Salesforce CMO Warns Against Unsupervised AI
Ariel Kelman, President and CMO of Salesforce, warned that the most dangerous marketing task to automate without human oversight is anything that directly interacts with customers. Speaking on the MarTech Podcast, Kelman stated that failures in AI agent implementation almost always stem from poor data foundations and a lack of change management. He also noted that an estimated 95% of generative AI pilots fail to deliver business impact because they lack the necessary context or groundwork.
- The 95% failure rate for generative AI pilots often stems from a "learning gap," where tools that perform well in demos fail to integrate with enterprise workflows and data realities. A primary reason for failure is not the AI technology itself, but poor data quality, lack of clear business goals, and weak change management. - Many companies misalign their AI investments, concentrating budgets on sales and marketing pilots where ROI is often lowest, while neglecting back-office automation in areas like compliance and internal workflows, which typically yield higher returns. - Human oversight is critical for mitigating risks like AI bias, which can alienate customers, and for maintaining brand voice, which AI cannot fully replicate. Human judgment provides essential context that data alone cannot capture, such as understanding the nuances of an industry event or a shift in competitive dynamics. - For customer-facing roles, AI's lack of emotional intelligence is a significant risk; it can misunderstand a customer's emotional state and provide responses that feel insensitive. This can lead to what is known as "silent churn," where customers leave due to frustrating automated experiences without formally complaining. - In creative workflows, human-AI collaboration is framed as augmentation, not replacement. AI tools excel at generating variations, automating repetitive tasks, and providing initial ideas, which frees up human creators to focus on artistic vision, emotional depth, and strategic decisions. - Practitioners are increasingly using multi-tool, node-based AI workflows to chain different models together for complex creative pipelines. Platforms like Krea and Freepik Spaces allow creators to visually connect different AI tasks, such as image generation, upscaling, and media processing, into a cohesive, automated sequence. - For developers building AI tools, a new category of CLI-based agents and AI IDEs like Cursor, Windsurf, and Warp are emerging. These tools integrate AI directly into the development environment for tasks like code completion, refactoring, and debugging, with some even allowing developers to bring their own models. - Salesforce itself is a primary user of its own AI technology, deploying agents that have already handled millions of customer support conversations. One internal SDR agent generated $60 million in incremental pipeline in three months by following up on leads that were previously unworked.
