Every ServiceNow customer is now being asked the same board question: "What are we doing with AI?" After two years of Now Assist deployments, we can report where the value is real, where it's theater, and what determines which one you get.
Where the payback is real
Case and incident summarization. The least glamorous feature is the most reliable earner. Agents inherit tickets with full context in seconds instead of scrolling history; handoffs stop losing information. On desks above ~2,000 tickets a month, summarization alone typically justifies the licensing.
Knowledge drafting. The eternal reason knowledge bases rot is that writing articles is nobody's job. Now Assist drafting from resolved incidents inverts the economics: agents edit instead of author. Our clients see knowledge coverage climb within a quarter — which then feeds deflection.
Resolution notes and change summaries. Compliance-quality documentation, generated in the flow of work. Auditors notice the difference before the CFO does.
Where enthusiasm outruns value
Customer-facing generative responses without quality gates; AI search layered over a neglected knowledge base; and any deployment where nobody defined what "working" means. GenAI amplifies the platform underneath it — including its problems.
The dependency nobody budgets for
Data quality. Summaries are only as good as the records they summarize; recommendations are only as good as the CMDB relationships they traverse. The best predictor of Now Assist ROI in our portfolio isn't industry or size — it's CMDB health score and knowledge hygiene at the start. Fix the foundation first, or budget for AI that confidently hallucinates.
How to start
A six-week measured pilot: summarization and knowledge drafting on your real volumes, baselined against handle time and deflection, with quality gates agreed up front. If the numbers clear the bar, expand. If they don't, you've spent six weeks learning instead of a year hoping.