Tactics for Budgeting at Hyperscale

The CFO role has evolved faster than the tools built to support it.

Most finance teams are still running infrastructure designed for a job that no longer exists. The reporting. The reconciling. The close that bleeds into the next month. That's not finance - that's overhead with a title.

Agentic Finance shouldn't multiply your output - it should eliminate the work that was never worth doing in the first place.

That's why I run Mostly Media on Brex - an intelligent finance platform with AI-powered agents that do exactly that. Expenses handled automatically, policy enforced before the spend happens, books closed in minutes. So I can spend my time on the work that actually moves the business.

If you’re looking to fill a finance or accounting position on your team, I’d love to help.

The person you need is reading this newsletter.

Get in touch with my recruiting arm here.

The first real budget I ever built was obsolete by February. That's when Covid hit.

The second real budget I ever built was obsolete by April. That's when work-from-home jolted our sales into action.

And the third, fourth, and fifth budgets I ever built were also wrong, because we kept (for some unforeseen macro miracle) outperforming plan.

By the sixth, I had learned nothing. I threw up my hands, decided I had no idea what came next, and set the targets enormous. Those were wrong too, just in the other direction.

I was running FP&A at a company going from $10M to $150M in revenue, and from about 150 people to 1,500, in under three years. I'd close the month, update the model, walk leadership through it, and feel pretty good for about nine days. Then sales would come in 40% ahead of plan, or we'd greenlight three new offices, or a channel nobody believed in would suddenly work, and I'd version up the model for the 142nd time.

So I did what I think every FP&A person in hypergrowth secretly does (and would never put in writing). I stopped believing my own plan.

I kept building it, presenting it, defending it (and grasping for any semblance of logic to explain something that was obviously not growing logically). But in my head I was running three “types” of plans at once.

I visualized us on three sets of rails. One train track where we kept catching fire, a second where things cooled off and went back to linear, and one in the middle that I didn't actually believe but seemed not too hot, not too cold. It was the least absurd one, so that’s what I presented, but also the one I had the least confidence in.

I thought that was a ME problem… a sign I wasn't good enough at forecasting yet.

Then I sat down with Marten Abrahamsen, the CFO of Vercel, at the New York Stock Exchange.

The company’s annual recurring revenue (ARR) has skyrocketed from $100 million at the beginning of 2024, as reported by The Information, to a run rate of $340 million by the end of February 2026, according to Forbes. They’re certainly somewhere well beyond that by now.

I asked him how he possibly runs an annual operating plan when they are moving that fast. Because it’s hard enough to budget for a company growing 30% y/y. How could it be feasible to plan for something growing closer to 300%?

And he described running those same three sets of rails I’d attempted to ride, except he'd done something I didn’t do. While I only ever pictured the tracks, he’d laid them, pre-negotiated them, and built the company to jump from one plan to another based on a set of triggers.

The standard FP&A move is to run one plan and keep versioning up. In January you look fresh to death. Pressed Brooks Brothers shirt, perfectly formatted 17 tab model, with assumptions highlighted in a mustard yellow hew. By Q3 you're Mayhem from the Allstate commercials, all banged up, dragging the corpse of your financial historicals across the floor to jam them into the model one more time. You've been in a car wreck. And instead of getting in the car that still runs, you keep dragging the totaled one toward the finish line. Because it's your car. You built it.

This works fine if you're growing 20% a year. Next quarter kinda rhymes with this one, the misses are small enough to explain, and last month is a decent guess at next month. None of that is true at triple digits. Last month is a stranger; Adele would call it someone that I used to know.

So what you've built is a bad weather report. It tells you it's raining, which you knew, because your socks are wet. What it can't tell you is whether it's safe to send the team out windsurfing tomorrow. And that's what you actually need to know to make forward looking decisions.

You might convince yourself you're forecasting, but what you're really doing is providing historical commentary to something that already happened, and nudging the next few months a few points depending on how the month went (and according to some logic that will sound intelligent).

And the real problem underneath all of this is that hypergrowth doesn't move the way your model assumes. It's not a straight line up. It's not even the clean hockey stick everyone draws on the whiteboard when they talk about growing in a “non linear fashion.” Yes, growth is accelerating, but maybe not week to week, and absolutely not day to day (IYKYK).

Marten described it as being in a rowboat. You row like hell and lurch forward. Then you sit dead in the water for two weeks while signups go flat-ish and you start rewriting the hiring plan (but not telling anyone yet). But then... a massive lurch forward, for no reason you can point to.

You oscillate between these hyper-aware states of “this is falling apart” and “I can't believe we're going this fast.”

