How Ramp scaled to 100+ agents with AI-powered support and saves 5,100+ hours annually

Building financial operations with customer experience at the core

Ramp is revolutionizing how businesses manage their finances. As an all-in-one financial operations platform combining corporate cards, expense management, bill payments, procurement, travel, and accounting automation, Ramp helps companies save time and money at scale.

Ramp supports more than 50,000 businesses with a CX team that, unlike most support organizations, reports directly into the product organization. This is a deliberate choice that ensures customer insights directly shape what the company builds. Ramp's 100+ agents aren't just solving customer problems; they're actively influencing the platform tens of thousands of businesses rely on.

The WFM foundation: Building operational excellence

When Ramp implemented Assembled's workforce management platform in 2022, they were scaling rapidly and needed better operational control. The team used Assembled to improve agent autonomy, fundamentally changing how they manage performance.

Compliance became proactive rather than reactive. Rule flagging helped the team catch non-compliant schedule requests before they resulted in costly penalties, protecting the bottom line while ensuring compliance.

VTO and extra work functionality removed hours of manual approval work while increasing agent engagement during heavy seasons. This allowed agents to pick up hours to drive down backlogs while freeing WFM to focus on strategy rather than administrative tasks.

The challenge: When every agent built their own workflow

With workforce management running smoothly, Ramp's attention turned to how agents were handling support work itself. Their commitment to AI wasn’t new, as outlined in their AI Principles, Ramp has embraced AI since its founding to build tools that save customers time and money.

Like many forward-thinking companies, Ramp provided access to various LLMs across the organization. But without structure, every agent developed their own approach. Agents faced constant context switching between Notion, Guru, Slack threads, and internal systems. After juggling multiple sources for a complicated ticket, agents needed mental breaks before diving into the next case.

The team tracked solves per hour as their primary productivity metric, alongside average handle time, quality assurance scores, and CSAT. With 310,000+ tickets handled annually across a 100-person team, every inefficiency compounded.

The build vs. buy decision

In early 2024, Ramp faced a critical decision: should they build their own AI support solution or buy one?

Ramp's AI principles are clear: "outcomes, not interfaces" — they believe AI should be embedded in workflows to get things done, not just thin chat layers on top of a product. They also believe humans should always have the final say over irreversible decisions.

The existing partnership with Assembled influenced their evaluation, but the technical requirements were clear: an AI Copilot had to live inside Zendesk to eliminate context switching.

Assembled's approach to AI also aligned perfectly with Ramp's own philosophy, making the choice easy.

Implementation: Building trust through design partnership

Ramp took a methodical approach to rolling out an AI Copilot, named Cal, understanding that agent trust could be lost as quickly as it's gained. This reflected their AI principle of "take ownership" — giving individuals ownership of their work while deploying AI systems responsibly.

Their team focused heavily on enablement. Rachael, an agent on the fraud risk team, built out best practice documentation specifically for the CX organization. They gamified early usage by sharing leaderboards of who was doing the most actions within AI Copilot, making adoption feel like a fun challenge rather than a mandate.

Leadership also addressed anxiety about AI automation: AI Copilot wouldn't replace agents but would free them to focus on increasingly complex work as simple tasks got automated away.

The transformation: From 45% to 75% adoption

The results speak to both the partnership and Ramp's enablement efforts. Adoption grew from 45% in May 2025 to 75% across 100+ agents in early 2026. Agent satisfaction scores rose to the high fours out of five, and the team consistently saves between 30 to 90 seconds per ticket.

Across 310,000+ annual tickets, these time savings add up to over 5,100 hours saved annually. Junior agents now ramp faster with an AI partner that helps them speak in Ramp's voice from day one, shifting training focus from memorizing processes to knowing when to trust AI Copilot's answers.

Ultimately, this focus on agent efficiency is rooted in Ramp's fundamental goal of helping companies save time and money at scale. By equipping agents to move faster without sacrificing quality, the time saved per ticket ensures that customers resolve their financial operations issues quickly and with minimal friction, maximizing the value they get from Ramp's products.

Real-world impact: Stories from the frontline

For Rachael on the fraud risk team, AI Copilot transformed how she handles complex cases. Recently, a customer needed help troubleshooting a device-related issue for mileage reimbursements, which typically requires multiple sources and extensive back-and-forth.

Using AI Copilot to summarize information and prompt for missing details, Rachael escalated the issue to the appropriate team with all necessary context in one clean handoff, resulting in minimal friction for both her and the customer.

And different teams use AI Copilot differently based on their workflows. Tier 1 agents heavily use macro search with added context for common issues. Tier 2 agents rely more on draft replies for complex multi-product questions. The fraud risk team primarily uses summarization for long investigative threads, while the accounting and bill pay team favors quick actions to save time on specialized work.

The virtuous cycle: How AI improves documentation

An unexpected benefit emerged: AI Copilot became a forcing function for documentation quality across the entire company. This aligned perfectly with Ramp's AI principle of transparency, providing easy-to-understand information on how their AI systems work.

Ramp consolidated all their knowledge into Notion to support AI Copilot's context needs. But the benefit flows both ways. When the AI struggles, it signals that documentation is confusing — not just for AI, but for human employees too.

Agents can now flag documentation issues directly through AI Copilot, creating a feedback loop that improves support quality and also helps sales, account management, customer success, fraud operations, risk teams, and any roles at Ramp that touch customers.

Empowering strategic partners

Beyond efficiency metrics, AI Copilot has shifted agent engagement. The once-quiet Slack feedback channel is now active with agents sharing tips, flagging issues, and suggesting improvements.

The team intentionally avoided requiring agents to become prompt engineering experts. Instead, curious power users optimize their workflows and share best practices. Rachael exemplifies this model, leading enablement efforts that benefit all agents.

What's next

As Ramp crosses the 100-agent milestone, they’re focused on expanding AI Copilot's capabilities while maintaining their principled approach to AI. On the WFM side, automation continues driving strategic impact and operational efficiency.

They continue to operate as design partners with Assembled, providing feedback that helps shape the product — the same collaborative approach that attracted them to the partnership in the first place.

‍