Advertising
The Harvard Crimson is the nation’s oldest continuously published daily college newspaper and the only daily publication of Cambridge, Massachusetts. We work with advertisers ranging from national companies to local businesses, from student recruiters to advertising agencies.
The media and advertising industry is a constantly changing landscape, and our business team is focused on innovative advertising solutions to stay ahead of the curve and deliver results to our clients.
Sponsored Content
The lab that's rewriting the rules of finance
Inside Ramp Labs, a group focused on AI experiments within Ramp’s Applied AI team:
I refreshed the analytics dashboard for the hundredth time.
It was the week before Thanksgiving, and we had just shipped Ramp Sheets, an AI-powered spreadsheet editor that my team had been grinding on for two months. Two months of late nights, of debugging formula engines, of demoing to skeptical colleagues who smiled politely and moved on. For most of that time, we were the only people at Ramp who believed this thing could work.
Now the numbers were moving. Fast.
The first thousand users signed up within hours of launch. Then the tweets started. Finance Twitter picked it up. VCs were sharing it. By the end of the week, we had crossed 2 million impressions. My phone wouldn't stop buzzing — messages from friends in investment banking, founders I hadn't talked to in years, people I'd never met asking how we built it.
I remember sitting in my apartment that night, watching the user count climb past 5,000, then 7,000, then 10,000. No paid marketing. No PR agency. Just a product that people actually wanted.
That was two months ago. Today, Ramp Sheets has become one of the most-used AI spreadsheet tools in finance. Teams at KKR, Thrive Capital, General Catalyst, Wharton, Stanford, and MIT use it daily.
This is the story of how we built it, and why it matters for everyone entering finance and tech today.
Something strange is happening in finance...
Told by Alex Stauffer, Lead of Ramp Labs
The most powerful professionals in the world, those who move billions of dollars, structure the deals behind every major acquisition, and forecast the financial futures of entire industries, still spend their days hunched over spreadsheets. They manually input data. They hunt for formula errors that could cost their firms millions. They copy and paste between endless versions of files named Q3_Model_v2_FINAL_revised_ACTUALLY_FINAL.xlsx.
There are 750 million monthly active Excel users globally. Eighty-nine percent of companies use spreadsheets for finance. The average finance professional spends 2.5 hours every day working in them. And yet, despite decades of software innovation, the fundamental experience has barely changed since Lotus 1-2-3.
That is, until now.
Unlike most teams at fast-growing startups, Ramp Labs wasn't given a roadmap. There were no KPIs to hit, no quarterly targets.
Instead, the team was set loose to explore the frontier of applied AI, to build things that seem impossible, break them, learn from the wreckage, and build again.
"Oftentimes the really interesting use cases only come up when you're doing pure exploration in the tech itself," Alexander Shevchenko, our engineering lead, recently explained at an event hosted by 8VC. "There's no direct metric we optimize for. We can just build something incredible."
The results have been wild. We put Claude Code inside RollerCoaster Tycoon, letting an AI agent manage park operations, hire mechanics, and adjust prices in real time. The resulting video went viral. People started asking: what the hell is going on at Ramp Labs?
But the RollerCoaster Tycoon stunt wasn't just for laughs. It was a proof of concept that demonstrated how AI agents can navigate complex interfaces, make financial decisions, and execute on goals in real time. The same technology that manages a virtual amusement park can, with the right infrastructure, transform how real businesses operate.
The bigger project was already underway. Our team had been embedded with Ramp's internal finance team, studying how they work. The goal was to automate their most tedious processes. What we discovered was both obvious and profound: finance professionals live in spreadsheets.
When Shevchenko reviewed Loom recordings of finance workflows, he found that at any random moment, there was a 95% chance the user was in Excel. Every process, forecasting, budgeting, modeling, and analysis, flowed through spreadsheets.
Our first attempt at automation generated Python scripts. The finance team's response was polite but firm: they couldn't verify code they didn't understand. In finance, you can't trust black boxes. A single error can cascade into millions of dollars in losses.
So we pivoted. Instead of replacing spreadsheets, we asked: what if AI could work inside spreadsheets, speaking the language finance professionals already know?
