Salesforce Expands Agentforce Across Customer Service, HR and IT

TLDR

• Salesforce is consolidating its AI strategy around Agentforce, moving beyond Einstein’s predictive capabilities toward AI agents that can answer questions, use enterprise data and execute workflows.
• Customer service is the leading Agentforce use case, with companies using agents to handle routine requests autonomously while routing more complex or business-sensitive interactions to humans.
• Enterprise data and workflows are critical to effective AI agents, and Salesforce recommends companies start with a specific business problem, ensure the necessary data is accessible and establish clear rules for human handoffs.

Salesforce is folding more of its AI capabilities into its agentic AI system called Agentforce, as businesses move beyond using generative AI to summarize information and begin allowing AI agents to perform customer service and other enterprise tasks.

Kishan Chetan, executive vice president and general manager of Agentforce Service at Salesforce, said the shift reflected an evolution from the predictive models associated with the company’s Einstein AI brand toward systems designed to interact conversationally with users and act on enterprise data and workflows.

“Agentforce now subsumes everything,” Chetan said in an interview with The AI Innovator, describing the company’s evolving product branding. Salesforce has renamed what was Service Cloud as Agentforce Service, while some older customers and products may retain Einstein references.

Chetan oversees Agentforce Service, including customer, field, IT and HR service, and said he also assumed responsibility for the broader Agentforce platform product line roughly three months ago.

The change is more than a rebranding exercise. Salesforce is betting that businesses will increasingly use AI not merely to generate text for employees but to resolve requests with less or no human intervention.

In customer service, Chetan said, the first priority for many Salesforce clients is handling more customer questions autonomously. A second is making human representatives more productive.

For example, he cited apparel maker Canada Goose, where AI agents resolved 89% of customer service queries during the sales peak season, and Formula 1, where 70% were resolved by AI agents, Chetan said.

Chetan said some customers have reported 60% to 70% productivity gains among human representatives with AI assistance.

Generative AI as a stepping stone

Chetan said companies often start with gen AI before moving to AI agents. In regulated industries in particular, organizations often begin with lower-risk gen AI applications, such as summarizing a customer conversation or drafting a response, before allowing software to act autonomously.

“You’re a little afraid of just letting an agent in the wild,” he said. “So people start there, especially in regulated industries, but they very quickly want to move to agentic.”

Salesforce is incorporating gen AI capabilities into Agentforce while allowing customers to start with narrower applications. The company is also trying to make agent behavior more deterministic, addressing the problem that gen AI models can produce different responses to the same request.

A central part of Salesforce’s pitch is that enterprise agents can be more useful than general-purpose consumer assistants like Alexa because they can be grounded in a company’s own data, policies and existing workflows.

“The consumer-facing agent is only trained on … the data that’s there on the public internet, which is a lot, so it’s still very good,” he said. “But where I think the real value will come to the fore is when an agent is trained on the enterprise data.”

“Enterprise agents are a very different ballgame because they’re very specific and they’re focused on your data or your customer’s data.”

An AI agent handling a travel reservation, for example, can draw on a company’s policies and other enterprise data as well as workflows already encoded in its systems. “If you combine them, you have good agents,” Chetan said.

Salesforce is applying this approach to AI agents used by companies including Blink, Engine and Williams-Sonoma, according to Chetan.

Chetan said businesses set rules governing when agents should continue handling a customer and when a human representative should take over.

A subscription cancellation could be routed automatically to an employee because the company wants an opportunity to retain the customer. A luxury retailer might similarly decide that customers expressing an intent to purchase an expensive product should always interact with a person.

Another other trigger is failure. If an agent determines it cannot answer a customer’s question, the system can escalate the conversation.

Salesforce doesn’t recommend making customers repeatedly fight with an AI agent that can’t answer their questions. Chetan said companies might allow an agent a couple of attempts to clarify a request, but “don’t try to do it five to 10 times because you’ll really frustrate the customer.”

Start with the problem, not the agent

Companies considering Agentforce should first identify what they actually want to automate, Chetan said.

One starting point is a company’s website: What questions are customers already trying to answer there? Another is its interactive voice response system, where menu choices can reveal the most common reasons customers call.

The next question is whether the information and workflows required to resolve those requests are accessible. A company might need a knowledge base to answer a question, or connections to operational systems to execute something such as changing an address.

“In order to solve this, what type of data would you need in order to get it? Is it easy to get the knowledge base? Is it easy to orchestrate your workflows?” Chetan asked.

Only then does conversational design enter the picture. “We strongly focus on getting to the PoC (proof of concept) pretty quickly and really tuning it because there’s a lot of subtleties that come into play,” he added.

Voice agents, for example, may need different tones in different markets. Chetan said a hospitality company’s U.S. customers might expect an enthusiastic interaction, while feedback Salesforce has received in Germany suggests a more matter-of-fact style may be appropriate there.

Some customers have deployed in less than 45 days, he said, though he cautioned that deployment times vary with project complexity.

Customer service takes up 80% of Agentforce’s business, but Salesforce is also pushing the technology into employee-facing IT and HR operations, according to Chetan.

One newer example is UNESCO, which Chetan said is using Agentforce for IT service across 68 offices globally. Employees can make requests such as reporting a laptop problem or getting system access, with the agent either addressing the request or routing it to the appropriate team.

UNESCO handles more than 1,000 such requests a month, according to Chetan. Salesforce didn’t provide a quantified productivity gain because the organization is still measuring the results, he said.