Shares in the artificial intelligence lending platform Upstart fell sharply on Wednesday after the company revealed its own AI model intentionally reduced loan approvals, causing it to miss key performance targets. The stock dropped 14.8% despite the company reporting stronger-than-expected profits for the third quarter.
The core issue for investors was not profitability, but a significant shortfall in loan origination volume. Upstart facilitated $2.9 billion in loans, well below the $3.3 billion analysts had anticipated. Company executives explained this was a deliberate action by their AI, which detected economic risks and automatically tightened its lending criteria.
Key Takeaways
- Upstart's stock declined 14.8% following its Q3 2025 earnings report.
- The company missed loan origination targets, reporting $2.9 billion against an expected $3.3 billion.
- Executives attributed the miss to their AI model, which they said "overreacted" to macroeconomic signals and tightened lending standards.
- Despite the lending slowdown, Upstart reported a net income of $32 million, significantly beating analyst forecasts of $10.2 million.
- Analysts expressed confusion over the AI's behavior, raising questions about the model's reliability and differentiation in the market.
An AI with a Mind of Its Own
Upstart's business is built on using artificial intelligence to underwrite loans more effectively than traditional credit scores. However, in the third quarter, that same technology became the source of its problems. According to company leadership, the AI model perceived elevated risks in the broader economy during the summer and took pre-emptive action.
Paul Gu, Upstart's Chief Technology Officer, explained during an earnings call that the model is designed to react with "speed and precision" to changing conditions. "A few months ago, that led the model to tighten on credit while certain risk signals were elevated before recently normalizing," Gu stated.
This automated conservatism directly impacted the company's ability to convert loan applications into funded loans. The company's conversion rate fell from 23.9% in the previous quarter to 20.6% in Q3. This slowdown occurred even as consumer demand for loans reached its highest point in over three years, indicating a missed opportunity for growth.
By the Numbers: Upstart's Q3 Performance
- Stock Price Drop: 14.8%
- Loan Originations: $2.9 billion (missed $3.3B target)
- Net Income: $32 million (beat $10.2M estimate)
- Earnings Per Share: $0.23 (beat $0.08 estimate)
- Conversion Rate: Down to 20.6% from 23.9%
CEO Calls It a 'Speed Bump'
CEO and founder Dave Girouard acknowledged the model's aggressive pullback, framing it as an overcorrection. He described the event as a "speed bump" in the company's path.
"I think in some sense, having a model that overreacts is better than having ones that underreact because it did revert," Girouard told investors, suggesting the AI's caution was preferable to reckless lending.
Despite the operational hiccup, Upstart delivered strong financial results on the bottom line. The company posted a net income of $32 million, a stark contrast to the $6.8 million net loss reported in the same quarter last year. This translated to a diluted earnings per share of $0.23, nearly triple the consensus estimate of $0.08.
Total revenue also saw a significant year-over-year increase of 71%, reaching $277 million. However, even this figure slightly missed the analyst consensus of $280 million, adding to investor concerns.
The Promise of AI Lending
Upstart and other fintech companies have long promoted AI as a superior tool for assessing credit risk. The technology promises to look beyond simple FICO scores to analyze thousands of data points, theoretically enabling lenders to approve more borrowers at lower risk. The recent actions of Upstart's model, however, highlight the potential for volatility and unpredictability when automated systems react to complex economic data.
Analysts Remain Skeptical
Wall Street analysts were not entirely convinced by the company's explanation. The AI's independent decision to scale back lending introduced a new layer of uncertainty into Upstart's business model, which is supposed to be its key advantage.
In a note to clients, John Hecht, an equity analyst at Jefferies, described the company's reasoning for the model's behavior as "confusing to us and is in contrast to the overall market." The sentiment was that while other lenders were operating under one set of market assumptions, Upstart's AI was acting on another.
David Scharf, a managing director at Citizens Bank, echoed these concerns. He noted that the incident "raise[s] questions regarding how differentiated [Upstart's] model is compared to other machine learning-based lenders." Scharf added that such "bumps or misreads inject a dose of incremental uncertainty" into the company's core value proposition.
Following the report, Citizens Bank slightly lowered its 2026 adjusted EBITDA estimate for Upstart from $316 million to $313 million, signaling a more cautious outlook. The market reaction suggests that for an AI-driven company, demonstrating control and predictability over its core technology is just as important as delivering profits.