94% Drop in US AI Tech Stocks: NANDA Report Highlights Low Success Rate of Generative AI Projects in Commercial Organizations
U.S. stocks in AI technology companies experienced a decline at the close of trading yesterday, with the NASDAQ Composite index dropping 1.4%. Notable losers included Palantir, down 9.4%, and Arm Holdings, which fell by 5%. According to reports, Tuesday marked the largest one-day market fall since early August.
The decline may be attributed to a report released by AI company NANDA, highlighting the high failure rate of many generative AI projects in commercial organizations. Initiated at the Massachusetts Institute of Technology Media Lab, Project NANDA is an organization aiming to build an “agentic web.” The research, available for download from their site, includes data from 52 structured interviews with enterprise decision-makers, analysis of over 300 public AI initiatives and announcements, and a survey questionnaire completed by 153 company leaders.
The study found that only 5% of generative AI pilots reach production and generate measurable financial returns, with the majority producing minimal impact on profit and loss metrics. After leaving pilot status, return on investment was measured over six months.
While AI is often deployed in front-office or customer-facing business functions, successful projects are primarily found in back-office workflows, resulting in savings largely from reduced reliance on third-party agencies and Business Process Outsourcing (BPO). The survey revealed little impact of AI projects on overall internal staff levels.
Though 90% of employees reported personal benefits from using publicly available AIs like large language models such as ChatGPT, these subjective advantages do not translate to institutional-level success. Around 40% of the surveyed companies subscribe to large language models (LLMs).
Many failed projects’ owners cited a lack of contextual awareness in generative AI models, which includes adapting to circumstances, evolving over time, and recalling past queries. NANDA emphasizes that establishing partnerships with organizations capable of providing such a system and ensuring its adaptation to the specific needs of the organization is crucial for success. The paper features quotes from interviews, including between 60%-70% agreeing with statements like, “The AI system doesn’t learn from our feedback,” and “Too much manual context is required each time.”
The media & telecom vertical experienced the most positive impact from generative AI, followed by professional services, healthcare & pharmaceuticals, consumer & retail, and financial services. The energy & materials sector currently exhibits a negligible rate of generative AI project launches, the paper states. In terms of business units, sales & marketing is where most projects are or were based, with finance & procurement being the least popular for AI implementations.
Generative AI is most commonly deployed in sales & marketing, while complex tasks like client management are less likely to be assigned to an AI system. Typically, summarizing a report or writing an email would be delegated to a human on 70% of occasions.
The published report’s language and lack of academic rigor suggest its origin and purpose may be more aligned with marketing than intellectual and technological discussion. The paper’s authors advocate for strategic partnerships with knowledgeable vendors to increase the chances of generative AI projects’ success, a partnership that, coincidentally, NANDA can offer one half of. They note “unprecedented opportunities for vendors who can deliver learning-capable, deeply integrated AI systems” in their conclusions.
The findings from the NANDA report paint a challenging picture for decision-makers responsible for generative AI implementations. However, the report’s underlying messages may be weakened by its publication’s intent, potentially influenced by partisan surveys from authors with vested interests. It appears more likely that the NANDA report reflects broader concerns on Wall Street about the practical effectiveness of generative AI as a business tool.
