ServiceNow’s Zurich release and the cumulative new-features release notes are dense by design. They catalogue product deltas across the platform: Now Assist skills, Voice AI on phone channels, agentic ITSM workflows, Build Agent for conversational app creation, CMDB Workspace changes, and stronger controls around machine identity and sensitive data.
Read as a feature list, Zurich looks like an AI adoption mandate. Read as an operating problem, it looks like every other major platform release: more ways to move work, more ways to create work, and more places where an unclear entitlement or an outdated knowledge article gets amplified.
The organisations that get value from Zurich will not be the ones that enable the longest list of skills. They will be the ones that decide, in advance, which service journeys are allowed to use assistance, which actions still need a named human, and how success will be measured against the current baseline.
Five Zurich themes that change the operating conversation
These are not a full release digest. They are the themes that change how service, platform, and risk teams have to talk to each other.
1. Now Assist and agentic ITSM move from assistive copy to workflow participation. Zurich materials highlight conversational change support, image-to-incident style intake, daily work-plan generation, and multi-step agentic playbooks. That is useful when the taxonomy, assignment rules, and CAB expectations already make sense. It is expensive when the agent inherits a request catalogue that still mirrors last year’s org chart.
2. Voice AI puts the phone channel on the same governance surface as chat and portal. Self-service and deflection on voice only help if the knowledge articles, identity checks, and escalation paths are current. A phone agent that cites obsolete reset instructions is not a productivity win; it is a new failure mode with higher emotional cost.
3. Build Agent lowers the cost of creating apps — and raises the cost of uncontrolled creation. Natural-language app building on the Now Platform is a real delivery accelerator when sandboxing, review, and promotion rules are explicit. Without those rules, Zurich makes it easier for well-intentioned teams to invent parallel workflows that never join the service catalogue your resolvers actually run.
4. CMDB and integration context remain the quiet dependency. Create-CI and Workspace improvements only matter if configuration items, identifiers, and freshness expectations are agreed with the systems that own the truth. An agent that triages against a stale CI relationship will sound confident and still send the wrong group.
5. Usage measurement and default skill behaviour are commercial and operational facts, not footnotes. Patch notes around Now Assist usage models and skills turning on by default are the difference between a controlled rollout and a surprise invoice — or a surprise change in how requesters experience the portal. Entitlements and default states belong in discovery, not in a post-go-live ticket.
ServiceNow documents these changes in product release notes and community posts. Treat the vendor page as the source of truth for what shipped. Treat your own request sample and licence inventory as the source of truth for what you should enable.
What to settle before you turn features on
Four questions are cheap before an upgrade window and expensive after a pilot has political momentum.
Which request types are in scope for assistance? Name the journeys — password reset, access request, laptop failure, standard change — not the platform modules. If the category tree cannot explain those journeys cleanly, fix the tree before you add Voice or an agentic playbook.
Who owns the record of authority for people, assets, and accounts? ServiceNow may coordinate the work while identity, HR, CRM, or asset systems own the facts. Write the freshness expectation and the failure behaviour when a downstream system is unavailable. An HTTP 200 is not a business completion.
Which actions may run without a human, and who signed that list? Classification and knowledge retrieval are different from creating a change, resetting access, or closing an incident. Consequential actions need named accountability. Role masking and machine-identity controls in the platform help only if your organisation has already decided the policy those controls enforce.
What is the baseline for the journeys you claim AI will improve? Volume, waiting time, rework, first-contact resolution, and successful completion for the specific request types in scope. Without that baseline, Zurich becomes a narrative programme: “agents are handling more tickets,” with no defensible comparison.
These questions mirror how we already frame ServiceNow work on the ServiceNow practice page: start from the service people are trying to receive, then assess the instance, entitlements, and integrations that make that service true.
How this maps to a bounded Emerge engagement
We do not treat Zurich as a reason to re-platform. We treat it as a reason to sequence three familiar steps.
Service-workflow assessment. Sample real requests from submission to resolution. Record transfers, waiting states, missing information, and external dependencies. Output a prioritised process map and an agreed definition of successful completion — including which Zurich capabilities are candidates for the first bounded change.
Enterprise integration. Connect approved context through documented interfaces, dedicated integration identities, and reconciliation. Keep ServiceNow as the coordination record when another system executes the business action. This is also where Salesforce Agentforce fits for organisations that want one front-office agent surface reading ServiceNow incidents, CMDB, and knowledge — see the ServiceNow + Agentforce integration guide.
Managed improvement. Turn on only what the assessment justified. Acceptance covers misclassification, unavailable approvers, duplicate submissions, integration failures, and withdrawn requests. Reporting separates volume, waiting time, rework, and successful completion so the team can improve the service without a single vanity metric.
AI may classify a request, retrieve approved knowledge, or prepare context for a resolver. It does not replace the need for current operating instructions, least-privilege access, and an owner for every knowledge article the agent is allowed to cite.
What this is not
This note is not a claim that Emerge has delivered every Zurich module, nor a ServiceNow partnership badge. Licensing, product entitlements, and instance configuration are confirmed in discovery. Illustrative workflows are not anonymised client outcomes.
It is also not a substitute for ServiceNow’s own release-notes summary. Use that page to see what changed. Use your operating baseline to decide what matters.
A practical next step
If you are planning a Zurich upgrade or a Now Assist expansion this quarter, bring three artefacts to the first conversation: a sample of recent requests (successes and failures), your current ServiceNow product entitlements, and the list of systems that own identity, asset, and customer facts. We will map those against the Zurich capabilities that are actually relevant — and leave the rest of the feature catalogue where it belongs: on the vendor’s page.
Discuss ServiceNow — assessment first, features second.