect-just clear, consistent, and present.
- Document your workflows and decision points. AI is only as effective as the process it’s supporting.
- Focus on persona-based journey mapping. Who does what, why, with which inputs?
- Identify where AI can offload cognitive or repetitive steps. Start with triage, classification, summarization, and solutioning.
- Invest in clarity, not volume. AI does not need 30,000 examples. It needs meaningful examples.
Conclusion: You’re More Ready Than You Think
Most ITSM organizations dramatically underestimate how prepared they already are for AI.
If you have:
- A functioning incident process
- A set of knowledge articles (even if imperfect or externally sourced)
- Resolution notes that show how work gets done
- Clear personas and workflows
…then you have everything you need to begin extracting value from Now Assist for ITSM and AI Agents.
Even if you don't have these items, generative skills can begin generating them on your behalf. The organizations realizing the fastest time-to-value are not the ones with the cleanest data-they are the ones with the clearest workflows and the courage to start small and learn quickly. AI is no longer a destination; it’s an operating model. And the sooner your ITSM organization invites Now Assist AI into its processes, the sooner you’ll see measurable impact on speed, accuracy, MTTR, and employee experience.