And yet, when you zoom out a year later and smooth the lines a bit, it turns out you were growing very fast the whole time, sometimes at an increasing speed, but it was never the clean line you thought it would be.

Last year Marten ran one model and kept begrudgingly updating it as they outperformed. They learned from their own success, and this year took a more dynamic approach. Now they can move from one plan to another in real time. Each plan has its own trajectory and its own resource allocations attached, and those resources unlock as the business hits certain trigger points.

What's important, and a real frame shift for me, is that he doesn't “pick” one plan to start the year. He gives everyone visibility into the paths, then waits to see which one the business is actually tracking to.

Back to his analogy…

I have three small children and I love the growth charts the pediatrician hands you. I see my son has grown two inches since his five month check up and start extrapolating him out to six-foot-five and small forward for the Boston Celtics in 18 years. (for context, I'm five-foot-eight, so this is mostly hope). The doctor can't tell you exactly how tall the kid will be. But she can show you the curve he's on and what percentile that puts him in.

So the monthly job, and the monthly conversation, changes for the FP&A team. Rather than rebuilding the forecast, you're replotting it. You figure out which rail you're riding and you go tell the rest of the business which one it is.

Oh, and you can't just keep this to your internal team. You have to be honest with the board about which track you're on and how you're moving resources as you slide between them.

Going full kimono like this sounds insane to some operators, but it's a lot easier than calling one finite number at the start of a twelve month stretch and defending it to the death. Marten puts the uncertainty on the table on purpose.

To do this right, you reframe the budget as less of a number you're trying to hit and more of a set of spending decisions you've already made, sitting there waiting for permission to fire.

Each plan has its own spending attached. The top plan funds things the base plan doesn't. And you don't fund the ambitious-plan stuff in January because you're feeling good about life. You fund it in March, when the business actually climbs onto that rail and trips the trigger.

In Marten's words:

When you hit a trigger you don't go relitigate the resource allocation or build a deck to get it approved. Finance and the rest of the org are already expecting you to spend, hire, and allocate against it. It's actually bad if you don't, because now you've got a gap in the P&L. You moved onto the rail for a higher revenue target and don’t want to underfund the demand gen that's supposed to feed it. You're starving the plan you just qualified for.

And you're almost never cleanly on one rail. That's fine, it's the normal state.

This lets a hypergrowth company move fast without drowning in its own success. Nobody's standing flat-footed waiting three weeks for a hiring plan to get unlocked. It's a continuous flow instead of a constant hurry-up-and-wait for green lights.

The example Marten gave me was regional events. Vercel historically ran a few big US conferences. This year the bet is to break that spend into smaller events around the world.

So the spend isn't a blank check. It's tied to a number they have to believe in:

And the good news is none of this requires you to guess right. Making a binary choice between this plan or that plan, knowing there's something like a 20% gap in spending between the two, is scary. This isn't one bet. It's a standing set of decisions, and you don't have to know which rail you're on in January. You just have to know what you'll do on each of them, and let the business tell you where it's going.

One caveat before you run with this, because I've watched people trip in hyper growth. Moving fast and funding against your triggers is great until it becomes a reason to skip the unglamorous work you need to put in to make sure the infrastructure can support it.

Marten got bitten by his own version of this. They were growing fast, short on people, and kept some processes manual on purpose.

Famous last words: we'll do it by hand for now, and if it gets huge, hey, that's a good problem to have.

He doesn't buy it anymore.

I learned this the dumb way. At the company going from $10M to $150M, we ran sales commissions out of an Excel workbook. Actually, not even a full workbook. It was one tab, with cascading rows into oblivion, and arithmetic logic. My man Ryan could knock out thirty people in an afternoon, no problem.

Then we hired into the plan we'd qualified for and ended up with a couple hundred people carrying a bag, half of them on custom plans somebody had promised in an offer letter, and that one tab became a live grenade. Me explaining to the CFO why a “good problem to have” was now a six-figure true-up that messed up our quarterly projections and Ryan's wife being mad he hadn't slept in a week. We went full dumpster fire mode, and pissed off the people who brought in deals.

What nobody tells you is that it gets harder the better you do. The more you hit plan, the more track you need to lay before the train reaches it.

You can't unlock a hiring plan in real time if you don't have the recruiting capacity to fill it. You can't spend into the top rail of software spend if you don't have someone in procurement to cut the deals. There are levels to the game, and you can't invite 300 more people to live in your hotel without first building the plumbing to handle them.

Apple | Spotify | YouTube

I sit down with Marten Abrahamsen, CFO of Vercel, at the NYSE to talk about running finance inside a hypergrowth AI company. We cover

Wishing you a plan you get to abandon on purpose,

CJ