Why Salesforce warns against AI pilots driven by corporate FOMO

This article summarizes an episode of Amplify’s video series featuring Arundhati Bhattacharya, President and CEO of Salesforce South and Southeast Asia.

Arundhati Bhattacharya, President and CEO of Salesforce India / Photo credit: Amplify

Enterprise AI often succeeds or fails because of the systems behind the agent, not the agent itself.

Arundhati Bhattacharya, president and CEO of Salesforce  South and Southeast Asia, believes clean data, well-defined workflows, cost discipline, and clear governance determine whether AI agents deliver real business value.

Agents need design before they can act on their own

AI agents put pressure on the systems beneath them. If data is messy or workflows are unclear, agents expose those weaknesses faster than traditional software.

An agent performs well only when it has the right context, follows defined processes, and allows people to review or correct its work. Companies should build that foundation in a practical order:

• Organize and govern the data layer so agents use the same facts.
• Turn work steps into systems the software can follow.
• Build agents on top of context and workflows.
• Integrate agents inside the work apps employees already use.
• Redesign processes and decide when agents need human approval.

Bhattacharya rejects the idea that agents can sit apart from the business. “It’s a stack. It’s not something sitting out there like a magic wand you wave… When the agentic layer relies on context and workflows, the output the agents give is insightful and useful,” she says.

Those choices shape the business case. Bhattacharya says companies can begin seeing results within 90 to 120 days if their systems are connected, adaptable, and built around measurable workflows.

Readiness starts with the data a company has today

Even well-designed agents depend on reliable data. Bhattacharya believes readiness begins with understanding the quality, location, and usefulness of a company’s existing information.

Her first readiness test asks whether data can support real decisions before organizations invest in additional AI tools or larger data platforms.

To pass this test and build a foundation, organizations must:

• Map where structured data sits and which systems feed it.
• Bring customer signals from emails, chats, calls, and documents into decisions.
• List and control data before choosing a model.
• Test whether data can improve one customer or employee decision.

Data readiness is a money issue, not an IT cleanup job. Poor data weakens personal offers, risk checks, and service fixes. Usable data can help a company keep customers by making routine interactions faster, more accurate, and more relevant.

FOMO leads to pilots no one owns

Prices decide whether agents become everyday tools

Trust needs checks and local choices