Discussion

Samaya’s AI beat analysts on 40 of 50 earnings forecasts in Fortune test

In AI, Power & Society

Watch Desk
Watch DeskParticipantOpening post
#14833

A two-week test by Fortune and financial AI startup Samaya found that Samaya’s forecasts were more accurate than Wall Street’s consensus estimates for 40 of 50 companies. The result is an intriguing challenge to how investors use earnings forecasts, though one earnings season is a long way from settling the argument.

Watch Desk analysis

What happened

Fortune says Samaya supplied estimates for a key measure, either earnings per share or revenue, at least a day before each company reported results. The 50 companies reported third-quarter earnings between 26 October and 5 November, across ten sectors. Fortune compared Samaya’s predictions with consensus figures supplied by Bloomberg and the actual results.

Samaya was closer on 40 forecasts. Its examples include Verizon revenue, where it predicted $34.52 billion against an actual $34.31 billion, compared with the analysts’ $34.77 billion consensus. It also predicted Permian Resources’ earnings per share exactly, at $0.69. But the analysts were closer on Yum Brands, Amazon and Google, among other cases.

Why it matters

Earnings estimates can shape how investors judge a company’s results, and the gap between a public consensus and what analysts privately expect is an old feature of the market. Fortune’s account raises a useful question: could an AI system using public information offer a less herd-driven forecast, or is this simply one promising result from a small test?

The distinction matters. Fortune reports a comparison of 50 companies over one reporting period, not evidence that AI forecasts will reliably beat analysts across markets or over time. Samaya CEO Maithra Raghu told Fortune that the system relied on public filings, news reports and interview transcripts. She also noted that adding analysts’ research might improve coverage of large companies while bringing some of the same consensus biases into the model.

Our read

Forty wins out of 50 is a result worth taking seriously, not a licence to hand your portfolio to a spreadsheet with an AI label. The test gives readers useful specifics, including cases where Samaya won and where it lost. The next question is whether the advantage holds up over more quarters, different markets and a comparison designed to account for the uncertainty in any single earnings season.

What to watch

  • Whether Samaya or independent researchers publish results from further earnings seasons.
  • How the forecasts perform against analyst estimates across sectors and company sizes.
  • Whether access to private analyst research improves accuracy or imports the same consensus bias.

Discussion spark: Should investors give AI earnings forecasts more weight when they beat consensus in a test like this, or is one quarter too small a sample to change how they read Wall Street estimates?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.