Salesforce shares key AI agent lessons from a million customer chats

Salesforce has processed over one million AI agent-customer conversations since launching its AI-powered support system in October 2024. The company’s Salesforce Help site, which receives 60M annual visits, now leverages AI agents to deliver multilingual, scalable support. Here are the three critical lessons learned:

1. Content Quality and Diversity Drive AI Success

• AI agents rely on 740,000+ content pieces, including structured (CRM data, product usage) and unstructured (forums, internal docs) sources.
• Outdated or conflicting content was a hurdle; Salesforce now uses human experts to curate and update materials.
• Key insight: Unstructured data (e.g., community discussions) provides context, while structured data enables personalization.

2. Balance Intelligence with Empathy

• AI agents need a “dynamic brain” (continuous learning via feedback) and a “caring heart” (emotional intelligence).
• Early restrictions (e.g., blocking competitor mentions) backfired; high-level guidance like “prioritize Salesforce’s best interests” worked better.
• Customers increasingly asked relationship-focused questions (e.g., “Who’s my Account Executive?”), prompting deeper contextual support.
• Human hand-off rates were adjusted from 1% to 4% after feedback showed customers still value human connections.

3. Empathy Must Come First

• Even flawless solutions fail if delivered poorly. AI agents now mirror human empathy, especially in crises (e.g., outages).
• Example: Responses begin with “I’m sorry you’re experiencing this” before technical troubleshooting.
• Result: AI agents became trusted partners, not just tools.

Key Takeaway

Salesforce’s hybrid approach—combining data-driven precision with human-centric empathy—proves AI’s potential in customer service. The company emphasizes iterative testing and balancing automation with human touchpoints.

For more AI insights, subscribe to ZDNet’s Innovation newsletter.