Now Assist

Implementing Now Assist: What Actually Works in Production

A practical guide to Now Assist implementation — which features deliver real value, activation requirements, the configuration that avoids common pitfalls, and honest expectations.

Now Assist is ServiceNow's AI layer built on a combination of their own LLMs and third-party models. Since Washington DC, it has been available in most ITSM-licensed instances. Since Xanadu, it's production-recommended. This guide covers what to actually implement, in what order, and what to expect.

Activate Before You Configure

Now Assist requires the Now Assist SKU (license) and activation of the Now Assist for ITSM plugin. Before doing anything else, verify:

  • Now Assist license is active on your account
  • Plugin com.glide.now_assist is installed
  • The Now Assist admin role is assigned to the right people

Navigate to Now Assist > Administration > Now Assist Settings to see what's active. If the plugin is installed but features are greyed out, the license needs to be activated by your account team.

The Features Worth Implementing First

1. Incident Summarisation

The highest ROI feature in Now Assist. When an agent opens an incident, Now Assist generates a plain-English summary of the record and its related activities. Agents spend 40-60 seconds reading context on every incident — summarisation brings that to under 10 seconds on long incidents.

Enable in: Now Assist > ITSM > Incident Summarisation

No configuration required — it works out of the box on incident records. The summary updates as the incident changes.

2. AI Search

Replaces keyword search with semantic search across your knowledge base. Users can type a question in plain English and get relevant articles, rather than having to guess which keywords to use.

Enable in: Search > AI Search > Configuration

Requires your knowledge base to be indexed. Initial indexing takes 2-4 hours for large knowledge bases. Quality of results depends entirely on the quality of your knowledge articles — poorly written articles produce poor AI search results.

3. Resolution Notes Generation

Generates a draft resolution note based on the incident's work notes and resolution actions. Saves agents 2-5 minutes per incident and improves resolution note quality and consistency.

Features to Approach Carefully

AI Case Summarisation for High-Volume Tables

On instances with thousands of open incidents, AI summarisation can increase database load. Monitor your instance performance metrics after enabling — if you see degradation, configure summarisation to trigger on-demand rather than automatically.

Now Assist in Flow Designer

The natural language flow generation is impressive in demos but requires careful governance in production. Generated flows need thorough review before deployment. Establish a review process before enabling this for your development team.

Data Considerations

Now Assist sends record data to ServiceNow's AI infrastructure for processing. Review your data residency requirements before enabling — if your instance has regulatory constraints on where data can be processed, you need to confirm Now Assist's processing locations are compliant.

Now Assist does not train on your data for other customers' instances. Your data is used to generate responses for your instance only and is not retained for model training.

Measuring Success

Set baseline metrics before enabling Now Assist so you can measure actual impact:

  • Average time to first response on incidents
  • Average time to resolution
  • Knowledge article deflection rate
  • Agent satisfaction scores

Measure the same metrics 30 and 90 days after enabling. Without a baseline, you cannot demonstrate ROI to stakeholders.

Honest Expectations

Now Assist is genuinely useful, but it is not magic. Set these expectations with your stakeholders:

  • Summarisation accuracy depends on the quality of work notes — if agents write poor work notes, summaries will be poor
  • AI Search quality improves over time as your knowledge base improves — it is not a substitute for a well-maintained KB
  • Change management is needed — agents need training on how to interpret and verify AI-generated content
  • Some features require substantial data to provide useful outputs — a knowledge base with 20 articles will not produce good search results

Implementation rollout strategy

Rolling out Now Assist to your entire user base at once is rarely the right approach. A phased rollout lets you validate quality, gather feedback, and address issues before they affect all users. Recommended phasing:

Metrics to track throughout: AI feature adoption rate (what percentage of available AI features are being used), quality scores (user ratings on summaries and suggestions), time-to-close comparison (are AI-assisted agents closing tickets faster), and knowledge base quality improvement driven by AI Search gap analysis.

Related: Activation guide · Now Assist for ITSM · AI Search · Code generation

Common implementation pitfalls

Several pitfalls consistently affect Now Assist implementations that skip the prerequisite assessment phase:

Governance and change management

Now Assist implementation is as much a change management challenge as a technical one. Users who are used to doing things manually may resist AI suggestions if the quality is inconsistent. Communicate clearly what Now Assist does and does not do, set appropriate expectations (AI summaries are a starting point, not a finished document), and create feedback channels for users to report quality issues. Quality issues are often fixable — poor work note writing is the most common root cause of poor summarisation quality, and addressing it is a training and process change rather than a technical change.

See also: Now Assist overview · Activation guide · ITSM features · AI Search

Building the business case for Now Assist

IT leadership often asks for ROI justification before approving Now Assist activation. The strongest business case focuses on three measurable outcomes: time savings for ITSM agents (measure the average time to write resolution notes manually vs accepting an AI suggestion — typically 3-5 minutes per ticket), knowledge base utilisation improvement from AI Search (measure knowledge deflection rates before and after AI Search activation — better search means fewer calls to the service desk), and developer productivity improvement from code generation (measure the time from feature request to code completion for new Business Rules and Script Includes with and without AI assistance).

These three metrics are all measurable with native ServiceNow reporting capabilities. Establish baselines before activation, measure at 30 and 90 days post-activation, and report the delta. A typical target for ITSM summarisation is 2-4 minutes saved per ticket on average — at high ticket volumes, this becomes hundreds of agent-hours per month. The Performance Analytics indicators you build to measure Now Assist ROI are also useful for demonstrating the broader value of your ServiceNow investment to leadership.

NowSpectrum covers every major area of the ServiceNow platform with practical articles written by people who build on it daily. The platform evolves quickly and staying current on releases, new APIs, and best practices is essential

Long-term success factors

The organisations that get lasting value from Now Assist share three characteristics: they measured baseline metrics before activating (so they can demonstrate ROI), they invested in change management alongside technical deployment (so users actually adopt the features), and they continue iterating based on usage data (adjusting configuration based on what is being used, what is being ignored, and what quality feedback reveals). Now Assist is not a one-time deployment — it is an ongoing capability that improves as you tune it to your specific data and workflows. Budget time for ongoing configuration refinement, not just initial deployment. Related: Now Assist overview · Activation · ITSM · AI Search · Xanadu requirements

Implementation quality determines how much value you extract from Now Assist. A technically correct activation with poor change management and no baseline measurement delivers less ROI than a phased rollout with adoption tracking and continuous tuning. Treat it as a programme, not a project — with ongoing measurement, iteration, and stakeholder communication as features mature and your organisation's AI fluency develops.

The most successful Now Assist implementations are led by practitioners who understand both the AI capabilities and the underlying platform deeply enough to configure them correctly and troubleshoot effectively. The NowSpectrum AI series — overview, activation, ITSM, code generation, AI Search — provides the complete technical foundation for that combination of knowledge. Subscribe to the weekly newsletter at newsletter.nowspectrum.com for ongoing updates as Now Assist capabilities expand with each new ServiceNow release.

Want the complete reference?

This article is part of the NowSpectrum knowledge library. Browse all products for cheat sheets, interview prep, and deep-dive reference guides.

Browse All Products →
← Back to all posts