TL;DR
The 2026 marketing budget has a token-shaped hole in it. The historical B2B split of roughly 45% headcount, 45% programmes, and 10% tech is breaking under two pressures at once: internal token consumption from your own agents and content ops, and external token-priced software you want to buy. Most CMOs went into 2026 without a forecasting model for either. This article gives you one, plus the conversation to have with your CFO before the next budget cycle.
For the better part of two decades, the B2B marketing budget has had a stable shape. Headcount around 45%. Programmes around 45%. Tech around 10%. Carilu Dietrich has written about this benchmark in her newsletter for years. It has held through SaaS booms, recessions, the move from outbound to inbound, and the rise of marketing operations. It is not holding now.
Tokens are the reason. And the 2026 marketing budget has a token-shaped hole in it because almost nobody planned for them.
What ‘Tokens in the Budget’ Actually Means
It means two distinct things that look like one line item but behave very differently.
1. Internal token consumption (your own agents and content ops)
You run agents internally. Claude Cowork for the marketing team. ChatGPT Enterprise for the wider company. An n8n workflow that uses GPT-4o for lead scoring. A custom agent that drafts campaign briefs. Each of those consumes tokens, and the bill arrives monthly. For a marketing team of 8 with active agentic tooling, the internal token bill is now plausibly $3-8k a month and rising.
The Nvidia line that should make every CFO pay attention: Jensen Huang has predicted that engineers will soon negotiate ‘how many tokens come with my job’ as part of their salary package. Nvidia engineers are already expected to consume token spend roughly equal to half their salary. Marketing is two years behind engineering on this curve, not five.
2. External token-priced software (AI-native vendors you want to buy)
The new AI-native marketing tools price differently. There is a platform fee, and then a usage charge denominated in tokens, calls, or actions. The vendor sets a ‘starter’ bucket and then bills consumption above it. The trick is that the vendor often cannot tell you what average usage looks like, let alone what small, medium, and large usage looks like. Carilu nailed this in her piece: ‘even the vendors selling you these tools often cannot tell you what average usage looks like.’
Result: you sign a contract with a platform fee that looks reasonable and a usage curve that goes vertical in month four. The CFO is rightly irritated.
Why the Old 45/45/10 Split Is Breaking
Three pressures are moving simultaneously.
Tech is rising. Early 2026 data shows tech allocations up two to three percentage points year on year, driven by AI tooling, data infrastructure, and orchestration. The 10% historical share is now 12-14% in AI-forward teams.
Headcount is splitting in two. Traditional SaaS teams are running flat headcount with rising expectations (AI is the productivity multiplier). AI-native teams are growing headcount in absolute terms, but the headcount-to-revenue ratio is dropping because revenue is outpacing hiring.
SaaS rationalisation is happening at the same time. Teams are cutting redundant point tools as AI-native platforms absorb them. So the tech line is going up, but inside it the mix is shifting from legacy SaaS (down) to AI-native usage-priced platforms (up sharply).
The CFO sees three lines moving at once and reasonably asks: what is the new normal? Most CMOs cannot answer.
A Token Forecasting Model You Can Use Tomorrow
Here is a model simple enough to defend to a CFO and rigorous enough to be roughly right.
Step 1: Inventory who uses what
List every AI tool your marketing team touches. Per tool, list which roles use it, how often, and on what tasks. This is the boring 30 minutes that most teams skip. Without it, every other step is guessing.
Step 2: Set a per-role token allowance
For each role, agree a monthly token budget. A content marketer running Claude or ChatGPT for drafts: $120-250/month. A demand gen lead running agents for lead scoring and outreach: $300-600/month. A marketing ops engineer running orchestration workflows: $500-1,200/month. These ranges shift by tool, but the discipline of setting an allowance is what matters.
Step 3: Forecast usage-priced platforms with three scenarios
For every AI-native vendor on your stack, model usage at low, expected, and high. The vendor will not give you the curve, so you have to build it. Take the documented unit cost, multiply by your forecast volume, and add 30% as the ‘we underestimated’ buffer.
Step 4: Surface the line to the CFO before it surprises them
Add a ‘Tokens and AI usage’ line item to your monthly forecast. Track it separately from SaaS subscriptions. The point is not perfection; the point is visibility. The CFO can absorb a new line that grows over the year. The CFO cannot absorb a surprise bill in month seven.
Step 5: Tie the spend to outcomes
This is the line that wins the conversation. Per dollar of token spend, what outcome do you expect? Faster brief-to-launch? Lower cost per qualified lead? Fewer agency hours? Translate token cost into the cost-per-outcome the CFO already grades you on. If you cannot, the budget gets cut.
What This Means for the 45/45/10 Split
The honest answer is nobody knows the new ratio yet. My working assumption, based on the engagements I am running in 2026: tech moves from 10% to 15-18%, headcount drops from 45% to 38-42%, programmes hold roughly steady at 42-45%. Inside tech, AI-native consumption rises faster than legacy SaaS falls, so the absolute tech spend is up.
If your 2026 budget still has 10% tech, you have probably not modelled tokens. That is the hole.
Frequently Asked Questions
How fast is the token line growing for a typical marketing team?
For teams that have adopted agentic tooling in 2025-2026, the line item has roughly doubled year on year. Most of the growth is internal consumption (agents, content ops, lifecycle automation) rather than external platform usage, though the latter is the faster-growing piece.
Should tokens sit under marketing budget or IT budget?
Marketing budget if the consumption is marketing-driven (your agents, your content ops, your platforms). IT budget if it is part of a company-wide AI platform contract. In practice, the line gets split, and the negotiation with the CFO determines who carries it. Better to claim it openly and instrument it well.
What is a sensible cap to put on internal token consumption?
Start with 1-2% of marketing budget for token consumption alone in 2026. Expect that to rise to 3-5% over the next two years as agentic workflows mature. Caps are useful for budgeting; they should not be useful for capping value creation. If the team is hitting the cap and producing outcomes, raise the cap.
How do I explain token spend to a CFO who has never seen this line item before?
Three sentences. ‘Token spend is what we pay for AI usage, internally and externally. It is replacing some agency hours and freelancer spend, and it is funding capabilities we did not have before. Here is the forecast, here is the cost-per-outcome it produces, here is the contingency we built in.’ Then walk them through your model. Most CFOs will be more reassured by the model than by a low number.