Why AI Agents Fail: The Data Deficiencies in CRM Systems

According to projections by the research firm Gartner, over 40 percent of agentic AI projects are expected to be canceled by the end of 2027. The primary cause of these failures is often not the AI itself, but the underlying Customer Relationship Management (CRM) systems.

The issue stems from CRMs being designed for a pre-AI era, based on the assumption that humans would reliably enter data. In reality, sales representatives often enter data inconsistently, push back closing dates repeatedly, and write notes that are personal reminders rather than structured records. Fields may exist in the interface but remain blank in the database.

While humans can compensate for these data gaps through inference and contextual understanding, AI agents cannot. Missing data creates a break in the information chain. If a closing date is repeatedly postponed without explanation, an AI agent may not know how to proceed. Similarly, ambiguous notes pose a challenge. These gaps, which humans navigate seamlessly, become the precise points where AI agents falter.

Keith Peiris, co-founder and CEO of Lightfield, a company focused on re-architecting CRM for the agent era, states that agents fail because the data they work with is incomplete, inaccurate, and lacks the necessary structure for comprehension. His company has secured $47 million in funding to address this problem.

Salesforce's "State of Sales" report indicates that manual errors are the top data problem for sales teams utilizing AI agents. Sales representatives spend up to 60 percent of their workweek on non-selling activities, including manual data entry, highlighting the critical need for updated CRM systems.