A deep dive into S&P 500 earnings disclosures from a 30-day window ending August 17, 2026 reveals a stark disconnect between AI rhetoric and financial reality. The study analyzed all 433 in-window reporters among the index's 500 issuers, and the results should make anyone building enterprise AI products think twice about customer promises.

Talk Is Cheap; Data Is Scarcer

Of 428 unique periodic filers, 253 companies—59.1%—explicitly mentioned AI or machine learning in their disclosures. That sounds impressive until you drill down: only 47 of those same companies (11.4%) named a specific internal financial or operating path tied to their AI initiatives. The vast majority are dropping keywords without substance. When it comes to spending, the numbers get even thinner. Just 18 out of 411 non-provider adopters said they spend on AI at all. Only three provided an actual dollar figure, and only one company—Incyte—isolated a ring-fenced AI-specific total. If you're building budgeting tools or cost allocation dashboards for enterprise customers, this should concern you. These companies either don't know what they're spending, or they're not telling.

Measured Results? What Measured Results?

The study found that 13 out of 411 adopters measured anything at all—usually operational metrics like time saved or model quality improvements. But here's the catch: those operational benefits have "no economic bridge" to realized financial gain without additional disclosure. Only two companies tied AI to a current issuer cost effect, and one of those (Cigna) was explicitly excluded from P&L impact because it involved customer medical costs. For all the talk about 10x productivity gains and ROI validation, zero comparable pairs exist in this dataset. The research explicitly notes that their return calculation is "NOT COMPUTED, not zero"—a crucial distinction for anyone citing this data. You can't prove AI doesn't work any more than you can prove it does from what's disclosed here.

The One Forward Guide Worth Noting

Broadridge was the only company out of 411 adopters to issue AI-only dollar guidance: $25 million in FY2027 AI-driven productivity. That's forward-looking, not realized results. For infrastructure teams evaluating where enterprise budgets are actually heading, this single data point is your best leading indicator—and it's still just a projection.

Key Takeaways

  • Nearly 60% of S&P 500 companies mentioned AI in earnings calls, but only 11.4% disclosed specific financial or operational paths
  • Only 3 companies provided dollar figures for AI spending; 1 isolated an AI-specific total
  • Zero comparable spend-to-return pairs exist—no company proved ROI through disclosed methodology
  • Operational improvements don't equal P&L impact without explicit financial bridges in filings
  • The only FY2027 AI-only guidance came from Broadridge at $25M, and it's not realized yet

The Bottom Line

If you're selling AI infrastructure, cost management tools, or ROI tracking software to enterprise buyers, this data is a reality check. Your customers are talking the talk but can't walk the walk because they genuinely don't have the measurement frameworks in place—or they're not disclosing them. That's either a massive market opportunity for better observability tooling, or evidence that the enterprise AI value chain has serious accounting problems we haven't solved yet.