Ramp founder Eric Glyman on the many ways AI is changing corporate spending

Ramp’s business today has expanded well beyond its original card interchange model. While cards remain the largest revenue source, multiple newer lines of business are growing rapidly and are on track to collectively comprise the majority of Ramp’s gross profit by the end of this year.

• Card interchange is the original and still-largest revenue stream, where businesses issue Ramp cards to employees and the company earns interchange on transactions.
• Bill payments and software, a roughly two-year-old line, is already over $100 million in annual revenue. It covers ACH, wires, checks, foreign exchange, multi-entity management, automated accounting, and procurement.
• Treasury, about a year old, holds several billion dollars in deposits across checking-like products and investment/money market offerings.
• Procurement and travel are the newest lines, with procurement being one of the fastest-growing.
• Ramp now processes more than 2% of all corporate and small business card transactions in the United States, and the company grew faster last year than the year before.
• The average Ramp customer grew revenue by 16% last year, roughly three times the US business average of about 5%, which Ramp attributes to the efficiency gains its platform drives.
• Ramp helps the average company cut expenses by about 5% per year, and Glyman frames the company’s mission as becoming a “digital brain” for organizations to allocate resources and eliminate wasted spend.

The correct expense policy is a question Glyman finds fascinating because most companies operate at one of two extremes: either completely permissive (the Netflix “no rules rules” approach) or rigidly prescriptive with detailed dollar limits and advance-booking requirements.

• Ramp can back-test the relationship between expense policy strictness and business outcomes like growth pace and margins. High-growth companies tend toward permissive policies with trust-but-verify oversight.
• The breakthrough of the past year is that companies can now write expense policies in plain English and have an LLM-powered agent review every transaction against that policy in real time. Ramp is processing over 100,000 expenses a day through agentic review with over 99% accuracy, higher than human reviewers.
• The legal requirement driving expense review is Sarbanes-Oxley’s separation of duties, which prevents self-certification but does not require that a human personally do the reviewing. An agentic system with a full audit trail satisfies the requirement.
• Glyman argues the deeper problem is instilling a founder’s moral code, where employees treat company money as their own, as companies scale from a few people to thousands. Neither extreme, draconian rules nor total permissiveness, captures the nuanced judgment calls that require context about business outcomes.

Bill payment has proven uniquely resistant to modernization. Despite decades of technology, most business-to-business payments still flow through PDF invoices with manually verified bank details.

• The reason is partly that accounts receivable and accounts payable are adversarial: controllers want to pay as late as possible and collect as early as possible. Checks persist partly because the float benefits the payer.
• Other payment networks have been upgraded in place: credit cards went from carbon-copy authorizations to real-time modem-based systems, and checks went from physical transport to digital scanning under the Check 21 Act. But the loose network of PDF invoices between businesses has resisted similar upgrades.
• Glyman points out that there is no “DNS for companies” that would let you look up a business’s verified bank details in a central clearinghouse, which would eliminate an entire class of spear-phishing attacks.
• He also highlights an inefficiency in the accounts receivable chain: a small business owed money by a company like Google, which can borrow at roughly Treasury-plus-100 basis points, may itself have to borrow at 18% because the lender doesn’t see that the receivable is effectively backed by Google’s creditworthiness. Better data sharing could close this gap.

AI and software engineering are blurring traditional roles within Ramp. Designers are shipping code, marketing reports into the CTO and produces work that Glyman calls some of the best he’s seen, and customer support agents push code to production.

• The time between identifying a problem and fixing it is shrinking dramatically. An engineer can ask an AI tool to change a button color and have it done and verified within minutes.
• Glyman raises the question of whether lines of code are becoming a liability rather than an asset. In the future, codebases might be written as outcome specifications, with models rewriting the underlying implementation each year as they get smarter, rather than accumulating technical debt in the traditional sense.
• This works for growth engineering teams that don’t require four nines of uptime, but not for safety-critical systems.
• The fitness function for software companies is shifting: if a product requires very few tokens to replicate (like a simple expense app), it will evaporate. Companies that do deeper work underwriting, financing, automated accounting are harder to replicate because the token cost to recreate the system exceeds the value it delivers.
• Glyman agrees that data moats are becoming more important than ever. He gives the example of vLex, a 25-year-old company that digitized every legal record in Spain and built a $20 million bootstrapped SaaS business that jumped to $100 million in a year once AI made that proprietary dataset actionable. Similarly, DomainTools runs daily WHOIS lookups on every website, building a historical record that is impossible to recreate retroactively even though each individual data point was free.
• He also points to what he calls the “dark matter moat”: the accumulated edge cases and local knowledge embedded in a product that make it extraordinarily difficult to clone. If you asked Claude to clone Ramp, you’d get something that looks right on the surface but misses millions of edge cases that represent years of real-world problem-solving. The example he gives is Waymo encountering a thundering herd problem when the power went out in San Francisco, a corner case they hadn’t anticipated.

Ramp’s spend data offers a real-time window into the US economy that Glyman believes is more informative than official government surveys.

