Inside the complex anatomy of agency AI bills

For the most part, agencies are carving out a piece of their existing tech budgets to fuel AI spend rather than creating a new line item on the P&L. According to research from Ramp, about 31% of companies spend more than $10,000 per month on AI, indicating AI is a formal budget line. All three of the agency execs Digiday spoke with declined to outline specific AI spend figures. 

AI billing line items 

At Markacy, the primary AI costs come from paying for enterprise Claude accounts for a team of 25-plus staff, Tucker Matheson, co-founder and co-CEO of Markacy, told Digiday. For teams of two to 150, Claude Team plans can range from $20 for a standard seat to $100 for a premium seat. For enterprise accounts, seat price starts at $20 per seat and usage cost scales depending on model and tasks, per Claude’s pricing. The agency additionally pays for ChatGPT for certain employees who request it, per Matheson. Rather than a dedicated AI tool budget, investments are treated as variable costs, he added. 

Crispin has a different approach. The Stagwell-owned agency uses around 100 enterprise AI tools across media planning, social listening, influencer discovery and more, according to Freddy Dabaghi, chief transformation officer at Crispin. Most of the agency’s AI deals are structured as flat, fixed fees with annual renewals to curb unexpected monthly costs, he added.

Then there’s Dept. The agency’s AI toolkit spans across Claude, ChatGPT and Google to account for varying client preferences, Jonathan Whiteside, global evp of technology at Dept, told Digiday. Those three are where the agency sees the biggest token costs, Whiteside added. 

AI costs have not yet reached a double-digit percentage of the company’s budget—at least that’s the case for Dept. Per Whiteside and other execs, human capital remains the dominant agency expense.

That means staff salaries remain a major line item on the agency P&L, overshadowing the costs of agencies’ growing AI toolbelts, per the three execs. 

“I still need people who can get on a phone with a CMO and be impressive,” said Matheson. 

‘Checks and balances’

AI costs are climbing. Agencies have thrown their weight behind AI tools to expedite work, streamline businesses and woo clients. 

Forrester recently published research pointing out 74% of agencies use generative AI to summarize documents and communications and 70% apply AI to research and competitive intelligence. Meaning, AI has gone from experimentation to operationalization, driving up costs, per the execs. With more hands on keyboards, there can be less predictability around those costs. 

“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,” said Whiteside. 

Dept’s checks and balances comes via an internal monthly AI Tool Committee that reviews AI tools to approve, test, drop or certify company-wide usage. The agency is also working to monitor costs via a system that helps staff decide which AI models are best for a specific task or client. 

Whiteside isn’t alone in his thinking. Tokens are hard to predict, he said, “because we just don’t know how many tokens an individual might consume on a monthly basis.” More companies and agencies are starting to scrutinize token costs and the ROI of AI.

Crispin takes an annualized approach to contracts. Because of enterprise agreements at the holdco level, Crispin has been able to set clear AI cost thresholds, Dabaghi said. 

“There is some usage based with the Google agreement, but it’s not completely Free Willy, where we were going to get completely screwed over,” he added. 

Crispin is also revising its tech stack—reviewing platform usage logs and tech vendors to cancel unnecessary subscriptions and maintain the tools the agency actually needs. 

The new agency business model 

Agencies aren’t shouldering the AI costs alone, actively looking to revamp the agency business model—that much was made clear at the Goldman Sachs’ Communacopia and Technology Conference last week. 

Omnicom CFO Phil Angelastro said the holdco is in the process of building out a business model to manage token costs, measure them and get reimbursed for the additional costs of work. 

“Ultimately, we’re going to move towards a more outcome-driven model from a revenue perspective, and we think that’ll be a long-term positive in the end,” Angelastro said at the conference. 

At least one agency exec, who spoke on the condition of anonymity, said their agency is experimenting with a model in which AI token costs are billed to clients as materials, separate from pre-set human labor costs. The agency is also experimenting with a model that combines the human cost of work and embedded token cost into client billing. 

In the near term, agencies have started looking at 2027 budget planning and how AI will impact it. 

“My gut is our tech budget will increase every year, as it has been the last probably decade,” said Dabaghi.