Agency AI Bills Grow Complex as Token Costs Become Unpredictable

The evolving cost structure of agency AI

Marketing agencies are confronting a new financial reality as artificial intelligence expenses grow more complex and harder to predict. Beyond simple software subscriptions, agency leaders now manage token-based pricing, enterprise licensing agreements, and the question of whether clients should share the burden.

According to agency executives interviewed by Digiday, a typical AI bill now includes three components: subscriptions to tools like Claude, ChatGPT and Google's AI offerings; tokens consumed during usage of those platforms; and employee costs for staff operating the tools. Research from Ramp indicates approximately 31 percent of companies now spend over $10,000 monthly on AI, suggesting the technology has become a formal budget line for many organizations.

The agencies Digiday spoke with declined to share specific spending figures, but all confirmed AI costs are rising as the technology moves from experimentation to operational deployment.

How three agencies structure AI spending

Markacy, with a team of over 25 staff, primarily pays for enterprise Claude accounts, with Claude Team plans ranging from $20 to $100 per seat depending on tier. The agency treats these investments as variable costs rather than a dedicated budget line, according to co-founder and co-CEO Tucker Matheson. ChatGPT subscriptions are added for employees who request them.

Crispin, owned by Stagwell, takes a different approach with approximately 100 enterprise AI tools spanning media planning, social listening, and influencer discovery. Chief transformation officer Freddy Dabaghi said most deals are structured as flat, fixed fees with annual renewals to avoid unexpected monthly fluctuations.

Dept deploys Claude, ChatGPT and Google tools to accommodate varying client preferences, with those three platforms generating the largest token costs, according to Jonathan Whiteside, global executive vice president of technology. AI spending has not yet reached double-digit percentage of the company's budget, with human capital remaining the dominant expense.

Why it matters

The shift to metered, consumption-based AI pricing creates financial uncertainty that conflicts with traditional agency business models built on predictable costs and fixed-fee client contracts. As Forrester research shows 74 percent of agencies now use generative AI to summarize documents and 70 percent apply it to research tasks, the operational integration of AI is driving costs upward while making them harder to forecast. This tension is forcing agencies to develop new governance structures and explore billing models that pass token costs through to clients—potentially reshaping how agency services are priced.

New oversight and billing models emerge

Agencies are implementing controls to manage the unpredictability. Dept established an internal monthly AI Tool Committee that reviews tools to approve, test, discontinue or certify for company-wide use. The agency is also building a system to help staff select the most appropriate and cost-effective AI model for specific tasks or clients.

"We do have more checks and balances with the AI tooling, purely because—number one, it's changing so fast. And secondly, the cost has been unpredictable in the past year," Whiteside said.

Crispin leverages holdco-level enterprise agreements to set clear cost thresholds and is auditing its tech stack to eliminate unnecessary subscriptions.

The question of who pays is also evolving. At Goldman Sachs' Communacopia and Technology Conference, Omnicom CFO Phil Angelastro said the holding company is building a business model to manage, measure and get reimbursed for token costs, moving toward outcome-driven revenue models.

One agency executive, speaking anonymously, said their firm is testing a model that bills AI token costs to clients as materials, separate from human labor. Another experiment combines human work costs with embedded token expenses in client billing.

Despite the growing complexity, human salaries remain the largest agency expense. "I still need people who can get on a phone with a CMO and be impressive," Matheson said.

As agencies begin 2027 budget planning, technology spending is expected to continue its decade-long upward trajectory, with AI adding a new layer of financial complexity to manage.

These details were first reported by Digiday.