1/ AI + CRM: @Salesforce just raised a red flag:
LLMs like GPT-4o are not enterprise-ready for CRM without strict governance.
In tests, models struggled with multi-turn context, hallucinated actions, leaked PII.
CIOs: AI without controls = A data breach waiting to happen.
LLMs like GPT-4o are not enterprise-ready for CRM without strict governance.
In tests, models struggled with multi-turn context, hallucinated actions, leaked PII.
CIOs: AI without controls = A data breach waiting to happen.
2/ New CRMArena Pro tests by Salesforce found GPT-based assistants fail at basic CRM hygiene:
– Can’t preserve thread state
– Confuse similar customers
– Leak sensitive notes
Want LLMs in customer-related workflows? IT leaders must bring A-game in AI ops, audits, and access.
– Can’t preserve thread state
– Confuse similar customers
– Leak sensitive notes
Want LLMs in customer-related workflows? IT leaders must bring A-game in AI ops, audits, and access.
3/ CIOs, don’t rush GPTs into the heart of your CRM stack.
Salesforce’s own research shows you must layer:
✅ Prompt engineering
✅ Full identity/data controls
✅ Fine-tuning for enterprise tone/policy/safety
Without them, LLMs risk become rogue actors in customer experience.
Salesforce’s own research shows you must layer:
✅ Prompt engineering
✅ Full identity/data controls
✅ Fine-tuning for enterprise tone/policy/safety
Without them, LLMs risk become rogue actors in customer experience.
4/ I’m actually bullish medium term involving AI in customer experience. But IT depts must educate themselves.
The details on CRMArenaPro and the gap between LLMs / enterprise CRM needs in a major new paper by @SFResearch’s @steeve__huang + team:
arxiv.org/abs/2505.18878
The details on CRMArenaPro and the gap between LLMs / enterprise CRM needs in a major new paper by @SFResearch’s @steeve__huang + team:
arxiv.org/abs/2505.18878
• • •
Missing some Tweet in this thread? You can try to
force a refresh
