You built a solid budget. You kicked off the year focused on growth and efficiency. But there’s a troublesome budget item popping up everywhere.

Not because you forgot it, but because you didn’t know how to size it… token costs.

If your team is doing real AI work, running agents, scaling content operations, personalizing at volume, you are consuming tokens. And unlike most costs in your budget, tokens don’t come with a fixed cost. They scale with usage, and usage is scaling faster than most first-year plans assumed if your team is pushing the envelope.

A few weeks ago, Nvidia CEO Jensen Huang predicted that engineers will soon negotiate “how many tokens come with my job” when they negotiate their salary. In part, that’s because Nvidia engineers are expected to use at least half of their salary in tokens! While Jensen’s lens is engineering, the underlying point applies to marketing too. Token consumption is becoming a proxy for how seriously a team is using AI, and right now, almost nobody has enough history to forecast confidently. Or how overall marketing tech budgets will be affected.

The amazing Ray Rike of Benchmarkit and I are running a Marketing Benchmark Survey to see how AI has changed marketing budgets in 2027. Have you decreased headcount in proportion to tech spend, anticipating this AI and token issue? By how much?

👉 Take the 2027 Marketing Budget Benchmark Survey — under 6 minutes, and we need 50 more respondents to hit statistical significance.

The Benchmarks That Have Held for Years

We’ve been tracking B2B marketing budget ratios for years, and they’ve been remarkably stable over time averaging out to:

• ~45% headcount

• ~45% programs

• ~10% technology

In Ray’s reports we break it down by company size, funding and growth rate since those nuances matter, but the consistency has been remarkable.

This year, all three lines are in motion as revenue / headcount is expected to grow radically, as AI replaces apps, but adds costs, and as media prices soar, AEO/GEO demands new budget, and organic traffic tanks.

What the Early Data Is Showing

The first signals from the survey, combined with my ongoing CMO conversations, are pointing in a few directions.

Technology budgets are up. Early data shows tech allocations up 2 to 3 points year over year, driven by AI tooling, data infrastructure, and orchestration. But it’s not a clean line. SaaS rationalization is happening at the same time, as AI replaces point solutions and teams consolidate their stacks. Whether your net tech budget is higher than planned depends a lot on how aggressively you’ve been cutting legacy tools alongside adding new ones.

Headcount is a tale of two companies. At traditional SaaS companies, headcount is roughly flat while expectations are rising. The same team is being asked to support more products, more segments, more pipeline, with AI utilization increasingly part of the performance conversation. But at AI-native companies, headcount is growing fast in absolute terms. The interesting thing is that the ratio is still inconsistent, because revenue is growing even faster than hiring, and organic traffic often outpaces paid at these companies. Either way you slice it, the headcount percentage of the total marketing budget is under pressure, just for very different reasons depending on the kind of company you're running.

Programs are holding, with continued efficiency pressure. CFOs want to see AI driving down cost per outcome, not just improving output quality.

The Cost Nobody Modeled Last Fall

Here’s the question I keep hearing: how do we actually budget for tokens?

There are two versions of this problem, and most teams are dealing with both at once.

The first is internal consumption. If your team is running agents, scaling content operations, or building personalization into customer journeys, you’re burning tokens. Possibly more than you anticipated.

The second problem is new AI-native software you want to buy. As you adopt AI-native software, you’re increasingly encountering platform fees plus token-based consumption charges. It sounds straightforward until you try to forecast it. Even the vendors selling you these tools often can’t tell you what average usage looks like, let alone what small, medium, and large usage looks like. And while throttling your own users or your own team is an option, it’s not a great one if your’e trying to drive AI productivity.

Why This Moment Is Worth Capturing

We’re at an interesting point: far enough into 2026 to have real signals, early enough that the data can still shape how people think about the rest of the year.

This year’s budget survey isn’t just about confirming the consistent benchmarks of the past. This is the first year that real spend on tokens, AI tooling costs, and SaaS rationalization all hit the budget simultaneously, and where headcount spend is really being offset by tech investments. How does it all roll up?

If your experience is relevant, and it IS, your data point matters.

👉 Take the survey here — under 6 minutes, and the full results go to everyone who participates.

Huge thanks to Ray Rike and the BenchmarkIt team for running this, and to Jon Miller for collaborating.

Carilu Dietrich is a former CMO, most notably the head of marketing who took Atlassian public. She currently advises CEOs and CMOs of high-growth tech companies. Carilu helps leaders operationalize the chaos of scale, see around corners, and improve marketing and company performance.