What followed was the hardest stretch of my time at Ramp.
Building an AI that can reliably manipulate spreadsheets sounds straightforward until you actually try it. Excel is deceptively complex: there are quadrillions of possible configurations, self-referential formula chains, cached values versus calculated values. Getting an AI to navigate all that without producing a wall of #VALUE! and #REF! errors is a nightmare.
For two months, we lived in that nightmare. We'd ship a version, test it against real finance workflows, watch it break in spectacular ways, fix it, and repeat the process. There were weeks where I genuinely wondered if we were wasting our time.
The hardest part wasn't the technical challenge, it was the loneliness. Most people at Ramp were focused on core product work. We were off in a corner building something that didn't have an obvious business case yet. When we showed early demos to colleagues, they were polite. Interested, even. But you could tell they weren't convinced that this was going to be a big deal.
I kept telling myself: if we get this right, it changes everything. Finance professionals spend their entire careers in spreadsheets. If we can make that experience 10x better, the market is enormous.
So we kept building.
The concept we landed on is elegantly simple: a spreadsheet interface you already know, but with an AI agent that can build models, clean data, write formulas, search the web for market research, and format presentations, all from natural language commands.
Ask it to "build a 13-week cash flow forecast for a 50-person SaaS company with $8M ARR," and it delivers a board-ready model in minutes. Ask it to "benchmark our engineering compensation against San Francisco market rates," and it searches the web, pulls current salary data, and builds a comparison table, complete with percentile rankings and gap analysis.
The key innovation is transparency. Unlike AI tools that spit out final answers, Ramp Sheets shows its work. You can see every formula it writes, every cell it modifies, every assumption it makes. Finance professionals can audit the output the same way they'd audit any spreadsheet, because it is a spreadsheet.
The use cases that have emerged have surprised even us. Investment banking analysts are using Ramp Sheets to build leveraged buyout models in a fraction of the time – tasks that used to take hours now take twenty minutes. Startup founders are using it to value their own companies before fundraising. FP&A teams are automating their monthly close processes.
One founder told us they modeled out three different hiring scenarios—"What if I hire three more people to the sales team this quarter? How does that affect my burn?"—in the time it used to take just to open their existing model.
It's become a quiet staple among people who take finance seriously.
Here's the uncomfortable truth: the skills that got previous generations ahead are being automated. The analyst who was valued for their ability to build a model from scratch in Excel now competes with AI that can do it faster. The associate who pulled all-nighters formatting pitch books now watches AI do it in seconds.
But here's the opportunity: the professionals who learn to work with AI, those who can prompt effectively, verify outputs critically, and apply judgment to AI-generated work, will be the most valuable people in any organization.
This isn't a future prediction. It's happening now. At Ramp, everyone, engineers, designers, finance professionals, marketers, uses AI to accelerate their work. Claude Code, Cursor, and our internal tools have 10x'd our shipping velocity.
Ramp Sheets is just the beginning. We're shipping new features constantly: templates for common use cases, saved prompts, cloud-based auto-saving. The product is being integrated deeply into Ramp's core platform, so finance teams can go from raw transaction data to polished analysis without leaving the ecosystem.
And Ramp Labs will keep experimenting. We're exploring generative user interfaces that adapt dynamically to each user. We're pushing the boundaries of video-based process mining — imagine uploading a Loom recording of your workflow and getting an automated process map in return. We're applying reinforcement learning to make our agents faster and cheaper.
The mission is ambitious: make finance work effortless. Get rid of the tedium that consumes knowledge workers' days. Let humans focus on judgment, strategy, and creativity — the things AI can't do.
Get involved today.
Also Check Out Ramp's Velocity Blog
Velocity is read by thousands of finance leaders and operators. Read a real blog post below!
Money talks. Now it thinks.
In the first year of business, every dollar is a decision.
A $500 software subscription is a discussion. A $5,000 research tool is a debate. Everyone knows who approved what — and whether it was worth it.
Ten years later, the same subscription has empty seats renewing on autopilot, and $5,000 of research slips away as miscellaneous. There’s also a $1.2 million SAP maintenance contract that nobody dares question because “it’s always been there.”
