In 1997, a Fortune journalist observed, “The simplest, most visible, most merciless measure of corporate success…has become this one: Did you make your earnings last quarter?” Those words hold up just as well today. Indeed, the current Q3 earnings season has featured the same, decades-old ritual: Investors wait to discover if a firm can meet or exceed a consensus target from Wall Street analysts—and then reward or punish the company’s share price accordingly. Now, that ritual could be in for a shakeup.
In October, I was approached by a financial AI startup called Samaya that claimed it could predict earnings outcomes more accurately than the analysts. Prove it, I said. I invited Samaya, which raised $43 million from prominent investors in 2025, to put its money where its mouth is and provide Q3 predictions for 50 big companies, ranging from Coca-Cola to Carvana, prior to their earnings releases.
The company accepted the challenge. This led to a two-week experiment in which we placed Samaya’s AI predictions alongside the consensus view of Wall Street analysts in a spreadsheet, and then recorded the actual Q3 earnings number. Samaya didn’t just beat the analysts, it smoked them. For 40 out of the 50 companies, the AI startup’s forecasts were more accurate than the Wall Street consensus and, in most of these cases, by a considerable margin.
The experiment represents a small sample over a single earnings period, but it has potentially large implications for investors, companies and analysts alike. More broadly, the exercise underscored, once again, how AI is transforming many aspects of corporate America.
The results
For anyone unfamiliar with earnings season, it goes like this: The season kicks off a few weeks after the close of a calendar quarter, and enters full swing when a cluster of big banks report their results, followed by a range of other sectors. Prior to each company’s earnings release, a handful of companies, including FactSet and Zacks, publish a consensus estimate that reflects the average of analyst predictions for a given company’s performance. When the actual earnings results go live, any meaningful divergence from “the Street’s” prediction leads investors to drive the price of the stock up in the event of a beat, and way down if there is a miss.
For purposes of the AI versus analysts experiment, Fortune and Samaya selected 50 companies that reported Q3 earnings between October 26 and November 5, covering ten different sectors ranging from health care to energy to IT. In each case, Samaya supplied an AI-based estimate for the most important metric (EPS or revenue) for a given company at least a day before its earnings release. For comparison, the exercise used an analyst consensus view supplied by Bloomberg.
When the first results rolled in, it became clear Samaya was not blowing smoke. The startup had forecast that Verizon would bring in $34.52 billion in revenue—a number closer to the actual figure of $34.31 billion than $34.77 billion, which was the consensus figure from the analysts. Likewise with UPS, where Samaya’s prediction for adjusted EPS of $1.65 was more accurate than the $1.62 consensus estimate. (The actual figure was $1.67.)
After losing to AI on the first four companies to report, the analysts struck back, producing more accurate estimates for the next two, S&P Global and PayPal. But from then on, it was a rout for Samaya, which supplied better picks for 36 of the remaining 44 companies. Its notable wins included Palantir (where it predicted $2.30 billion in revenue versus the actual figure of $2.31 billion and $2.18 billion for the analysts) and energy giant Permian Resources, where Samaya forecast the actual EPS figure of $0.69 exactly, while the Street had predicted $0.625.
Samaya’s most notable miss came with Yum Brands. The AI startup’s EPS forecast of $1.42 fell well short of the food giant’s actual EPS of $1.56, while the analysts came closer with a forecast of $1.49. Samaya also lost to the analysts when it came to forecasting the degree to which tech giants Amazon and Google would fall short of revenue expectations.
Samaya CEO Maithra Raghu points out that, where the Street consensus proved more accurate, it was typically in the case of very big companies that are the subject of large volumes of private research. Most of this research did not form part of Samaya’s training material, she said, explaining that the startup relied instead on publicly available data like filings, news reports, and interview transcripts.
Raghu says that Samaya, whose clients include banks and hedge funds, is likely to add analyst notes to training data for earnings in the future. But she also expressed a concern that doing so could bias the AI by teaching it to internalize the same sort of desire for consensus that defines many Wall Street analysts.
Earnings forecasts in the AI era
“I think there is safety in numbers,” said Sheraz Mian, Director of Research at Zacks, one of the firms that compiles analyst consensus figures. Mian is an analyst himself, so he is well familiar with how a Wall Street consensus emerges—a process that is rooted equally in human nature and what is often a tacit pressure game by companies.
According to Mian, there are times when an analyst may believe that a company’s quarterly earnings will fall well outside what most people expect. But he says that there are incentives not to go out on a limb and say so. Namely, the potential acclaim that comes with correctly publishing an unorthodox forecast is unlikely to outweigh the reputational damage an analyst will incur if such a forecast is wrong. On balance, there is an unspoken incentive to stick the herd.
This instinct to conform is reinforced by corporate investor relations teams, which are facing powerful incentives of their own when it comes to earnings targets. In an observation that is as true today as when it was written, the Fortune story from 1997 noted how, “The price of missing a quarter has risen sharply, particularly among high-priced growth stocks. In the growth stock fraternity, ‘missing by a penny’ now implies the height of corporate boneheadedness.”
And indeed, the history of earnings in recent decades shows that, while exceeding forecasts typically offers stocks a modest boost, missing those forecasts—even by a little—results in investors punishing the stock by a disproportionate measure. One result of this phenomenon, Mian explains, is that IR teams monitor individual analysts closely, and are quick to call them up, especially if they believe a forecast is too optimistic.
Even though a rule change by the Securities and Exchange Commission in 2000, known as Regulation Fair Disclosure, forbids companies from providing material information—including their forthcoming earnings—to analysts, they are still able to exert considerable soft pressure, says Mian.
“In the long run, they provide optimistic numbers that get walked down to beatable targets at the end, so a higher proportion of companies end up meeting or beating expectations right at the close,” says Paul Hribar, a professor of accounting at the University of Iowa.
Hribar also observes that, in response to all this, there has long been the idea of a “whisper number” on Wall Street—a figure that is closer to what people actually believe than what is stated in public predictions. But if AI models like Samaya’s gain traction, it’s an open question as to whether the current earnings ritual of coming up with a “consensus” number, or even a whisper version of it, will persist.
After all, the current earnings regime is highly rooted in human incentives and bias—qualities that are not shared by insentient AI models. It’s possible, of course, that the AI models will in time internalize these biases as well, and reflect them in their output—a possibility that Raghu, Samaya’s CEO, explicitly noted. But it’s equally possible that the growth of AI-based forecasts could blow up the cozy analyst-IR consensus altogether.
The apparent success of Samaya’s model also raises the question of whether Raghu, a former member of the foundational Google Brain AI team, might decide to ditch startup life in favor of trading stocks. She says the question is “not uncommon” but that her plans for now are instead to keep working on Samaya’s technology.

