72% of CIOs Are Breaking Even or Losing Money on AI — Highspot Says the Problem Isn't the Model

Highspot’s read on why heavy AI spending isn’t translating into results for go-to-market teams starts with a gap between confidence and outcomes: its 2026 GTM Performance Gap Report found 96% of organizations claim clear growth strategies and 98% say execution is standardized, yet only 53% report consistent outcomes. Layered on top, Gartner data cited in the piece shows 72% of CIOs are breaking even or losing money on their AI investments, while separately 76% of GTM leaders say AI adoption is outpacing their operating model and 85% say they now have more performance data than they can actually use. Highspot’s core argument: AI doesn’t fix broken execution, it amplifies it — a $500 million business with a 5% forecast miss loses $25 million, and inconsistent processes just get automated faster and at greater scale.

That diagnosis matches what a separate Gartner CFO survey found in July: of 204 finance leaders surveyed, 45% said their AI investments lean toward productivity and efficiency, while only 20% said their investments lean toward improving decision quality — a split Gartner frames as misaligned with what boards actually want, since boards weight growth and decision-sharpening AI more heavily than pure cost-cutting. The same survey found functions that invested in “Upend” initiatives — AI projects creating new value propositions, products, or markets — were more than twice as likely to report high realized value than productivity-only investments were.

Read together, both point to the same fix: less spend chasing efficiency for its own sake, more spend on AI wired into decisions that actually move revenue. Highspot’s own recoverable-value estimate puts a number on the gap — organizations believe more consistent execution could lift performance by an average of 8.5% within 12 months, worth roughly $85 million for a $1 billion business — which is a way of saying the money being left on the table isn’t a model-capability problem, it’s an execution-architecture one.