AI

AI startup beats Wall Street analysts 80% of the time on earnings

Samaya's forecasting model outperformed human consensus estimates for 40 of 50 major companies in Q3 2026 experiment.

Omega Editorial · November 19, 2026 · 3 min read

Key takeaways

  • Samaya's AI model beat Wall Street analyst consensus on 40 of 50 Q3 2026 earnings predictions across ten sectors
  • The AI excelled at mid-sized companies but struggled with tech giants that receive extensive private analyst research
  • Wall Street analysts face subtle pressures to conform to consensus views, potentially limiting forecast accuracy
  • Corporate IR teams influence analyst estimates through soft pressure despite SEC rules against sharing material information
  • AI forecasting models may disrupt the decades-old earnings ritual if they prove consistently more accurate than human analysts

AI outguns Wall Street in earnings forecasting test

A financial AI startup has demonstrated it can predict quarterly earnings more accurately than Wall Street's traditional analyst consensus—winning 40 out of 50 head-to-head matchups in a controlled experiment.

Samaya, which raised $43 million in 2025, accepted a challenge from Fortune to forecast Q3 2026 earnings for 50 major companies before their results were released. The startup's AI-generated predictions were then compared against Bloomberg's analyst consensus figures. When actual earnings were reported between October 26 and November 5, Samaya's forecasts proved more accurate 80% of the time, often by considerable margins.

The test covered ten sectors and included household names from Verizon and UPS to Palantir and Amazon. Samaya predicted Verizon's revenue at $34.52 billion versus the actual $34.31 billion—closer than analysts' $34.77 billion estimate. For Permian Resources, the AI nailed the exact EPS figure of $0.69, while the Street had predicted $0.625.

Why it matters

The quarterly earnings ritual has remained largely unchanged for decades, with companies and their stocks rewarded or punished based on whether they meet analyst consensus targets. If AI models prove consistently superior to human forecasters, the entire earnings ecosystem—from how companies manage investor relations to how markets price expectations—could face fundamental disruption. The experiment also highlights how AI's lack of human bias might actually be an advantage in financial forecasting, where analysts face subtle pressures to conform to consensus views.

Where AI stumbled

Samaya's model wasn't perfect. Its most notable miss came with Yum Brands, where it forecast EPS of $1.42 versus the actual $1.56 (analysts predicted $1.49). The AI also underperformed when forecasting how much tech giants Amazon and Google would miss revenue expectations.

Samaya CEO Maithra Raghu, a former member of Google's foundational Brain AI team, noted that analyst consensus typically proved more accurate for very large companies that are subjects of extensive private research. The startup's model relied on publicly available data including SEC filings, news reports, and interview transcripts—but not proprietary analyst notes.

The human factor in Wall Street forecasts

The experiment exposes long-standing dynamics in how analyst consensus forms. Sheraz Mian, Director of Research at Zacks, acknowledged that analysts face incentives to stick with the herd rather than publish contrarian forecasts. The potential reputational damage from an incorrect outlier prediction typically outweighs any acclaim from being right.

Corporate investor relations teams reinforce this conformity through what Paul Hribar, an accounting professor at the University of Iowa, describes as a pattern of "optimistic numbers that get walked down to beatable targets" as earnings approach. While SEC Regulation Fair Disclosure forbids companies from sharing material information with analysts, soft pressure remains common.

Raghu said Samaya may incorporate analyst notes into future training data but expressed concern that doing so could teach the AI to internalize the same consensus-seeking behavior that defines Wall Street forecasting. Whether AI models will eventually adopt human biases or disrupt the cozy analyst-IR relationship remains an open question.

These findings were first reported by Fortune, which conducted the experiment with Samaya over a two-week period during Q3 2026 earnings season.

  • #earnings forecasting
  • #financial ai
  • #wall street analysts
  • #samaya
  • #machine learning
  • #investor relations

This is an original analysis by the Omega editorial team. Source reporting: AI Watch.

Want systems like this working for your business?

Book a Call

More in AI