How Ramp applies its engineering mindset to customer experience.

Ramp is on a mission to save its customers the two things they care about most: time and money. With an integrated platform that spans corporate cards, expense management, procurement, travel, and accounting automation, the company has reimagined how finance teams operate. But that transformation extends beyond software features. For Ramp, delivering on its promise requires a customer experience that's just as robust.

"Money is one of the most sensitive parts of any business, so performance and trust are non-negotiable," says Ben Levick, who leads customer operations and conversational AI products at Ramp. "Support isn’t just about resolving issues—it's about helping customers get more value from Ramp."

From homegrown to high-impact

In 2023, Ramp built an in-house AI assistant to deflect simple inquiries. While the results were promising, the technical burden of maintaining it became clear. "We got to a point when building multi-step workflows and user-specific actions where we realized that support automation wasn't our core competency," Ben explains. "We needed a partner whose focus was doing this at scale."

A tipping point came when Ramp started to scale its AI agent beyond Q&A. "Once we moved into personalization—looking up user data, taking actions for customers—we knew we had to move beyond our internal tooling," Ben says. "It wasn’t sustainable to manage that level of complexity ourselves with a small team."

Why Sierra

Ramp needed an AI agent that could evolve as quickly as its own product. One key requirement: the ability to retain full engineering control without sacrificing speed. "Sierra's Agent SDK stood out because it let us write customer journeys as code, track changes, and build sophisticated logic without giving up our development workflows," says Ben. "It was the first solution that truly met our technical bar. It wasn’t a handoff of control. It felt like an extension of our engineering practice."

Just as important was the collaboration. Rather than acting as a vendor, Sierra partnered closely with Ramp to co-develop the agent. Ben describes the Sierra team as "culturally aligned and deeply hands-on," enabling Ramp to move quickly while learning to build and maintain journeys independently. "They’re not just shipping features—they’re teaching us to fish," he adds. "That’s made all the difference."

Confidence built in

For Ramp, one of the biggest advantages of working with Sierra is knowing that trust and safety are handled at the infrastructure level.

“If we built our own system end-to-end, we’d have to own every dimension of abuse prevention and risk mitigation ourselves,” Ben explains. “With Sierra, we get to focus on building great customer journeys—knowing that safeguards like jailbreak detection, abuse monitoring, and supervisory models are already in place.” That peace of mind extends to conversation quality as well. Sierra’s orchestration layer and triage capabilities help prevent drift or rogue behavior, ensuring the agent stays on task without becoming overly rigid. “It’s a balance we don’t have to manually tune ourselves,” Ben says. “And that allows our team to stay tightly focused on the definition, design, and continuous improvement of the journeys themselves.”

Ramp also benefits from Sierra’s testing suite, which simulates conversations to test agent performance at scale before launch. This AI-driven testing framework supports regression checks, predictive analytics, and outcome-based evaluations—making it easier to validate changes, avoid unintended behavior, and anticipate business impact. “It gives us a level of testing we couldn’t build ourselves,” Ben adds. “We can ship confidently knowing the agent’s behavior is stable, reliable, and aligned with our goals.”

Designing for every user

Ramp's agent supports a wide and growing range of both user-initiated and admin workflows. Ramp cardholders can lock or replace a card with ease, or get quick explanations for transaction details and failures. The agent handles collecting and validating address information, streamlining the process for users who need to update their records—it can even ship a physical card. For admins, the agent can coordinate bulk actions to help streamline work, such as issuing multiple cards or pulling specific transaction data from across a business. These automations make operational tasks faster and more reliable.

All of these are built using the Agent SDK, ensuring that the journeys are easy to maintain, version, and expand. Ramp has also been experimenting with different interface styles to suit different users. “Some of our cardholders just want to get in and out fast,” Ben notes. “Others—like admins—want visibility and control. We’ve designed the agent to flex for both.”

Real results, real impact

Since launching the agent, Ramp has achieved a 90% case resolution rate through automation—dramatically reducing the volume of routine tickets and helping prevent support backlogs. This has allowed Ramp’s support specialists to focus their time and expertise on the most complex, high-value cases where their impact on the customer experience is greatest. “The Sierra agent takes care of the straightforward issues,” Ben says, “so our team can concentrate on the harder problems that really benefit from a human touch.”

Ramp also maintains a rigorous quality assurance loop, feeding data from Sierra back into its product ops and knowledge management teams. These teams manage not only product enablement for human agents, but also for AI. “We give product operators ownership of AI success in their domain,” Ben explains. “They’re reading transcripts, updating knowledge, and shaping journeys—just like they do for our support team.”

The team has already started rolling out their voice agent, which Ben says has evolved quickly: "Even six months ago, voice agents still felt stuck in the uncanny valley. But Sierra's voice experience now handles context, follow-ups, and tone in a way that feels genuinely helpful."

Advice for others

Ben believes that AI agents are only as good as the teams that support them. At Ramp, knowledge management is embedded within the product org, with dedicated product ops partners tasked with enabling AI the same way they enable their support team. Weekly feedback loops help prioritize updates, whether it's refining a knowledge article or scoping a custom journey.

That emphasis on cross-functional ownership has made a big difference. "When an AI agent fails, it’s often not because the model isn’t smart enough—it’s because the underlying knowledge is vague, or the prompt isn’t specific," Ben explains. “If you empower your team to be curious rather than skeptical, you’ll see major gains.”

His second piece of advice is to not underestimate support teams. “They often have the most context about what customers actually need,” Ben says. “If you invest in helping them work with AI—whether through process, training, or tools—they become your strongest operational thinkers. AI agents aren't set-it-and-forget-it. But if you invest in the right tooling, team structure, and partnerships, the payoff is massive.”

What's next

Ramp is now expanding the agent across all support channels and exploring non-support use cases, where AI can support a growing conversational layer across the product. They're also interested in more autonomous patterns—agents that don't just respond, but observe and act based on defined signals. “In the future, we will see AI agents that monitor for certain conditions and then act—like a workflow trigger that lives inside a conversation,” Ben explains.

Ultimately, Ramp sees AI agents not as a novelty, but as a durable part of their operations. “We’ve only scratched the surface,” Ben says. “The more we build, the more opportunities we see to integrate AI deeply into how we serve and empower our customers.”