You used to run the business. Now the business runs you. And if you do question an expense, you get a spreadsheet, not an answer.
Every founder is adamant this will not happen to them — but it inevitably does. You start building a product. You end up building a bureaucracy.
Take the average publicly listed SaaS company. Big teams, lots of bloat. They’re growing at 16% per year. What about just the top 10? The best performers. They’re growing at 30% YoY. This is all public data.
So when I tell you Ramp's underlying profitability is growing 153% year over year, that sounds absurd. It’s 10x faster each year than the median publicly traded SaaS company. Our revenue was $500 million 12 months ago, over $1 billion today.
A whole new class of companies have come along. Heavyweights that move like lightweights.
Getting big no longer means getting slow. Let me explain.
The Age of “Thinking Money” (2025 –)
For millennia, money talked — but it didn’t think.
Then AI happened. Suddenly, money was no longer just a number in an Excel sheet. For the first time it understood context, could reason, and act.
Imagine a dollar wants to leave your company. Before “thinking money” it could simply walk out. No memory of what it funded, or knowledge if it was spent wisely.
Now, that same dollar has intelligence. So before it leaves it checks for fraud and if you have permission to spend it. It has memory. So as it moves it leaves a complete audit trail and updates budgets. And it can reason. So once it’s spent it tells you if it was well used, underused, or wasted.
That’s one dollar. The average publicly listed SaaS spends over five hundred million of them each year. What if each one a) was only spent if it should be, b) audited itself instantly, and c) flowed only to the highest-impact projects?
You had a bureaucracy. Now, you have a business again.
If money thinks, what does finance do?
Rather than have me explain, let me show you.
In October, Ramp’s AI made 26,146,619 decisions across more than $10 billion in spend. This is “thinking money” with intelligence, memory, and reason. Here are a few of those decisions:
-
Our policy agent prevented 511,157 out-of-policy transactions, saving $290,981,801.
-
Our treasury agent moved $5.5 million from idle cash to 4% investments.
-
Our fraud agent blocked a $49,000 AI-generated fake invoice.
Our travel agent saved $113.34 for Zain on his trip to New York.
What do all these decisions have in common? They’re objective! Put every accountant in America in a room, give them context and 10 minutes, and they’d all agree. But in practice, thousands of small would-be improvements slip through every day.
Now that all of this is automated, your finance team is free to make a smaller number of higher-leverage decisions. They’re strategists — not clerks. Not catching policy violations, they’re designing policy that enforces itself. Not coding transactions, they’re working out how best to allocate a $500 million budget.
Big decisions. Not the small stuff.
The return of economic productivity
Forty years ago, Robert Solow said “You can see the computer age everywhere but in the productivity statistics.”
And for decades he was right. From 1947 to 1973, U.S. productivity grew 2.8% a year. In the last three decades, just 1.4%. But that was the age of “money talks.”
Now it’s the age of “thinking money.” When thinking money automates the small stuff, companies run better. The median Ramp customer saves 5% while growing revenue 12% year over year. They’re closing books in days instead of weeks, running leaner teams, and compounding efficiency gains quarter after quarter.
The promise of the computer age — widely available, cheap intelligence — is only now coming to life, and helping everyday companies get more from their time and money. Finally, we’re proving Solow wrong.
A whole new growth curve
I’d started by telling you "there's a whole new growth curve.” Well, I’d like to end by explaining it.
All the way back in 1967, there was a computer scientist named Melvin Conway. He was frustrated by how slowly software projects were moving at IBM, so he started researching the problem.
Two months later, he published the paper that introduced the now-infamous Conway’s law.
“Your product mirrors the system of the organization that built it.”
In plain terms: the reason IBM’s software was slow and complicated was because IBM was slow and complicated.
So, what if your staff no longer had to book travel, email receipts, update budgets, or chase approvals? What if there weren't three layers of management between spend and strategy?
No, you won’t magically start growing at 100% YoY. But you're building the kind of organization that can.
That’s the second growth curve. That’s “thinking money.”
The Crimson's news and opinion teams—including writers, editors, photographers, and designers—were not involved in the production of this article.