Earnings forecasts matter more than earnings themselves. Markets often move less on what a company reports than on how results compare to expectations. As artificial intelligence becomes embedded in future trading applications, financial analysis, and forecasts are increasingly shaped not just by human judgment, but by algorithms trained on vast pools of historical data and corporate language.

The promise is clear: AI can process more information, more consistently, than any human analyst. The risk is subtler. AI can improve forecasting accuracy in some settings while simultaneously distorting expectations through false precision, correlated models, and narrative feedback loops.

Whether AI fixes or breaks earnings forecasts depends less on the models themselves than on how they are used.

Where AI genuinely improves earnings forecasts

AI’s strongest contribution to forecasting is scale. Traditional analysts are constrained by time and coverage. Machine-learning models are not. They can ingest decades of financials, thousands of earnings calls, and streams of news and disclosures without fatigue.

Academic research supports this advantage. A large international study published in The Accounting Review found that machine-learning models outperform traditional statistical benchmarks, particularly for firms with volatile earnings, limited analyst coverage, and non-U.S. listings, contexts where human attention is sparse, and uncertainty is high. In these environments, AI reduces baseline error by identifying nonlinear relationships humans often miss.

AI is also effective at converting qualitative disclosures into measurable inputs. Natural-language models can track changes in tone, uncertainty, and competitive references across earnings calls at scale. LSEG has noted that transcript analytics increasingly rely on advanced NLP to detect early risk signals that are difficult to capture manually (reported by Reuters, 2024).

In short, AI improves forecasts when the problem is one of information overload, not strategic judgment.

Where AI starts to distort expectations

The distortion begins when AI outputs are treated as objective truth rather than probabilistic estimates.

Most AI forecasting systems produce point estimates that look precise: EPS to the cent, revenue growth to two decimals. That apparent precision can mask real uncertainty. When these outputs feed directly into models, investor notes, or trading strategies, they encourage overconfidence, even when the underlying prediction error remains high.

There is also a crowding effect. As more funds and research teams rely on similar datasets and pretrained models, forecasts become correlated. When expectations shift, they shift together. This increases the risk of sharp, synchronized revisions rather than gradual adjustment.

Market commentary has already begun to reflect this concern. The growing use of AI-driven sentiment and earnings tools raises questions about whether markets are becoming more efficient, or simply faster at reacting to the same signals. Another analysis pointed out that AI optimism is often priced into stocks well before earnings impacts materialize, increasing downside risk if results disappoint.

Humans still outperform where judgment matters

Importantly, AI does not consistently beat human analysts in well-covered firms. A comparative study of U.S. stocks from 2013–2016 found that gradient-boosted machine-learning models did not outperform analyst consensus forecasts when analyst coverage was strong and information access was rich (Farrell et al., 2018).

This makes intuitive sense. Human analysts retain advantages that are hard to encode:

• Direct interaction with management
• Understanding of strategic inflection points
• Judgment during regime shifts (rate shocks, supply disruptions, regulatory changes)

During periods of structural change, historical patterns, the backbone of most AI models, become less reliable. Humans can override models faster than models can relearn reality.

The real risk: narrative feedback loops

The most underappreciated risk is not model error, but feedback.

If markets reward AI-driven growth narratives, companies adapt their language. If language shifts, models adapt in turn. Over time, signals degrade. What began as insight becomes noise, amplified by repetition.

This is not hypothetical. Financial regulators have long warned about model risk in valuation and forecasting. U.S. banking supervision guidance (SR 11-7) emphasizes that models used in financial decision-making must be governed, validated, and challenged, not treated as black boxes.

AI forecasting systems that lack these controls do not just make mistakes. They institutionalize them.

So, does AI fix or break earnings forecasts?

AI fixes earnings forecasting when it expands coverage, standardizes signal extraction, and quantifies uncertainty.
AI breaks forecasting when it creates false precision, synchronized expectations, and narrative-driven confidence divorced from fundamentals.

The evidence points to a hybrid future. AI narrows baseline error and speeds analysis. Humans retain the edge in interpretation, regime change, and accountability.

The organizations that benefit most will not ask whether AI can replace analysts. They will ask where AI should stop, and where judgment must begin.

Conclusion

AI is neither a forecasting savior nor a market villain. It is a force multiplier. Used carefully, it improves discipline and consistency. Used carelessly, it accelerates overconfidence.

In earnings forecasting, the difference between fixing and breaking the system comes down to humility: forecasting ranges instead of points, treating models as inputs rather than answers, and remembering that intelligence without context has always been a poor guide to markets.

Benzinga Disclaimer: This article is from an unpaid external contributor. It does not represent Benzinga’s reporting and has not been edited for content or accuracy.