A recent study led by Salesforce AI scientist Kung-Hsiang Huang reveals that Large Language Model (LLM) agents struggle with key CRM functions. They achieve a 58% success rate on simple tasks, but this falls to 35% for complex, multi-step tasks. Their handling of confidential information is particularly poor, lacking a built-in awareness of privacy. Although agents excel in single-turn workflows, they face challenges in acquiring necessary information through proactive dialogues. Marketers should be cautious of these limitations, especially in nuanced customer interactions which require dynamic problem-solving.
AI agents had a success rate of 58% on single-step tasks but only 35% on multi-step tasks, highlighting significant limitations in complex scenarios.
One of the biggest takeaways for marketers: Most large language models have almost no built-in sense of what counts as confidential, posing risks in customer interactions.
Collection
[
|
...
]