Inside Salesforce’s Struggles With Its Flagship AI Product

Last fall, Salesforce debuted an AI agent so remarkably human-like it seemed like a vision from the future.

In a sleek demo, the company showed how luxury retailer Saks Fifth Avenue was using Salesforce’s new flagship AI software Agentforce to create “Sophie,” a charming, patient customer service representative. A Salesforce executive dialed a Saks hotline, the chipper-voiced Sophie picked up, and then deftly recommended a sweater based on his order history and talked through shipping options.

Sophie reflected the promise of AI agents, which have had no bigger booster than Marc Benioff.

Soon after unveiling Agentforce, the Salesforce CEO wrote an essay in Time Magazine, which he owns, declaring that agents would unleash “a revolution that will fundamentally redefine how humans work, live, and connect with one another from this point forward.” And he would lead this revolution, he boasted. In an interview with Business Insider around the same time, he said that Sophie was already live on Saks’ website. “That is what our customers are able to do” with Salesforce’s AI tools, he said, unlike those from one of his chief competitors. “Microsoft doesn’t have these examples, actually.”

In the weeks after Agentforce’s debut, Salesforce’s stock surged by more than 50%, peaking at an all-time high last December.

But a year later, shares are slumping as competitors’ stocks keep climbing higher. Some Salesforce investors, analysts, clients, and even employees say Benioff may have placed too many chips on a bet that is yet to pay off.

Salesforce’s public disclosures show that fewer than half of the company’s 12,500 Agentforce customers are paying. Of its total customers, less than 2% were having more than 50 Agentforce conversations per week as of this summer, according to people with knowledge of internal reports. A Salesforce spokesperson said those numbers are just one cut of data that doesn’t represent actual adoption or how the tool is being used.

Sophie is also not live as it debuted as. Today, when a customer calls Saks’ hotline, a robotic voice answers, and responds to basic commands before transferring customers to a human representative — an experience virtually indistinguishable from the technology that banks and airlines have been using since the 1980s. In August, Saks announced it’s partnering with Amazon and AI company NLX on a “new AI-powered virtual voice assistant named ‘Sophie.'”

Inside Business is today’s must-read story on what the forces shaping business, tech, and innovation mean for you. Each piece delivers insights to help you make smarter career, money, and life decisions.

Adam Evans, an executive vice president at Salesforce who heads Agentforce, tells Business Insider he can’t “comment on exactly what’s happening with Saks,” but that “actual tangible return on investment value is being created across thousands of our customers.” A company spokesperson adds that Salesforce customers have used Agentforce in more than 3 billion instances to date and that customers from Williams Sonoma to Singapore Airlines to the South American retailer Falabella “are seeing value as they ramp up use cases.” PepsiCo transformation officer Athina Kanioura tells Business Insider that the company is using Agentforce to help 1.5 million retail stores manage their orders of its products and is targeting 5 million stores by the end of 2026.

The struggles with agents and generative AI in general aren’t unique to Salesforce. An MIT report in July found that despite $30 billion to $40 billion in enterprise spending on generative AI, 95% of organizations report no return on investment. The management consultancy Gartner recently predicted more than 40% of agentic AI projects will be canceled by the end of 2027 due to “escalating costs, unclear business value, or inadequate risk controls.”

It’s not that Salesforce is necessarily behind competitors when it comes to agents, says Gil Luria, head of technology research at D.A. Davidson. It’s that the company “bet the farm” on an early and unproven technology and neglected their core business. “Salesforce became so focused on their AI product while the rest of their business … was decelerating very sharply,” he says. “They were paying too much attention to this new thing and not enough to what got them there.” A Salesforce spokesperson said Agentforce is part of its core business and “enhances every Salesforce application by making it agentic.”

“Salesforce is uniquely vulnerable because we don’t have a cloud business to fall back on,” one senior employee says. “Microsoft has plenty to fall back on, Amazon has plenty to fall back on.”

If you are a normal business with normal admins, you do not have the expertise to set this up.Kristi Valente, a Salesforce administrator since 2008

Agentforce is Salesforce’s No. 1 priority, according to a draft of the company’s annual strategic plan for its upcoming fiscal year. The annual plan is typically finalized by February. “Win the enterprise agent wars” is listed as the first method of success for Salesforce, dubbed “the agentic enterprise” in the internal document.

The company’s stock is down more than 20% year to date, while peers like Oracle and Microsoft are up around 55% and 24%, respectively. Salesforce’s revenue growth rate earlier this year slowed to single digits for the first time in its history as a public company. Another pressure point: An activist investor that pushed Salesforce into significant cost-cutting in 2022 (before selling many of its shares) increased its stake by nearly 50% this summer.

Related stories

The stakes are high for every Big Tech company in the agent wars. For Salesforce, losing may be catastrophic.

Inside the company, some current and former employees say there’s been constant struggle for the teams scrambling to deliver on Benioff’s public promises of what their AI products can do.

Demos of software to customers or at Dreamforce, Salesforce’s annual blowout conference, often showcase concepts that are months or years away, and the company sometimes shifts direction before these products are built.

“It’s very, very difficult — even for people working on the products — to know the difference between what we say in a demo, what’s on a road map, and what’s actually in production,” one senior employee said. “It’s a full-time job just figuring that out.”

Some insiders say the current version of Agentforce, requires significant technical skill from Salesforce’s clients. Kristi Valente, a director of sales operation for a large appliance and electronics distributor, has been a Salesforce administrator helping companies use its software since 2008. She says the Agentforce demo her company received made it look like the technology was a lot easier to set up and less expensive to use. Her company purchased it in April and has yet to set it up.

