ServiceNow Now Assist Consulting

What a Now Assist Consulting Engagement Covers

A Now Assist consulting engagement with Oblytech focuses on implementing ServiceNow’s AI-powered conversational assistant into your IT support operations. Now Assist helps support agents by suggesting relevant knowledge articles, common resolutions and automation actions while they handle incidents. We assess your current support process, knowledge maturity and tool integrations. We identify where AI assistance would reduce handle time and improve accuracy: first-line agents searching for solutions, senior agents with extensive domain knowledge who could mentor others, or complex incidents where multiple tools and data sources must be coordinated.

Implementation includes configuring Now Assist to understand your service model and common issue categories, training it on your knowledge base and standard procedures, and integrating it with systems agents already use. We measure impact through agent productivity metrics and customer satisfaction improvements. The engagement length depends on scope and readiness, from assessment to productive use.

Typical Starting Problems and Opportunities

Support teams often spend significant time searching for solutions. A new agent asking how to reset a user password might search your knowledge base with vague terms and find 50 articles, reading through them to find what applies to their situation. An experienced agent knows the procedure but has to context-switch away from the incident to document it for others. Knowledge exists but is fragmented and hard to discover. Now Assist changes this by presenting relevant knowledge and procedures in context while agents work on incidents.

Another opportunity is consistency. Different agents resolve the same issue differently because they learned different approaches or forgot documented procedures. Now Assist surfaces the recommended approach, so agents follow it and customers get consistent resolution. Escalations decrease because agents follow procedures more consistently and understand which incidents require specialist involvement.

Knowledge Base Preparation and Optimisation

Now Assist learns from your knowledge base. Articles must be well-written and findable. We audit your knowledge base to identify gaps, outdated content and inconsistent structure. We work with knowledge owners to fill critical gaps: how-to articles for common issues, troubleshooting procedures, decision trees for triage. Articles are tagged and categorised so Now Assist can understand which article is relevant to which situation.

We also prepare ServiceNow configurations that Now Assist uses as context. The service model must be clear so Now Assist understands which services agents support. Incident categories and types must be defined consistently so Now Assist maps incoming incidents to the right knowledge and procedures. CI relationships in the CMDB help Now Assist understand which components might be affected by incidents.

AI Assistant Configuration and Personalisation

We configure Now Assist to understand your organisation’s common terminology and processes. If you call a particular system by a nickname internally, we teach the assistant to recognise it. If your incident resolution process follows a standard triage flow, we configure that flow in Now Assist so it suggests the right steps. We integrate Now Assist with incident forms so it populates fields based on agent input and incident context.

Personalisation extends to individual agents. Experienced agents see different suggestions than new agents. Suggestions respect agent skills and preferences. An agent who specialises in network issues gets network-focused suggestions; a database specialist gets database-focused articles. This personalisation makes Now Assist more helpful and agent-specific rather than generic.

Integration with Support Tools and Workflows

Now Assist works best when integrated with tools agents already use. We configure it to appear in the incident form so agents see suggestions while working on incidents. We integrate it with your CMDB so it accesses current asset and configuration data for context. We connect it to your monitoring systems so incident suggestions include relevant metrics and logs. Automation suggestions show agents one-click actions: run a script, reset a password, restart a service, all from the Now Assist interface without switching to another tool.

Escalation workflows are improved with Now Assist. When an agent cannot resolve an incident, Now Assist suggests when to escalate and to which team based on incident type and complexity. Escalation handoff includes relevant context the Now Assist gathered during the initial interaction, reducing investigation time by the specialist team.

Measuring Success and Adoption

Key metrics for Now Assist success include mean time to resolution (MTTR), first-contact resolution rate, and agent satisfaction. Agents using Now Assist suggestions typically close incidents faster and with higher confidence. We track adoption: are agents using Now Assist suggestions or ignoring them? If usage is low, we investigate: suggestions may be irrelevant, the interface may be intrusive or agents may not trust AI recommendations. Feedback drives tuning to improve suggestion quality.

