With flat budgets, AI investment is funded by moving money, not by new money. The expert question isn't "how much AI?" but "what mix gets us ready, and what do we stop?"
The benchmarks, and their limits
Gartner's 2026 CMO Spend Survey (401 marketing leaders, mostly at $1B+ companies): AI averages 15.3% of marketing budgets, 21.3% in organizations with mature AI readiness; marketing budgets average 7.8% of revenue, 8.9% in AI-ready organizations. Gartner's release does not say how the rest of the budget splits across martech, media and labour, so treat any channel mix quoted in the press as unconfirmed.
Benchmarks are descriptions, not targets. Your right AI share depends on your readiness, your category and how AI is counted (does it include AI specialists' salaries? AI features in platforms?).
A readiness-first allocation logic
- Measure your true AI share, including AI inside people, agency and martech lines.
- Score readiness on data, governance, process, skills, measurement and stack control.
- Create a reallocation pool from lower-value spend: unused licences, duplicative agency work, media that fails incrementality tests.
- Split the pool by readiness: if readiness is low, most goes to foundations (data, guardrails, training, measurement); as readiness rises, more goes to AI tools and scaled use cases.
- Tie every AI line to a metric and a review date.
Where savings come from
AI-driven production savings (Klarna reported about $10 million a year in 2024) and incrementality tests that expose ineffective spend are the two most common sources. Martech utilisation is a third: Gartner found in 2023 that marketers used only 33% of their martech stack's capability.
Avoid
- Matching a benchmark percentage without readiness.
- Funding tools from the foundations budget.
- Counting "time saved" as budget saved.