If Salesforce CEO Marc Benioff’s claims that AI performs up to half of his company’s work are true, Salesforce appears to be way ahead of its customers.
In an interview with Bloomberg last week, Benioff said that “digital labor” now handles 30% to 50% of the company’s workload. However, Salesforce users in attendance at the Agentforce Boston event late last month appear to be working on a much smaller scale, as evidenced in discussing their own fledgling efforts to launch agentic AI projects.
At customer roundtable sessions and in keynotes, many Salesforce users said they were still in the early phases of setting up agentic AI projects. Their discussions focused on issues such as how to determine the scope of work that agents they were considering would eventually do, and how to test them.
They also discussed common challenges of pulling data from systems outside Salesforce so AI agents can aggregate customer data and perform time-saving tasks. Examples included managing healthcare appointment scheduling or summarizing a customer’s recent banking activities so a human contact center agent doesn’t have to look up bits of data across several applications for every call.
University of Massachusetts Amherst is in the process of evaluating agentic AI to help its more than 30,000 students and 1,900 faculty navigate their interactions with the administration. UMass started using Salesforce for student recruitment around 2017 and has since expanded its footprint to cover employee service and HR, as well as student services such as housing and financial aid, said Rachel Shipman, Salesforce product manager for the university’s Amherst, Mass., campus.
Shipman said an agent could potentially help students solve problems that now require a human case manager, like resolving housing issues or answering questions about bills.
Another possible use case under consideration is an agent to help students arrange study abroad programs or internships — activities that require considerable effort to pull off but also have a high correlation with student success and degree completion. It can be intimidating for a student considering these opportunities to set up an exploratory meeting with a person in the right office, but an AI agent would lower that barrier to entry, according to Shipman.
“They can start with a chat [that can] lead to an appointment with a very friendly staff person that really wants to help,” she said.
It is an initiative just kicking off. That will be really critical to getting us to the point where we can do some of these very interesting and cool [agentic AI] use cases.
Rachel ShipmanSalesforce product manager, University of Massachusetts Amherst
Beyond that, Shipman’s team imagines potential agents that could perform proactive outreach with students at crucial parts of the semester. They could send reminders to meet with their academic advisors or make their next round of course selections and help them keep on track by answering questions and setting up meetings on their behalf.
Like many companies building AI agents that would take on oft-repeated tasks, UMass has its own barrier to entry: Data for training the agents, and then ongoing access to data needed to manage cases and take actions. The university recently launched a data strategy to unlock information from traditional higher-ed applications, some of it structured and some not, and make those assets available for use outside those systems.
“It is an initiative just kicking off,” Shipman said. “That will be really critical to getting us to the point where we can do some of these very interesting and cool [agentic AI] use cases.”
Captain Rochester flies again
Meanwhile, Salesforce customer and the world’s largest distributor of semiconductor products, Rochester Electronics, holds about 15 billion chips and related parts at any given time. The company has facilities in most major global markets, and services customers in verticals as diverse as the automotive industry, healthcare and civil aviation.
Rochester Electronics’ need for an AI agent extended back into the pandemic years, when a worldwide semiconductor shortage necessitated giving customers access to real-time inventory data. Having consolidated e-commerce operations on Salesforce and connected it to its ERP prior to last year’s October Agentforce release, the company quickly released its own AI agent, dubbed Captain Rochester, the following month, according to Colin Strother, executive vice president at Rochester Electronics.
Captain Rochester was a long-retired superhero who had been previously used by the company for ads, comic books and even trading cards, fighting a battle against unauthorized sale of gray-market semiconductors, Stother said. Reincarnated as an AI agent, Captain Rochester can quickly find information on any of the company’s 250,000 products — whose descriptions carry somewhere between 5 and 10 million product attributes.
Data was a challenge for the company in setting up its agentic AI, he said. In the same way that many companies want a 360-degree view of the customer, Captain Rochester needs a continuously updated 360-degree view of products and knows every single thing about every single product. It’s constantly ingesting product information that flows into catalogs through semiconductor manufacturers’ APIs.
“It’s been a monumental effort to get all of that information in a clean format,” Strother said.
For now, Captain Rochester is — like a lot of agents — a much more capable, generative AI version of what was once a rules-based chatbot. It searches with an eye on customer intent and can handle natural language queries, he said.
But Strother envisions autonomous sales and service agents to come that will perform tasks that enhance the customer experience and drive customer success.
“That’s what’s on my mind,” he said. “Is that achievable? Not sure.”
Don Fluckinger is a senior news writer for Informa TechTarget. He covers customer experience, digital experience management and end-user computing. Got a tip? Email him.