Customer satisfaction often improves as resolution quality and consistency improve. Repeat incidents decrease because agents follow documented procedures more consistently. Documentation improves because Now Assist’s use of knowledge base highlights gaps, driving agents and knowledge owners to fill them.

Now Assist learns from every interaction, becoming more effective over time. Feedback from agents who rate suggestions helps tune the model. Popular suggestions are working; unpopular ones might need tuning or removing. Your knowledge team continues to add and improve articles, and Now Assist becomes better at recommending them.

Deciding where agent suggestions are allowed to act

The useful question about an AI assistant in an ITSM queue is not whether it can suggest a fix. It is which kinds of suggestion an agent may act on without a second check. A suggestion to read a knowledge article is low risk. A suggestion to run a remediation script on a production server is high risk, even when it is usually correct. Before any configuration starts, the service owner should sort the suggestion types the team expects to see and decide the level of human control for each. Governance for these controls is set out in the Now Assist for ITSM governance model.

Confidence in a suggestion is not the same as correctness. A suggested resolution can read well, cite a relevant article and still be wrong for the configuration of this particular service. The control is therefore not a confidence threshold. It is a rule about what the agent must verify, and that rule differs by the consequence of acting on the suggestion. Access restrictions for assistant content are covered in the Now Assist for ITSM security and access guide.

Preparing the service model and knowledge before switching suggestions on

An assistant is only as specific as the data behind it. Before suggestions go live, check the inputs the assistant reads: the service model, the incident categories and types, the links between incidents and configuration items, and the knowledge articles it can draw on. Many early weak suggestions trace back to one of these, not to the assistant itself. Knowledge accuracy, a common source of weak suggestions, is covered in a guide to knowledge quality review.

1. Confirm that categories and types are used consistently, and retire any category that agents use interchangeably with another.
2. Check that configuration item relationships on the main services are present and current, so suggestions about affected components are not built on stale links.
3. Review the articles the assistant is most likely to cite for the top topics, and fix the ones that would give the wrong step.
4. Build a test set from recently closed incidents, with the resolution the agent actually used, and compare the suggestions against it before any agent sees them.
5. Record the results of the test set as the baseline, so later tuning can be judged against something.

Measuring whether suggestions help or distract

Adoption is the first thing to measure and the easiest to misread. High acceptance can mean agents trust the assistant, or that the easiest suggestion is always the one clicked. Look at reclassification after acceptance, reopen rate for resolved incidents where a suggestion was used, and the number of times agents dismiss a suggestion and then do the same thing a different way. Dismissals carry information: a topic with consistently dismissed suggestions needs either better inputs or a decision to switch suggestions off for that topic. Metrics that separate real help from noise are in the Now Assist for ITSM KPI guide.

Does the assistant need access to every incident?

It needs access to the records it is meant to learn from and act on. Limit read access by group where the incidents contain restricted information, such as security investigations, and check that the restriction holds in the suggestions as well as in the record view.

Should agents be able to turn suggestions off themselves?

Allow it per topic rather than per agent. A personal switch hides the signal that a topic is failing, while a topic-level switch gives the service owner a clear lever and a clear record of what was turned off and why.

How do we handle a suggestion that cites an article we know is out of date?

Treat it as a content defect. Retire or correct the article, then check whether the same article is cited elsewhere. The fix is in the knowledge, not in the suggestion logic.

Who approves a change to what the assistant may automate?

The service owner approves the list, and the change manager confirms that each automation action has a rollback path. Keep the list short at first and widen it only after the reopen rate for those actions is stable.

When to hold off on assistant suggestions

If the categories and service model are still being rebuilt, suggestions will be built on moving targets and the team will stop trusting them early. If the knowledge base is largely out of date, the assistant will surface outdated steps with confidence. In both cases the earlier work is data and knowledge quality. Once the test set shows acceptable suggestions on the top topics, the assistant can be switched on for those topics first. Support arrangements for the assistant are described in the Now Assist managed support service.

The suggestion controls are easier to scope before anyone sees a live suggestion.