ServiceNow Now Assist for ITSM: Guides and Services

What Now Assist for ITSM Covers on the Now Platform

Now Assist for ITSM is a generative AI capability embedded in Incident Management, Problem Management, and other ITSM modules. It helps incident responders and service desk agents resolve issues faster by providing real-time suggestions, knowledge article recommendations, and automatically drafted responses. Now Assist learns from your historical incidents and knowledge base to understand your environment, your common issues, and what works for your organization. When an agent logs a new incident, Now Assist can recommend similar past incidents, suggest knowledge articles that might solve the issue, and even draft an initial diagnosis or resolution steps for the agent to review.

Now Assist reduces mean time to resolution (MTTR) by providing agents with expert recommendations, accelerates knowledge search so agents find relevant articles without complex queries, and improves first-contact resolution by suggesting solutions before escalation.

How Now Assist Relates to Other ITSM Modules

Now Assist works across Incident Management, Problem Management, Change Management, and Knowledge Management. In Incident Management, it suggests solutions based on similar incidents and knowledge articles, enabling faster resolution. In Problem Management, it helps identify patterns in recurring incidents and suggests root-cause investigation paths based on historical data. In Change Management, it can recommend similar past changes and help impact analysis. In Knowledge Management, it improves article search so agents find answers faster without needing exact keywords. Now Assist amplifies the effectiveness of these modules by putting expertise at the agent’s fingertips in real time.

How to Use this Section

Assess your readiness for Now Assist. The tool learns from your historical data: if you have well-categorized incidents and maintained knowledge articles, Now Assist will be more effective. If your historical data is sparse or poorly structured, you may need to clean up and enrich your knowledge base first. Start with Incident Management, where the volume of incidents provides the most learning data. As your data quality improves, add Now Assist to Problem Management and Change Management.

Implement Now Assist in phases. Phase one: enable agent-assist features so agents see recommendations while handling incidents, improving their efficiency. Phase two: enable automated knowledge search so relevant articles are surfaced without requiring exact keyword matches. Phase three: use Now Assist analytics to identify gaps (e.g., common incident types with no knowledge articles) and prioritize knowledge-base improvements. Phase four: monitor Now Assist effectiveness and gather agent feedback to improve recommendations.

Typical Starting Problems

Service desk agents often spend time searching for relevant knowledge articles or trying to remember how similar incidents were resolved in the past, extending resolution time. Complex environments with many systems and configurations make it hard for new agents to gain expertise quickly. High-volume ticket environments make it difficult to meet SLAs when agents are context-switching between systems. Now Assist addresses these challenges by putting historical incident knowledge and curated articles at the agent’s fingertips in real time, reducing search time and improving consistency.

What Oblytech Provides for Now Assist

Oblytech helps organizations assess readiness for Now Assist, prepare and enrich historical incident and knowledge data, configure Now Assist features aligned to your ITSM processes, and train agents on how to use and trust AI recommendations. We help you monitor Now Assist effectiveness (agent adoption, MTTR improvement, first-contact resolution improvement) and iteratively improve by adding more training data or refining knowledge articles based on actual usage patterns.

Data quality is critical for Now Assist effectiveness. Ensure incidents are consistently categorized with correct short descriptions and detailed descriptions. Now Assist learns from this historical data; garbage in means garbage out. Invest time in cleaning up historical incident data before enabling Now Assist. Remove duplicate incidents and consolidate similar incidents so the training data is clean.

Now Assist improves over time as more incidents are added and feedback is collected. Monitor which recommendations agents accept vs. reject. If agents consistently reject Now Assist suggestions, the model may need retraining or the recommendations may not be suited to your environment. Gather feedback and use it to iteratively improve Now Assist recommendations.

Sorting possible uses by what could go wrong

A generative assistant can help with many ITSM tasks, but the tasks do not carry the same risk. A suggested knowledge article that an agent reads before answering is low risk, because a person still decides what to send. A drafted resolution that is written straight onto a record, or a routing suggestion that moves work between teams, carries more consequences. Sorting candidate uses by the cost of a wrong answer is a useful first step, and it often reorders the wish list. The table below sets out how to group the common uses.

Deciding the order of adoption

The order matters more than the feature list. Start with uses where a person sees and judges every output, because those give you the evidence you need to widen the scope. Automatic actions come last, and they should come with a documented rollback. This sequence also builds the trust that agents need, since they learn how the assistant behaves before it touches their work.

1. Pick one queue with a stable volume and an experienced supervisor who will judge outputs.
2. Enable summarization or article suggestions, and record how often agents accept, edit or ignore each output.
3. Read a handful of outputs every week alongside the supervisor, and write down the failure types you see.
4. Fix the failure causes that sit in your data or knowledge base before adding a new use.
5. Only then consider recommendations that change fields, and enable them one field at a time.

Signals that the adoption is healthy

• Agents edit the output but keep using it, which can show the assistant is useful even when imperfect.
• Supervisors can explain why a given output was wrong, which means the failures are being categorized rather than ignored.
• The share of outputs that agents reject falls over time for the same use, after knowledge fixes are made.
• Escalations caused by a wrong suggestion are traceable to a specific cause, not a general sense that the tool is unreliable.

If these signals are missing, the response is often to narrow the scope rather than widen it. A smaller set of uses that agents trust can outperform a broad set they have stopped reading.

When to hold off on the assistant

Some environments should not start with generative features at all. Where the knowledge base is badly out of date and nobody owns it, every suggestion can be a guess dressed as an answer. Where regulated or sensitive records are the main workload, the data-handling review has to finish first, and the assistant’s output needs the same controls as any other record change. Where the service desk is about to change tooling, it is often wiser to finish that move before adding a new layer.

Questions to settle before a pilot

Settle these questions in writing so the pilot can end with a clear decision rather than a general impression. Who may switch a use on or off? Which records and fields may the assistant read? What does a supervisor do with a bad output, and where is it logged? What result would make you stop, and what result would justify the next use?

Does the assistant replace the knowledge team?

No. It depends on the knowledge team to keep articles accurate and current. If anything, the assistant makes gaps in the knowledge base more visible, because agents see the missing answer immediately.

Will changing categories break what the assistant learned?

It can. A category restructure changes the patterns the assistant relies on, so schedule the change, record the date, and re-run the advisory comparison afterwards rather than assuming the old results still hold.

Should agents see the same answers as managers?

Verify this in testing rather than assuming it. The answers should respect the access a user already has to records and knowledge, and a test with a restricted account is the only reliable way to confirm that.

Can we run the pilot without changing any records?

Yes, and for a first pilot it is the safer choice. Read-only suggestions still show whether the outputs are useful, and they keep the audit trail simple while the team learns.

How long should a pilot run before it is judged?

Long enough to cover the normal variety of work in the chosen queue, including at least one peak period. Set the length with your service owner rather than with a standard duration.

Next step

The sorting table above is a practical place to begin. Read the note on knowledge quality reviews, the related guides on the plain-language overview of the feature set and on knowledge management for the data side of the decision.