• The Census Bureau has reported that only a single-digit percentage of US businesses have adopted AI, but Ramp’s data from over 55,000 businesses shows the majority are already paying for AI tools like ChatGPT, Anthropic, Cursor, or Cognition.
• GDP growth has reaccelerated from the 1-2% range to 4-5%, and Glyman argues the growth is more durable than critics suggest, even accounting for subsidies from the Big Beautiful Bill and tariff effects.
• Businesses are getting more efficient at finding tools that help them grow. Glyman notes that “a penny saved is a penny earned” understates the value mathematically: the average American business has an 8% profit margin, so saving one penny is equivalent to earning 12 pennies in revenue. This efficiency gain explains why Ramp customers materially outgrow the average US business.

How Ramp saves businesses money involves both hard-dollar savings and time savings.

• Time savings: Employees and managers no longer spend hours reconciling expenses line by line. Glyman notes that people chronically undervalue their own time, and freeing up even one person to do higher-value work is significant.
• Single-use cards and merchant blocking: Ramp pioneered the ability to create one-time-use cards and to block entire merchants across all cards simultaneously. Companies can negotiate exclusive deals (e.g., Uber only, no Lyft) and enforce them automatically. They can also set spend caps per vendor that auto-decline after a threshold, triggering a conversation with the vendor’s sales team.
• Contextual controls: Cards can be programmed to auto-decline in certain contexts (e.g., an Uber ride on a Saturday) but allow the employee to text a justification that reactivates the card.
• Vendor data and benchmarking: Across hundreds of millions of purchases, Ramp knows what businesses pay per seat for SaaS tools. Before a customer renews a contract, Ramp can show them in real time that they are paying 20% more than the market rate, or that the timing of their renewal (January 1 vs. December 31) significantly affects pricing.

Ramp’s strategy and future differentiation is evolving as the company scales.

• In the early days, Ramp’s advantage was velocity: moving faster than banks to ship product. As the company has grown, the differentiation has shifted toward selling time rather than selling money. Competing financial institutions offer lower cost of capital or better rewards; Ramp offers to make expenses, accounting, and procurement just “done.”
• Glyman is exploring whether Ramp can aggregate buyer demand to negotiate bulk discounts for its customer base, similar to how Costco uses collective bargaining power with suppliers. Ramp already has visibility into where billions of dollars flow and could theoretically direct spend toward preferred vendors.
• He also sees an opportunity in what he calls “merchant-specific balances”: pre-committing spend to a vendor in exchange for a discount, similar to how gift cards work. This gives vendors demand predictability, which is valuable even to companies like Amazon with very low cost of capital.
• Glyman frames the philosophical question as where to invest the next marginal hour: negotiating a better FedEx rate for customers, or automating all of accounting for them. He believes the latter is a vector no financial institution can compete on.
• He notes that Ramp powers more than 2% of US corporate and small business card transactions, which sounds impressive but means 98% of spend is still not on Ramp, suggesting enormous remaining opportunity across cards, bill payments, and other payment methods.

Capital One, where Glyman previously sold his company Paribus, is one of the most important and underappreciated founder-run financial institutions.

• Founded by Rich Fairbank and Nigel Morris, Capital One’s origin traces to the 1980s when they pitched banks on the idea of using data to extend credit to people just below the traditional credit score cutoff, charging higher interest rates until borrowers proved themselves. Banks rejected the idea until Signet Bank in Virginia let them run it as an internal division.
• They called their approach Information-Based Strategy (IBS), using data pockets to test targeted offers at scale. The division became so profitable it was spun out as Capital One in 1994.
• Capital One was wildly experimental, at one point the largest customer of the US Postal Service due to the volume of direct mail offers. They wandered into areas like cell-phone financing and healthcare financing before refocusing.
• After buying a bank (Hibernia Bank in Louisiana, post-Katrina), they had to reconcile their culture of radical experimentation with the constraints of being a regulated institution, where regulators can show up the night a bank fails.
• Capital One’s talent pipeline has been extraordinary: nearly every modern fintech’s head of risk came from Capital One. The company hired smart people rather than banking people, creating a talent brand similar to McKinsey’s, where working there became a credential that opened doors elsewhere.

Treasury and where businesses keep their money: Glyman sees a massive opportunity because the current system is deeply inefficient for businesses.

• The national average interest rate on business checking accounts in the US is 0.07%, even though the federal funds rate is far higher. Institutions capture the spread and share almost nothing with customers.
• Ramp Treasury’s rapid growth is driven by a vastly better deal: customers keep minimal walking-around money in checking and move the rest into higher-yield accounts, with funds flowing back into checking only as needed (e.g., the day before payroll).
• Glyman believes the competitive process will push yields higher over time, whether through new chartered institutions, neobanks, or stores of value outside the traditional banking system like stablecoins.
• He also argues that “money can think”: as dollars become more intelligent, able to determine when and under how they should be spent, more capital will be put to work rather than sitting idle. For a business making an 8% profit margin, deploying capital into the business earns far more than any overnight rate, so smarter allocation means more dollars in flight and less sitting in bank accounts.
• He agrees with the broader macro point that better information about counterparties should reduce the cost of financing across the system.