[object Object]

The Now Assist pilot delivered measurable savings on incident summarization within six weeks. The pilot also surfaced a governance gap that took longer to fix than the pilot itself — nobody had configured PII filtering on the prompts and the audit team flagged it on the first review. Now Assist is real value waiting for the discipline of treating it like any other production system. The patterns below are what separates a successful enablement from a security finding.

What Now Assist Is

Now Assist is the platform’s branding for generative AI features across the surface — case summarization, knowledge article drafting, code suggestion in App Engine Studio, chat deflection in Virtual Agent, change risk insights, RCA assist on Problems. Some features run on ServiceNow’s default model; others can use partner LLMs via BYO LLM. The 2026 packaging includes most surfaces in the Professional and Enterprise tiers; Agentic AI is gated to Enterprise Plus.

Common Now Assist surfaces:
  Incident summary (ITSM)
  Case summary + response draft (CSM)
  HR case summary + policy article suggest (HRSD)
  Knowledge article draft from resolved incidents
  Code suggestion in App Engine Studio
  Change risk insight on change form
  RCA contributing-factor draft on problems

Use Cases That Actually Pay Back

Best-fit: agent-assist summarization of long tickets (saves real agent time), knowledge article drafting from resolved incidents (improves KB freshness), Virtual Agent response improvement (lifts containment), reporting narrative generation (saves analyst time on weekly summaries). These use cases save measurable agent time and have ROI you can defend in a steering committee.

Avoid: using Now Assist as the decision-maker for high-stakes routing, financial determinations, or data that cannot tolerate hallucination. The features draft and suggest; humans review and decide on anything consequential.

Governance You Can’t Skip

Now Assist data governance is the feature most administrators under-configure. Set data access controls (which roles can invoke which features), LLM region (residency for the prompt and response data), retention (how long prompts and outputs are stored), and PII filtering (redact known sensitive patterns before they reach the model) before turning features on for users. The defaults are reasonable starting points and not necessarily right for your regulatory profile.

Governance configuration checklist:
  - role-gated feature access
  - region for default and BYO endpoints
  - prompt/response retention period
  - PII redaction patterns (custom + built-in)
  - audit log retention
  - cost cap per feature

The audit log records every Now Assist invocation including the model, version, prompt, response, user, and timestamp. Review weekly at first to validate governance is working; monthly at steady state. The AI Control Tower (Zurich and later) is the central surface for this.

Skill Kit for Custom Use Cases

Now Assist Skill Kit lets you author custom AI skills — prompt templates, grounding data sources, allowed actions, output schemas. Use it for domain-specific summarization your industry needs (regulated reporting language, vendor-specific terminology, internal process language) where the generic skills do not fit. Start with a narrow, high-volume use case so the cost-benefit is clear.

// Skill invocation pattern (conceptual)
var skill = new sn_assist.Skill('IncidentSummaryHealthcare');
var result = skill.execute({
  incident_sys_id: current.sys_id,
  redact: ['PHI', 'PII'],
  audience: 'clinical_operations'
});
current.work_notes = '[AI summary] '+result.text;

Cost Control

Now Assist is metered per call (or per token, depending on feature and tier). Each invocation consumes credits or token allowance. Monitor usage in the AI Control Tower or Now Assist Admin. Restrict high-cost skills to specific personas until you understand the cost-benefit. Set per-feature spend caps with alerts at 50% and 80% of the cap; surprises at month-end are common when teams enable broadly without instrumentation.

Common Failure Modes

Features enabled across all users on day one — produces a wave of unaudited usage and an unpredictable cost spike. Stage rollout. Custom skills built without a feedback loop — the prompts drift from current data and the outputs degrade quietly. Quarterly review of skill performance is non-negotiable. PII filtering configured for the obvious patterns (SSN, credit card) and missing the domain-specific ones (employee ID, account number formats) — extend the filters to your actual data shapes.

What Changed in 2026

The April 2026 announcement reframed Now Assist from a discrete add-on into the default interaction model across tiers. Agentic AI on Enterprise Plus extends the surface from suggestion to autonomous execution. The AI Control Tower (Zurich) provides the governance plane that earlier releases required custom-built dashboards for.

Implementation Sequence

Pick one feature on one user group for one month. Validate cost, accuracy, and user adoption. Configure the governance plane (audit, retention, PII, cost caps) before broad enablement. Expand to the next feature in priority order. Trying to enable all Now Assist surfaces simultaneously produces governance debt and cost surprises that erode trust in the program.

What to do this week: enable the audit log review for any Now Assist feature already turned on and read the last week of invocations; the patterns you find are your governance baseline.

[object Object]
Share