“If you are a normal business with normal admins, you do not have the expertise to set this up,” she said. “We hired a consultant and even the consultant is confused. It’s too new for anyone to be an expert.”

In a statement, a Salesforce spokesperson said that since launching Agentforce, “we’ve learned a lot about the complexities of building agents,” adding, “We are committed to partnering with our customers as they embark on their technical journeys to becoming agentic enterprises.”

“We’re learning not every agent or use case is the same,” says Evans, the head of Agentforce. “A lot of the agents haven’t gone to scale because prompts alone are not enough. You need more.” For example, Salesforce has learned clients may not want an agent to be as creative as a chatbot when it comes to answering direct questions from customers, like how to process a return for an online order. For this reason, Salesforce has updated Agentforce with a “reasoning engine” to make sure the agent’s answers and behaviors are more predictable and reliable.

Some of Salesforce’s own salespeople said they struggled to grasp how the AI product they’re selling works or what it can do.

Related stories

Salesforce required Agentforce training for salespeople, asking them to build, for example, a concierge agent at a fictitious hotel chain that could help a patron with things like booking a restaurant or renting beach chairs. Agentforce in one of those trainings required a “tremendous amount of work” even to set up basic functions like dinner reservations, an employee who participated in the training said. “Anyone who took that training knows Agentforce isn’t ready for primetime.”

“We invest in comprehensive, hands-on Agentforce training for all 75,000+ of our employees because we recognize the significant shift that large language models represent,” the Salesforce spokesperson said in response to the complaint. “This new software environment is fundamentally different from traditional design, which is why change management and knowledge transfer have never been more important.”

Kash Rangan, Goldman Sachs’ top software analyst, said at a briefing following the investment bank’s technology conference in September that analysts in general are giving “zero credibility” to the optimistic comments from software companies like Salesforce and executives like Benioff when it comes to AI. “They have a long way to go,” Rangan said of Agentforce. “It needs to scale.”

Rangan added that Benioff, prone to grandiose predictions and bullish forecasts, was more measured and not “extremely optimistic” at Goldman’s conference. “They are in the fight, and he’s being realistic about where they are right now,” Rangan said.

Despite that note of sobriety, and despite Agentforce’s struggles, Benioff has leaned into his signature quality: Extravagance.

During that same Goldman talk, Benioff discussed an ambitious target for Salesforce: $100 billion in annual revenue, nearly three times the revenue for its most recent fiscal year.

The grandiosity continued a month later at Dreamforce — the greatest, and most surreal show in SaaS — a three-day jamboree sprawling across downtown San Francisco that’s part Davos, part Disney World, part Radisson banquet hall, part Burning Man, and entirely Benioff.

The conference, which 50,000 people attended and which one executive once told BI costs Salesforce around $100 million per year, is studded with happy hours, concerts, celebrities, and even mascots. Signs heralded Gen Z-coded slogans like “You have an agentic aura,” and “Dreamforce, it’s a vibe.” Movie star Matthew McConnaughey, whom Salesforce pays a reported $10 million a year to be a “creative director,” held a session where he read poetry with journalist Maria Shriver. A “special guest” turned out to be Trump’s AI czar David Sacks, who posted a photo with Benioff kissing him on the cheek. There was a gift shop where people lined up to buy plushies of one of Salesforce’s mascots, “Codey the Bear,” and copies of Benioff’s 2019 biography “Trailblazer.” There was a Metallica concert. There was a Hawaiian conch-blowing ceremony. Gold sequined sweatshirts were handed out to at least one attendee in an annual tradition for achievements in the community of people who use Salesforce. “I hate that I’m in a cult,” one Salesforce customer said while describing this “Golden Hoodie” award.

“Dreamforce is a marketing event, it is a sales event, not an engineering event,” one senior employee said. “It is to create hype and invite big customers and take them out to parties.”

At Benioff’s AI agent-focused keynote address, his salesmanship was on full display. He paused his speech several times to shake hands with celebrities like musician will.i.am and titans of business like Michael Dell. He talked about the death of his “close personal friend,” famed primatologist Jane Goodall, and quipped to a PepsiCo executive in attendance how much his friend “Matthew” (McConnaughey) loves Doritos.

He said his company has not been “perfect” in rolling out the Agentforce product. Later, at a briefing for reporters, Benioff shut down a question about the lack of adoption of Agentforce. “It’s the fastest-growing product in our history,” he said. “There’s never been a faster-growing product. That we have that much already is way beyond anything we could have expected.”

At Dreamforce, the company also hosted its first investor day in three years. Benioff took the stage with the company’s senior leadership team and cracked jokes while revealing new guidance — like a slightly more measured $60 billion revenue target by 2030, and returning to double-digit revenue growth. It won over some analysts. “We came away feeling incrementally better after the investor day,” William Blair analysts said in a note. “Despite slow agentic AI adoption, we don’t think Salesforce is a technology laggard,” Macquarie analyst Steve Koenig wrote in a note after the session.

How much time can the Benioff show continue to buy Salesforce as it seeks to compete in the agent wars against companies like Microsoft, Google, and Amazon with seemingly unlimited resources and mammoth, diversified businesses to fall back on? Several Salesforce salespeople told Business Insider that in the days after Dreamforce, it’s already getting easier to sell Agentforce. “Hype needed to build,” one said.

Ashley Stewart is a chief technology correspondent at Business Insider. She reports on enterprise technology companies including Microsoft and Amazon Web Services from Seattle.

Contact this reporter via email at astewart@businessinsider.com or Signal at +1-425-344-8242. Use a personal email address and a nonwork device; here’s our guide to sharing information securely.

Business Insider’s Discourse stories provide perspectives on the day’s most pressing issues, informed by analysis, reporting, and expertise.