Enterprise Technology Studies Uncover Challenges in Agentic AI Delivering Business Value
Agentic AI is often touted as the key to improving business operations. However, the reality is more complex. A large portion of AI projects fail to generate meaningful revenue. Reports dating back to 2019 from institutions like MIT and Gartner estimate that between 70% and 85% of AI initiatives do not deliver expected value. Despite advances, these figures continue to be cited even in 2025, suggesting that challenges remain significant.
The hype around agentic AI, especially driven by companies like Salesforce promoting it as a new kind of digital employee, raises an important question: can this technology truly deliver value? If so, under what conditions? Earlier frameworks have suggested focusing on business problems characterized by rich data, dynamic environments, and some allowance for errors, such as those framed by the Cynefin model. While useful, these frameworks are not fully scientific or definitive.
Insights from Business Intelligence Tools
Recent research sheds more light on this issue. Two scientific studies available on Arxiv—one from Carnegie Mellon University and another by Salesforce Research—offer a clearer picture of where agentic AI succeeds and where it falls short. Unlike vendor case studies that highlight successes, these studies take a more balanced approach, exploring agentic AI's effectiveness through simulations of standardized business scenarios.
Examining the Studies
Both studies assess the capabilities of large language model (LLM)-based AI agents in business contexts. The Salesforce Research project, CRMArena-Pro, focuses specifically on CRM tasks across both B2B and B2C settings. It identifies nineteen common tasks within CRM systems and organizes them into four categories of business skills. This approach allows for targeted analysis of AI performance in realistic operational environments.
Such research is essential for determining when and how agentic AI can add measurable value. It moves beyond hype to provide evidence-based insights that business and research professionals can use to guide AI implementation strategies.
For those looking to deepen their understanding of AI applications in business or seeking practical training, resources like Complete AI Training’s latest courses offer structured learning paths aligned with current industry realities.
Your membership also unlocks:
