How Salesforce’s Own Help Portal Became a Testbed for AI Agents
By trialing its AI agents on live customer support, Salesforce surfaces what works—and what doesn’t
Salesforce isn’t shy about how it’s using its products, especially its agent development platform. When it comes to embracing AI agents, the enterprise tech firm has openly touted that it’s “drinking its own champagne.” Since Agentforce’s launch, CEO Marc Benioff has used every opportunity to showcase how the service is helping bolster his company’s help portal, the online destination customers visit when they have questions about using Salesforce products. Nine months later, Salesforce is taking a bit of a victory lap, revealing that AI agents have “handled” one million support tickets, proving that the technology is viable for any organization.
“It’s a very significant metric for us,” Bernard Slowey, Salesforce’s senior vice president of digital customer success, tells me. “The reason that we do what we do, which is have an Agentforce on our help portal, is that we should be the showcase of what this digital labor force is.” He echoes a term his boss has used in the past, referring to the team behind Salesforce’s help portal as “customer zero.”
“When we launched [seven] months ago, we’ve learned so much along the way about things that didn’t work—you’ve probably seen some of the feedback yourself of things where it didn’t give a great answer in certain scenarios,” Slowey explains. “Now that we’re at a million, we’re seeing [a] strong resolution rate. We’re seeing the percentage of handoff to humans at a number that we feel pretty comfortable about. And all of these learnings, we’ve been able to feed into our product team to help improve Agentforce as a product. It also means that my team and I…feed our learnings into our enablement teams so that they can help our customers get Agentforce implemented—that’s the most important thing.”
One million is a nice, superficial milestone for any company to hit, but it’s critical to dig deeper into the number. Salesforce has declared that its bots have handled one million requests, but how were they resolved? Did its customers have a positive or negative experience? What was the Return on Investment?
Examining Salesforce’s help portal activity from January to May in 2024 and 2025, Slowey contends traffic grew by two percent, or around 400,000 visits, the equivalent of around 18.8 million people who visited the site in that time frame. “Normally, what we see is, year over year, our help portal traffic rises and our case volume follows a similar trajectory—more customers are using Salesforce products, we’ve had new releases, etc.,” he reasons. “But if you look at our case volume at the same time, we’re actually down the human case volume. We’re actually down five percent—that’s 27,000 cases. What’s the difference? We launched Agentforce on the help portal in that time period.”
He goes on to reveal that Salesforce has seen 82,000 fewer customer cases than the company initially forecasted. This led Salesforce to reassign 500 support engineers to different roles within the company. “Because we haven’t had the demand that we thought we were going to have, because…maybe some of that was improved products…but 82,000 against the forecast and 500 people, I’ve never seen us make productions like that in my time here,” Slowey claims.
Challenges Using Agentforce with the Help Portal
While Benioff may make it sound like Salesforce’s implementation was smooth, Slowey concedes that there were some difficulties. At the start, he points out concerns about the quality of the answers being provided by the AI agents to customers. The team ultimately realized there were two things causing this problem: First, the right content wasn’t imported into Data Cloud—customers were asking the agent developer-related questions when it was not designed for those queries. Second, Salesforce had to deal with “content collisions,” where there were too many articles talking about the same thing. Slowey’s team was forced to start culling content.
“We started looking at, like, ‘okay, this content hasn’t been viewed in over a year. Just kill it.’ That was my rationale to the team,” he shares. “And so, I tell customers, you’ve got to get your content hygiene right. You have got to get the right content, but you also have to get your content hygiene right, because in a world of agents, content is key, and everything you do.”
Salesforce also realized that it was being “super restrictive” with its agent, meaning that it was being treated like “an old school chatbot” (“if this, then that; if this, don’t do that”). “What we needed to do was allow the LLM to be an LLM,” Slowey asserts. “We went into the instructions, we deleted that guardrail, and inside our global instructions, we said ‘you are a customer support representative of Salesforce. Put the best interest of Salesforce in everything you do.’ Problem solved.”
Why Being Customer Zero Matters
Salesforce’s claim isn’t just a flashy milestone—it’s a strategic signal to other organizations exploring digital transformation about what’s possible when generative AI is deployed at scale inside the enterprise. At a time when many companies are still experimenting with AI in isolated pilots or narrow use cases, Salesforce casts itself as proof that AI-enabled operations are possible. Put differently, the company is showing that it’s practicing what it preaches.
The achievement could also be viewed as a response to some of the complaints levied against Agentforce—it’s more marketing hype versus being product-ready. While it puts a lot of fanfare around the platform, Salesforce can point to the success of its help portal as evidence that Agentforce delivers results. This win is a must-have, as the tech firm faces growing competition from companies such as Sierra, run by its former co-CEO Bret Taylor, as well as Zendesk, Intercom, ServiceNow, Amazon, Microsoft, and Meta.
The pressure to demonstrate that its agentic product has legs is why Benioff and Slowey are both referring to the customer success team as “customer zero.” “We pride ourselves on being customer zero, which is, we’re using this product before it’s available…so we can learn, give feedback, and don’t customize,” Slowey says. “We want to take it out of the box and turn it on.” He goes on to add that every Salesforce product his team uses is readily available to all customers—it doesn’t have privileged access to unreleased platform features.
“We have no customization. What we need to do needs to be out of the box so that we can go to a customer and say, ‘You can do this yourself. Here’s how to do it,’” Slowey elaborates. “Now, we also want to make sure that…everything we do is on our product roadmap. So, even thoguh I say no customization, we try and be two to three months ahead of where the product is, so that we can be testing this thing, understanding if it’s working, and then if not, give feedback to the product team and say, ‘Hey, you need to fix this before we can give it to the customer.’ So, it’s a healthy tension.”
Consequently, the “customer zero” approach enables Slowey’s team to inform other Salesforce teams on how to assist organizations in modernizing their technology stack. This includes those in its partner ecosystem, its professional services group, and support teams.
What About the Human Impact?
While Salesforce celebrates this milestone, one must wonder how AI success impacts the human worker. After all, with thousands of people being laid off due to their companies embracing AI, it’s easy to assume that with Agentforce reducing the case volume, has anyone lost their job? Salesforce isn’t immune to this trend. So, while the enterprise software giant touts this benchmark, one must wonder how this affects the knowledge worker?
While Salesforce celebrates this milestone, it raises a critical question: What does AI success mean for human workers? As companies across industries lean into automation—and layoffs increasingly accompany these shifts—it’s fair to ask whether Agentforce’s ability to reduce support case volume has come at a cost. Has anyone lost their job? Salesforce certainly isn’t immune to broader workforce trends.
“For us, it was never about eliminating roles,” Slowey asserts. “It was always about redeployment…We want to be able to touch as many customers as we can touch with our humans.” He emphasizes that the 500 support engineers who were reassigned were placed in “more outbound-type roles.”
“It was always about, can we increase capacity in other teams so that we can do more, like help drive adoption, help customers that are [having trouble] with Agentforce? That was always our goal from the start, and [Benioff] has been very clear about that. It’s not about eliminating roles. We’re redeploying some of these roles into other parts of Salesforce.”
He concludes by saying, “For us, it’s about the AI and the human together that drives the customer success. We’ve got incredible support engineers—very tenured, very knowledgeable—we’re always going to need them. What will happen is that you’ll probably see in this world of agents that the volume that ends up with the human gets harder because it’ll become the more complex scenarios that really the [AI] agent can’t handle and hands over. I think that’s what will happen.”
Voice and Native Apps Coming to the Help Portal
The Salesforce help portal is currently only text-based, but that will soon change. Slowey reveals that the company will be adding voice capabilities to its agent, allowing customers to interact with it verbally if they choose. Moreover, in the near future, Salesforce is launching native iOS and Android help portal apps.
“I never would have launched an app in my time here because…people have enough apps,” he explains. “But imagine that you can have a conversation with your phone app. Maybe you’re inside your CRM and you’re trying to do something you can’t figure out. I can just talk to my help agent and it’s voice back and forth, telling me what I need to do. So that’s what we’re getting ready to launch is these native iOS and Android apps with voice in it.”
The Quest for Customer Service AI Continues
Slowey doesn’t believe AI has been perfected for customer success, but it’s coming close. But he offers this message to business leaders who have some trepidation in embracing AI:
“I talked to customers a lot and…a lot of them are afraid to go customer-facing with agents. There are all kinds of internal use cases. They’re starting with around productivity enhancements. And I facetiously say to them, ‘Are your support engineers perfect? Do they get their answers right?’ I run a large support organization. I will tell you they do not. They’re humans—we all make mistakes. The agent…it’s not perfect. It’s making mistakes. But that allows us to see why it made the mistake. Failure is the biggest learning. We can see why it made a mistake [and] we can improve.”
He acknowledges that while AI may never be perfect, “it’s getting better, week over week, month over month.”