Author End-to-End Tests in Natural Language with KaneAI
Plan, Author, and Evolve E2E Tests for Web and Mobile Apps in Natural Language.
ServiceNow Testing: What to Test and Where ATF Stops
What to test across ITSM, catalog, HR and AI agent flows, where ServiceNow's Automated Test Framework stops, and how to gate a family release with evidence.
Last Updated on:
ServiceNow ended the second quarter of 2026 with 658 customers paying more than $5 million each in annual contract value, according to its Q2 2026 results.
The next ServiceNow upgrade is the Brazil family release, expected to reach general availability in November 2026 according to NowBen's Brazil key dates.
Teams at that scale run incident routing, HR onboarding and change approvals as customizations on a platform that ServiceNow upgrades underneath them. ServiceNow testing is how a team proves those workflows still behave correctly before each upgrade reaches production. If you are choosing tools rather than planning coverage, the ServiceNow testing tools roundup compares nine of them side by side.
TL;DR
ServiceNow testing is the practice of verifying that an instance's workflows, catalog items, integrations and AI agents still behave correctly after every configuration change, update set and family release. It exists because ServiceNow upgrades the platform underneath your customizations twice a year, and a broken approval or SLA reaches every employee who uses it.
What Should You Know Before the Next Upgrade?
- Does ATF cover platform UI or connected systems? No. The Automated Test Framework runs server, form, catalog and REST tests inside the instance, but it cannot test platform UI such as dashboards, and it stops at the instance boundary, so it never reaches Salesforce, Workday or a real phone.
- Can ATF run on production? No. ServiceNow disables the property that runs automated tests on production by default and says to test only on development, test and other non-production instances.
- ServiceNow Brazil release: The Brazil family release is expected to reach general availability in November 2026, with a Release Testing Preview at least 60 days earlier, which is the window for running a full regression on an upgraded sub-production clone.
- Is a correct final record enough for an AI agent? No. An AI Agent Studio agent can reach the right record through an unexpected tool call, so review its decision log in the Testing view as well as the end state.
- Plain-English UI tests: TestMu AI's Kane CLI runs a ServiceNow flow described in plain English, re-finds an element that moved after an upgrade, and fails the run when the outcome itself is wrong, such as a wrong assignment group.
What Is ServiceNow Testing?
ServiceNow testing verifies that the workflows a business runs on its ServiceNow instance, such as incident routing, catalog requests, HR cases and change approvals, still work after a configuration change, an update set or a platform upgrade. It covers the instance itself and the systems its flows reach, such as Salesforce, Workday or SAP.
The work splits by layer. Developers unit-test business rules and script includes, admins and process owners check forms and catalog items, and QA leads own the regression testing suite that runs before each upgrade. Each layer needs a different tool, so most teams pair ServiceNow's own framework with browser and device testing that runs outside the instance.
Why Is ServiceNow Hard to Automate?
ServiceNow changes underneath every customization on a fixed schedule, and much of its UI is generated rather than hand-built, so a test pinned to an exact field position or element id tends to break when a release lands.
- Releases and update sets - two family releases a year, plus update sets moved between instances in between. Any of them can rename a field, move a catalog variable or change platform behavior under your business rules.
- Generated UI - classic UI loads record forms in an iframe, Next Experience pages are assembled from web components that sit behind shadow roots, and element ids embed sys_id values unique to each instance. The frame switching this forces is shown step by step in automating ServiceNow with Selenium.
- Conditional forms - UI policies and client scripts show, hide or make catalog variables mandatory based on earlier answers, so the form under test depends on the data entered one step before.
- Asynchronous outcomes - approvals, SLAs and notifications fire through Flow Designer after the triggering action, often on a different record, so a check that runs too early fails a flow that works.
- Flows that leave the instance - an incident can open a Salesforce case and an HR request can write back to Workday or SAP, and tests that run inside ServiceNow stop at that boundary.
When a suite needs repair after every release, regression checks drift back to manual clicking inside the upgrade window, and the change advisory board signs off without a recorded result.
What Should You Test in ServiceNow?
Write each test as the outcome a person would check by hand, in one or two sentences. The table maps the main ServiceNow areas to a test written that way, the behavior it proves, and who usually owns it.
| Area | Example test in plain English | What it proves | Usual owner |
|---|---|---|---|
| ITSM | Log a P1 incident against the email service CI; check that it lands in the Messaging queue with a resolution SLA attached | Assignment rules and SLA definitions | Process owner |
| Change | Submit a normal change; verify CAB approval routing and that it cannot move to Implement unapproved | Approval policy and state model | Developer |
| Service Catalog | Order a laptop that needs manager approval; verify the REQ and RITM are linked and a fulfilment task reaches the right group | Catalog variables, UI policies and the fulfilment flow | Admin |
| HR Service Delivery | Onboard a new hire; verify the HR case, its tasks and the write-back to Workday | Case templates and the HR integration | HR operations |
| Customer Service | Open a case from the customer portal; verify queue assignment and entitlement in the agent workspace | Portal and workspace behavior together | QA lead |
| Cross-system | Create an incident; verify the matching Salesforce case exists and updates sync both ways | The integration contract end to end | Developer |
| Custom scoped apps | Trigger a Flow Designer flow in a scoped app; verify its outputs and business-rule side effects | Custom logic your team owns | Developer |
| Now Mobile | Approve a request on iOS and Android; verify the record state updates | Mobile approval flow on real hardware | QA lead |
| Virtual Agent | Ask Virtual Agent to reset a password; verify the topic it picks, the actions it takes and the resulting record | Conversation design and its side effects | Platform owner |
Start with the flows that would stop work if they broke after an upgrade, usually incident routing, catalog fulfilment and change approval, and add the rest once those run on every release. For mobile approvals, the Appium steps in ServiceNow with Appium cover the native app, and a ServiceNow test case template keeps each case's preconditions, steps and expected result in one format.
