josh whiteAgent Spend Is a P&L Line Now: OPEX or Variable Marketing?
On October 5 I asked the CFOs on X a question I keep coming back to. Do we need to start splitting up usage invoices for agents between OPEX and variable marketing? If I have an agent dedicated to creative strategy for paid media, that's a variable cost in my book.
I still think that's right, and I think most small brands are going to get this wrong for a while because the bill shows up as one software line and gets dumped into G&A with everything else.
why the line matters
At a $1-10M CPG brand, the number I watch is contribution margin. That's revenue minus COGS and the variable costs that move with each order, like shipping, fulfillment and payment fees. Variable marketing comes off right after that, and OPEX sits below both. OPEX is the stuff that doesn't really change whether you ship 5,000 orders this month or 8,000.
If a cost that really moves with your acquisition spend gets parked in OPEX, two things happen. Your nCAC (new-customer marketing spend divided by new customers) looks lower than it is, and so does payback, which is nCAC divided by monthly contribution margin per customer. Then you end up scaling spend into a payback that isn't real.
Quick illustrative example. Say you spend $40k a month on new customers and get 800 of them, so nCAC is $50. Add $2,000 a month of agent usage for creative strategy and ad reporting and nCAC is really $52.50. At $10 a month of contribution margin per customer, payback goes from 5 months to about 5.3. That's not a crisis, but agent usage tends to grow as you lean on it more, and it's a lot easier to set up the split now than to unwind a year of mixed-up books later.
the software companies are having the same argument
AI companies are fighting about this on a much bigger scale, because for them inference is the product. Hyphenate's piece on cost of revenue lays out a default where production inference serving paying customers is cost of revenue, and inference for internal tools is operating expense, classified by the function that uses it. Their test is where the inference runs and whom it serves, regardless of what the workload is called.
We're not selling inference, so on a GAAP P&L most of our agent spend lands in operating expense, marketing included. That "by the function that uses it" idea maps really cleanly onto a management P&L though.
how I'd split it
Variable marketing (counts in nCAC). Any agent dedicated to acquisition goes here. That means creative strategy for paid media and ad reporting, and I'd put creator seeding in there too since it's an acquisition program. I'd treat a dedicated acquisition agent the same way I'd treat an agency retainer and include it in new-customer marketing spend, even if part of the bill is a flat monthly plan.
Variable fulfillment or CX (in contribution margin per order). A customer service agent that's priced per conversation or per ticket moves with order volume, so it belongs up there with your other per-order costs.
OPEX. The bookkeeping agent, inventory planning, the founder's chief of staff agent, general seats that everybody shares. These are real costs but they don't scale with a new customer, so they stay below contribution margin.
If you're not sure where an agent goes, ask two questions. Is it dedicated to acquisition, or would the cost go up if you doubled ad spend next month? If either answer's yes, I'd put it in variable marketing.
why I'd set it up now
Pricing on this stuff is going to move. After spending a few grand on Grok Bot usage I said I believe the $120/mo teams plan is subsidizing usage by around 90% to encourage adoption, and the future price of usage was the first thing on my list of risks for anyone building on agents. That's my read and not something the platforms have told me. If prices do go up, you want to know which part of the P&L takes the hit, and right now most brands can't tell.
The practical part is pretty simple. Give each function its own workspace, account or API key so the invoice splits itself, which is basically what Hyphenate tells AI companies to do (split spend at the account or tag level instead of estimating after the fact). Then add a couple of sub-accounts to your chart of accounts, something like "AI agents, acquisition" under variable marketing and "AI agents, G&A" under OPEX, and allocate monthly.
This is management reporting. How it's classified for your tax return is a separate conversation with your CPA, and I'd have that conversation before you change anything in the books.
If you're building out the agents themselves, I wrote up how I'd structure them in The Agent Org Chart for a $1-10M Brand.
If you want help getting your chart of accounts set up for agent spend, or getting the agents themselves set up, book a call at joshwhite.co/book-a-call and we'll sort out where your spend should actually sit.
Example numbers in this article are illustrative, not data from any company.
the growth engine
Want your nCAC and payback live every week?
On the Operator tier I build the Growth Engine, a live model that syncs your marketing performance daily and ties it to your P&L and cash forecast. The payback number you plan with ends up being the one in the books.
Not ready for a call? Grab the nCAC & Payback Calculator, it's free.