It's tempting to start an AI transformation conversation with a specific model or vendor. The programs that produce measurable outcomes almost always start somewhere else: the workflow the model would sit inside, and whether that workflow's data and process are clean enough to automate in the first place.
A useful filter is to ask three questions of any candidate process: is the decision repeatable, is the underlying data trustworthy, and is there a human in the loop who can catch and correct errors early. Processes that pass all three are strong automation candidates. Processes that fail the second — dirty or inconsistent data — need a remediation step before any model gets near them.
On the ServiceNow platform specifically, this usually means CMDB and catalog data quality work comes before AI-assisted triage or predictive intelligence features, not after. Skipping that step is the single most common reason AI pilots stall after the demo.
Start with one workflow, instrument it well enough to prove the before/after numbers, and only then scale the pattern to the next process. Enterprise-wide AI strategy documents are useful for direction; they rarely substitute for one well-chosen pilot.