Note: Describe one of these flows in plain English and run it from your terminal or CI with Kane CLI, which re-finds elements that moved and fails the run when the outcome is wrong. Create a TestMu AI account
What Does ATF Test, and Where Does It Stop?
The Automated Test Framework (ATF) is ServiceNow's built-in tool for creating, managing and running automated tests on an instance. According to ServiceNow's ATF overview, its step categories cover server-side JavaScript unit tests (including Jasmine), forms, Service Catalog items in the platform and in Service Portal, and inbound REST requests, and ATF tracks and cleans up the data a test creates so a run leaves no permanent change.
Per ServiceNow's ATF design considerations, ATF prevents custom UI testing of ServiceNow AI Platform features, so you cannot test dashboards or graphical designers with it, and tests should run only on development, test and other non-production instances to avoid data corruption or an outage. The property that runs automated tests is disabled by default to stop accidental runs on production.
ATF tests run inside the instance under test and clean up after themselves. That makes ATF the right place for server-side logic and in-platform regression, and the wrong place to confirm what happens in a connected system or on a phone.
| Check | ATF | Browser and device testing outside the instance |
|---|---|---|
| Business rules and script includes | Yes, through server-side JavaScript unit tests | Only through their visible effect on a form or record |
| Dashboards and graphical designers | Not supported | Yes, by driving the rendered page |
| Flows into Salesforce, Workday or SAP | Stops at the instance boundary | One test can follow the flow into the connected system |
| Now Mobile on real phones | Out of scope | Yes, on a real device cloud |
Keep ATF for what it does well and cover the other rows outside the instance. When a browser test fails, the ServiceNow integration logs the incident from the test session straight into ServiceNow through the Mark as Bug button, so the defect lands in the same queue as production issues.
How to Test ServiceNow Before a Family Release
How Do I Test ServiceNow Upgrades?
Test a ServiceNow upgrade on a clone of production that you upgrade before production does. Rerun your regression suite for the critical flows there, then work through the skipped records list, where the upgrade left your customized records unchanged, and retest every record you merge or revert before the production upgrade is approved.
ServiceNow's guide to processing skipped records (Brazil release docs, updated September 10, 2026) tells admins to review each skipped customization and to thoroughly test every change they make to those records after the upgrade.
For Brazil, NowBen's schedule puts a Release Testing Preview at least 60 days before general availability and Early Availability at least 45 days before it. That gives a team roughly two months to run its regression against the new release before production moves.
- Clone, then upgrade - copy production into a sub-production instance and upgrade that copy, so tests run against your real configuration rather than a demo instance.
- Run ATF first - server-side regressions in business rules, script includes and REST endpoints are the cheapest to find and fix.
- Run the browser and device suite - point it at the same clone and cover the table's flows on the browsers and phones employees actually use.
- Review every changed step - when a test re-finds a field that moved, confirm the field moved and the logic did not.
- Treat a wrong outcome as a defect - a wrong assignment group or a missing SLA after the upgrade is the bug the run exists to catch, never a step to repair.
- Attach the evidence - add each run's screenshots, logs and verdict to the upgrade change request so the CAB approves on recorded results.
Repeat the run after every update set that touches a covered flow as well as at each family release. The suite then becomes a gate every change has to pass, instead of a project that restarts twice a year.
How Do AI Agents Change ServiceNow Testing?
ServiceNow's AI agent testing starts from All > AI Agent Studio > Testing: you choose the agent or agentic workflow, describe the task, and the run shows a process map of the agents' tasks and a decision log of each agent's reasoning, which you can download. To check many runs at once, automated evaluations use execution logs.
An agent plans its own route through tools and records, so two runs of the same request can take different valid paths to the same result. A scripted ATF test expects the same sequence on every run, and a check on the final record alone can pass while the agent called a tool it should not have touched.
- Decision log review - read which tools the agent called, not only whether the ticket or record ended in the expected state.
- Permission scope - test what the agent is allowed to reach, since an autonomous agent can touch more tables and workflows than a scripted test case does.
- Drift regression - rerun the same scenarios after any change to the model, the agent's instructions or its tools, because a route that passed before can change with no change to your tests.
