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.

👋 Hi, it's CJ Gustafson and welcome to Mostly Metrics - my newsletter for CFOs about the metrics and financial strategies behind the world's best businesses.

As a reminder, I wrote you a free book on annual planning. It’s pretty much my Magnum Opus or whatever they call it.

Ok, grab your zyn and let’s begin.

Before you model a buck, you gotta pull five things:

Everything downstream is built on these, so it's worth getting them rock solid before you start.

Ideally you begin this season working off the latest and greatest forecast you're currently using to make decisions, and this exercise becomes a more granular reforecast on the next 15 months rather than a blank canvas (you'll see why I say 15 and not 12 in a second).

The September P&L is the final month I've traditionally included in my "locked" baseline. This means I'm really turning the screws on months 10, 11, and 12 of the current year and 1 through 12 of the next (or 15 total months for those keeping score at home). You have to draw a line in the sand to start. And if your board is functioning right, you need to get the plan approved around Thanksgiving during the Q3 board meeting.

Now, the calendar I'm describing was built for the way planning has always worked. Up until our AI overlords took over.

But nothing about September is holy now that we can reswizzle numbers to our heart’s delight. The cutoff was always a concession to how long it took to gather five messy inputs and scrub them to be trustworthy, with a backstop of year end for when you need to go pencils down (otherwise these things drag on forever). Timeboxing annual planning is a necessary evil, due to the sheer amount of corralling it took (both of people and data).

That constraint is loosening. If your headcount, cost, and attainment data is clean and piped in, and you have AI doing the assembly instead of an analyst named Chester doing it by hand, these five inputs turn from a once-a-year foraging in the woods exercise into a standing feed you can refresh every month. While you still anchor the annual plan on September, you have the ability to rev the engine across a broader swath of materials (even when they don't agree).

Here are the five data pillars.

This should be an export from your HRIS system (Workday, Hibob, Bamboo HR, etc.). It should include every full-time employee and contractor, with their department, base salary, and any bonus they're eligible for. This is your "butts in seats baseline."

Note: if you're doing your planning in an FP&A tool like an Abacum, Anaplan, or Aleph (that's a lot of A's), you'll be able to pipe this in via API.

This list should be segmented into two buckets: positions actively being recruited for, and positions approved in the current year budget but not being actively worked on yet.

For all you FP&A folks, this is where you get to do some gambling. You'll need to make a call as to the probability of each position being filled before year end. And if it doesn't, you decide if it deserves an initial placeholder in the baseline for next year (e.g., it will still get filled at some point). The first bucket will have a higher close percentage applied than the second, as the time to open and fill a role likely pulls you into the next year.

This piece of advice is critical: where I've seen budgets fall off the rails is when open headcount gets "trapped" in no man's land. They either get excluded from the baseline but get filled before year end (so there's a real-life person working at the company you haven't contemplated for), or they don't get filled by year end but the department leader assumes you still had that role rolling over into the next year's budget envelope (so they're one short when they receive it).

The net effect of both scenarios is you eventually create an additional headcount out of thin air that wasn't contemplated. That's why budget envelopes for the following year need to be presented as a "theoretical max to not exceed" instead of a "number of additional heads," to capture anyone who gets caught in the middle. You always get screwed when working off "additional" or "incremental."

You don't have conclusive numbers on the year yet. That's right, we're working on next year's budget before this year is solidified. And it could change if your year end is heavily backend loaded (I've worked at places where 35% of our new biz was in Q4, of which 20% was in December, and of which 10% was in the last week of the year).

Many companies circumvent this problem by just waiting to formally approve their budgets in late February or early March during their Q4 board meetings. I think that's wrong. The worst thing you can do to any company in hypergrowth mode is delay decision making. Whatever you gain in accuracy by waiting to get your board plan approved once the year is over, you lose multiples over in momentum.

The continuous approach softens the objection that used to make people wait. The reason you held off was those last three months you couldn't see, so a fall plan felt like it was built on shaky ground. When your actuals feed in monthly and the model rebuilds itself as they land, that blind spot shrinks every few weeks instead of sitting frozen from Thanksgiving to January.

The plan still gets approved on September numbers. But now you close the gap on those last three months on a weekly basis, rather than defending a stale forecast that ripples through next year's projections.

To establish a proper baseline, revisit what you thought the targets for account executives were last year, see how close individual reps came as well as the median team members. Then have a discussion internally if you think they'll be going up or down.

Some hypothetical figures:

If you work at an AI company, multiply by 10. Just kidding, but not really.

Things that will impact your decision making include territory splits and new product launches. As you cut down on a rep's territory, you need to give them more to sell to a smaller subset of accounts. Otherwise they're getting the short end of the stick.

You want a full trailing 12 months to get the full picture, not just the nine year-to-date months. That means LTM from September.

I like to look at costs by category at the following levels:

One preview before we start building. You'll notice attrition isn't in these five inputs, and that's on purpose. I don't budget for it. Any gap between people coming and going is a tailwind to the plan, and it's a gift you want to find later in the year, not a given you bake in up front. There's a right way and a wrong way to handle attrition, and it's worth its own section, so I'll come back to it in full when we get to headcount… which you can also read now you eager beaver by download the freaking book below.

On this episode of Run the Numbers, I sit down with Matt Hedberg, a sell-side equity analyst at RBC who has covered software for more than 20 years. My friend Matt is wicked smahhht.

We break down:

Wishing you competitive moats that don’t disintegrate over night,

CJ