CUSTOMER Q&A: Ramp

How Ramp is turning finance into a smarter operating system

Ramp started with a simple idea: financial products should be aligned with the customer. Instead of encouraging businesses to spend more to earn more rewards, Ramp set out to help businesses keep more money in their bank accounts. Over time, that mission expanded from saving money to saving time, helping automate expenses and accounting, and manage payments.

In this interview from Plaid Effects, Plaid’s Co-founder and CEO, Zach Perret spoke with the Ramp’s Co-founder and CEO, Eric Glyman, about rethinking business finance from the ground up by using data, automation, and AI to help companies save money, save time, and turn financial workflows into work that happens behind the scenes.

This conversation has been edited for clarity and length.

Perret: Ramp counts its age in days. Why?

Glyman: It started as a happy accident. In our first board meeting, we didn’t have revenue yet, and I said, “We’re only 133 days old.” We put it on the cover of the deck and moved on. At the next board meeting, I updated the number and realized it had been 66 days. Did we get the same amount done as the previous 66? Did we get more done or less?

We were seven people competing against companies with tens of thousands of employees. We couldn’t outwork them. So we had to figure out which hours really mattered—the hours where we got 10x done, had better sales meetings, and made decisions that shaped the strategy. Counting days became a way to set tempo and say no more often to things that were less important.

Perret: Going back to the beginning, what was the early idea behind Ramp?

Glyman: The kernel of the idea was that many financial services providers are structurally misaligned with their customers. In corporate cards, the dominant model was built around convincing businesses to spend more to earn more points. But when you really listen, people say, “I just want more in my bank account.” If you help someone avoid spending $100 on a subscription they no longer use, that is $99 more in their account than if they got a dollar back on the purchase.

The enduring mission was to save customers money and have aligned incentives. But one thing we missed early was the importance of saving people time. For companies, time quite literally is money. An hour not spent doing expenses is an hour spent selling the next customer, working on the next product, or moving the business forward.

Plaid: We’ve been talking about intelligent finance. You’ve written about programmable money. What does that mean to you?

Glyman: If you look at a lot of financial processes, especially in a corporate context, they make no sense. Take a business expense. You might use one system to pay, another system to submit the receipt, and then someone else reviews it. There can be eight people involved in one purchase, even though the whole thing was digital in the first place. You would never design it that way today. A lot of what Ramp does is look at these systems from the ground up and ask: Are they designed for the capabilities of today’s world?

When people think about money, they often think about a dollar bill or a coin. But money is information. Payments are instructions moving between databases and ledgers. If you understand money that way, fintech is increasingly about doing work, saving time, and acting on intent.

Perret: What was the moment when you realized AI was going to change how the company works?

Glyman: The first big use case came from sales, not product development. We had a sales development rep who was excellent at booking meetings, but a lot of his day was spent preparing to do the work. He would look up which companies raised money, check LinkedIn, guess emails, and write copy for outreach. Before you know it, half his day was gone preparing, and only the last part generated revenue.

We looked at that and thought, these sound like structured processes an algorithm could do. Around that time, GPT-3 had come out, and we started using it to write better copy and turn unstructured data into good emails. Within a year, he was booking about eight times as many meetings per day. The lesson was not, “How do we adopt AI?” It was, “How do we make people better at their jobs?” We’ve applied that lens to engineering, marketing, and other functions by asking where the bottlenecks are and how AI can help.

Perret: Tell us about Glass. What is it, and why build it internally?

Glyman: Everyone is using AI now, but some people are “mega AI-pilled” and flying, while others struggle to get it working in their environment. The first experience can be overwhelming: connecting apps, handling permissions, worrying about data privacy, or using command-line tools.

Glass is an internal interface that comes preconfigured. It has the right permissions, connects to internal systems like Slack and Notion, and includes skills people across the company can use. If your colleague builds a useful skill, you can use it too. When someone gets better, everyone gets better. It’s a better harness that makes it easier for people to actually go do work.

Perret: You’ve talked about token spend becoming a major concern for companies. Why does it matter?

Glyman: I think token spend is going to become a third major category of business spend.

Historically, companies managed two things: people costs and vendor costs. Very soon, there will be a third category: asking models to do work at marginal cost. For many companies, this was not in their five-year plan—or even their one-year plan. On one side, spend is far higher than people expected. On the other side, people are saying, “I can’t work without this. It makes me so productive.” Both are true.

Ramp offers spend management, so helping companies understand and manage token spend is a natural problem for us to work on. Companies need to understand the nature of the spend, how to categorize it from an accounting perspective, and how to optimize it as it becomes a major source of work.

Perret: You deploy a lot of AI tools in production. How do you think about the mistakes that inevitably happen?

Glyman: A useful mental model is accounting. Customers need to close their books accurately and stand up to an audit, but the audit process assumes people will make mistakes. That is why there are reviewers, first lines, second lines, and outside auditors.

We try to design systems that are not only tolerant of mistakes, but that anticipate mistakes. One approach is using models to check each other. You ask one model a question, then have another model grade whether the result makes sense. Having systems work together can produce a much higher degree of accuracy. A lot of software design is moving in that direction: models working together, checking each other, with clearer benchmarks, accuracy targets, and success conditions.

Perret: What do you think fintech will be talking about in a few years?

Glyman: As the cost of knowledge work goes down, the interesting question becomes: What business models and products become possible when thinking is less expensive, intelligence is more accessible, and skills that used to be scarce are available to more people? That should lead to many more businesses being created and many more useful things we can do on behalf of people. 

Once models can categorize transactions out of the box, the better question is: Now that we’ve categorized all the transactions, what are we going to build? What are we going to do for people? That is where the useful, interesting work begins.