Virtual Agent conversations need a different check: whether the bot answered accurately, kept context across turns and escalated when it should. TestMu AI's Agent Testing connects to a chat agent's endpoint, runs multi-turn conversations as real users would, and scores each one for hallucination, completeness and context awareness. The chat-agent types it lists include IT helpdesk bots and HR assistants, the same jobs a password-reset or onboarding topic in Virtual Agent performs.
How to Run ServiceNow Tests With KaneAI and Kane CLI
Process owners and developers need the same tests in two different places. KaneAI lets an admin, HR operations lead or manual tester write a ServiceNow flow in plain English, runs it, and re-anchors steps when the UI changes, so maintenance becomes reviewing a heal instead of rewriting a selector. Its tunnel support reaches sub-production instances behind a firewall.
Kane CLI runs the same kind of plain-English flow from a terminal, a CI job or a coding agent such as Claude Code or Cursor, and returns a verdict a pipeline can gate on. For SSO-protected instances, a run can reuse a Chrome profile that is already signed in, so the test never scripts the SAML round trip.
- Cached replay - a flow saved as a test.md replays its unchanged steps from cache without consuming LLM credits.
- Step-heal and flow-heal - when one step's target moves, step-heal re-finds that element and leaves every other step untouched; flow-heal re-reasons the path only when a step cannot be healed, and never changes the acceptance criteria.
- Wrong results fail - a wrong total, assignment group or SLA is reported as a bug and is never healed into a pass.
- Evidence file - every run writes a single .evidence file holding per-step screenshots, console output, network traffic and the pass or fail verdict, which a release manager can add to the upgrade change record.
- Exit codes - 0 on pass, 1 on fail, 2 on setup or auth errors and 3 on timeout, and the --agent flag switches output to NDJSON a script can parse.
- Benchmark - Kane CLI ranked #1 of 7 tools at 76.7% on the PRISM 2026-Q4 browser-testing benchmark, which scores 50 real browser scenarios on whether a pass means the flow worked.
- Now Mobile coverage - the same approval flows run on TestMu AI's real device cloud of 50,000+ real Android and iOS devices.
Pointed at the public knowledge base on Now Support, ServiceNow's own support site, a one-line objective searches for ATF articles and checks the results. The last NDJSON line is the verdict, with the values the run read off the page:
kane-cli run "Go to https://support.servicenow.com/kb. Search the knowledge base for 'Automated Test Framework'. Verify that the search results list shows at least one article whose title mentions Automated Test Framework or ATF." --agent --headless | tail -1
{
"type": "run_end",
"status": "passed",
"one_liner": "searched for Automated Test Framework articles on ServiceNow Support",
"final_state": {
"url": "https://support.servicenow.com/kb?id=kb_aisearch&query=Automated%20Test%20Framework",
"automated_test_framework_article_title": "Async updates via Automated Test Framework",
"atf_article_title": "Why guest user testing is not supported in the Automated Test Framework (ATF)"
},
"reason": "Objective completed",
"duration": 64.9,
"result_code": 100
}
# exit code 0Change the expectation to something the page does not show, such as a first result that mentions Salesforce, and the same command exits with code 1. The failure record also classifies the cause, here a test-data issue rather than a product bug, so the reviewer knows to update the test before filing anything:
{
"type": "run_end",
"status": "failed",
"one_liner": "The search completed successfully, but its first result was not the article the test expected.",
"reason": "Final verification failed: \"the title of the first article in the search results contains the word 'Salesforce'.\"",
"duration": 75,
"result_code": 710,
"verdict": { "family": "automation_bug", "category": "test_data_issue" }
}
# exit code 1On your own instance, the objective becomes a row from the table above, run against the upgraded clone, with the exit code deciding whether the change request moves forward.
Conclusion
Pick the three flows from the table that would stop work if they broke, write each as one plain-English objective, and run them against an upgraded sub-production clone before Brazil reaches general availability. Install Kane CLI with npm install -g @testmuai/kane-cli and follow the Kane CLI introduction to run the first one, then see ServiceNow testing with KaneAI when process owners need to author the rest without code.
Testing web or mobile apps without writing code? KaneAI by TestMu AI turns plain-English instructions into tests.
Author
Reviewer
Maneesh Sharma is Chief Operating Officer at TestMu AI, where he leads the company's revenue functions across go-to-market, sales, RevOps, customer success, and the partner ecosystem, driving its global expansion. He brings over 28 years of experience across product, go-to-market, and business operations. Before TestMu AI he was Managing Director and General Manager for GitHub across India and APAC, Head of Partner Sales and Ecosystem at Adobe, Chief Revenue Officer at WizIQ, and Head of Products and Solutions at SAP India. Maneesh holds an MBA from IIM Bangalore and a B.E. in Electronics and Communications from NIT Surat.
ServiceNow Testing FAQs
Did you find this page helpful?
More Related Blogs
TestMu AI forEnterprise
Get access to solutions built on Enterprise
grade security, privacy, & compliance
- Advanced access controls
- Advanced data retention rules
- Advanced Local Testing
- Premium Support options
- Early access to beta features
- Private Slack Channel
- Unlimited Manual Accessibility DevTools Tests
















