A field report rather than a market forecast. Where adoption really landed by function, which pilots died and at what stage, what the surviving deployments have in common, and how spend correlated with measurable outcome. Method and sample size are stated before any chart.
Sample size, how respondents were sourced, and the self-selection bias that comes with surveying people already interested in AI. Any report that puts the chart before the method is asking you to trust a number whose provenance it is hiding, and this category is full of them.
Use the failure-stage data to plan your own pilot backwards: if pilots die at evaluation, build the eval suite before the agent. Use the ownership finding to decide whether you have a named owner before you start. Those two decisions matter more than which vendor you pick.